diff --git a/.copyright.hook b/.copyright.hook index de97ce90ac4a80d8d0532ea8898385b2e5781ce0..09afff2072df3384a429d01d06188218ae6e85d1 100644 --- a/.copyright.hook +++ b/.copyright.hook @@ -9,7 +9,7 @@ import subprocess import platform COPYRIGHT = ''' - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved. Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. @@ -49,12 +49,17 @@ def generate_copyright(template, lang='C'): LANG_COMMENT_MARK = "//" lines = template.split(NEW_LINE_MARK) - ans = LANG_COMMENT_MARK + COPYRIGHT_HEADER + NEW_LINE_MARK + BLANK = " " + ans = LANG_COMMENT_MARK + BLANK + COPYRIGHT_HEADER + NEW_LINE_MARK for lino, line in enumerate(lines): if lino == 0 or lino == 1 or lino == len(lines) - 1: continue - ans += LANG_COMMENT_MARK + line + NEW_LINE_MARK + if len(line) == 0: + BLANK = "" + else: + BLANK = " " + ans += LANG_COMMENT_MARK + BLANK + line + NEW_LINE_MARK - return ans + return ans + "\n" def lang_type(filename): @@ -62,6 +67,8 @@ def lang_type(filename): return "Python" elif filename.endswith(".h"): return "C" + elif filename.endswith(".c"): + return "C" elif filename.endswith(".hpp"): return "C" elif filename.endswith(".cc"): @@ -77,10 +84,13 @@ def lang_type(filename): elif filename.endswith(".proto"): return "C" else: - print("Unsupported filetype") + print("Unsupported filetype %s", filename) exit(0) +PYTHON_ENCODE = re.compile("^[ \t\v]*#.*?coding[:=][ \t]*([-_.a-zA-Z0-9]+)") + + def main(argv=None): parser = argparse.ArgumentParser( description='Checker for copyright declaration.') @@ -89,9 +99,14 @@ def main(argv=None): retv = 0 for filename in args.filenames: - first_line = io.open(filename).readline() - if "COPYRIGHT" in first_line.upper() : continue - original_contents = io.open(filename).read() + fd = io.open(filename, encoding="utf-8") + first_line = fd.readline() + second_line = fd.readline() + if "COPYRIGHT (C)" in first_line.upper(): continue + if first_line.startswith("#!") or PYTHON_ENCODE.match( + second_line) != None or PYTHON_ENCODE.match(first_line) != None: + continue + original_contents = io.open(filename, encoding="utf-8").read() new_contents = generate_copyright( COPYRIGHT, lang_type(filename)) + original_contents print('Auto Insert Copyright Header {}'.format(filename)) diff --git a/CMakeLists.txt b/CMakeLists.txt index 00996cb7ed5cc573c42b69be6db369c3654d6d1a..e8ea828dd2a25f5f47b03e92ae86e083d4425dc9 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -31,9 +31,6 @@ if(NOT CMAKE_CROSSCOMPILING) endif(NOT CMAKE_CROSSCOMPILING) find_package(Git REQUIRED) find_package(Threads REQUIRED) -if(NOT ANDROID AND NOT IOS) - find_package(Boost QUIET) -endif() include(simd) @@ -42,7 +39,7 @@ option(WITH_GPU "Compile PaddlePaddle with NVIDIA GPU" ${CUDA_F option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND}) option(WITH_MKL "Compile PaddlePaddle with MKL support." ${AVX_FOUND}) option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" ON) -option(WITH_TESTING "Compile PaddlePaddle with unit testing" ON) +option(WITH_TESTING "Compile PaddlePaddle with unit testing" OFF) option(WITH_SWIG_PY "Compile PaddlePaddle with inference api" ON) option(WITH_STYLE_CHECK "Compile PaddlePaddle with style check" ON) option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON) @@ -55,6 +52,8 @@ option(WITH_COVERAGE "Compile PaddlePaddle with code coverage" OFF) option(COVERALLS_UPLOAD "Package code coverage data to coveralls" OFF) option(ON_TRAVIS "Exclude special unit test on Travis CI" OFF) option(WITH_C_API "Compile PaddlePaddle with C-API(Prediction)" OFF) +# TODO: Only compile PaddlePaddle fluid version by WITH_FLUID option. +option(WITH_FLUID "Compile PaddlePaddle fluid only(TODO)" ON) option(WITH_GOLANG "Compile PaddlePaddle with GOLANG" OFF) option(GLIDE_INSTALL "Download and install go dependencies " ON) option(USE_NNPACK "Compile PaddlePaddle with NNPACK library" OFF) @@ -107,6 +106,10 @@ if (WITH_C_API AND WITH_PYTHON) "different Python interpreter from compiling.") endif() +if (WITH_C_API) + set(WITH_FLUID OFF CACHE STRING "Disable install fluid when compile the C_API" FORCE) +endif() + if(MOBILE_INFERENCE) set(THIRD_PARTY_BUILD_TYPE MinSizeRel) else() @@ -134,6 +137,7 @@ include(external/openblas) # download, build, install openblas include(external/mkldnn) # download, build, install mkldnn include(external/swig) # download, build, install swig include(external/warpctc) # download, build, install warpctc +include(external/boost) # download, build, install boost include(external/any) # download libn::any include(external/eigen) # download eigen3 include(external/pybind11) # download pybind11 @@ -158,7 +162,6 @@ include_directories("${PADDLE_SOURCE_DIR}") include_directories("${PADDLE_SOURCE_DIR}/paddle/cuda/include") include_directories("${CMAKE_CURRENT_BINARY_DIR}/proto") include_directories("${CMAKE_CURRENT_BINARY_DIR}/go/pserver/client/c") -include_directories(${Boost_INCLUDE_DIRS}) set(EXTERNAL_LIBS ${GFLAGS_LIBRARIES} diff --git a/CODE_OF_CONDUCT.md b/CODE_OF_CONDUCT.md new file mode 100644 index 0000000000000000000000000000000000000000..54131b48eca463aef817a4b96ba1b64de4b60aab --- /dev/null +++ b/CODE_OF_CONDUCT.md @@ -0,0 +1,46 @@ +# Contributor Covenant Code of Conduct + +## Our Pledge + +In the interest of fostering an open and welcoming environment, we as contributors and maintainers pledge to making participation in our project and our community a harassment-free experience for everyone, regardless of age, body size, disability, ethnicity, gender identity and expression, level of experience, nationality, personal appearance, race, religion, or sexual identity and orientation. + +## Our Standards + +Examples of behavior that contributes to creating a positive environment include: + +* Using welcoming and inclusive language +* Being respectful of differing viewpoints and experiences +* Gracefully accepting constructive criticism +* Focusing on what is best for the community +* Showing empathy towards other community members + +Examples of unacceptable behavior by participants include: + +* The use of sexualized language or imagery and unwelcome sexual attention or advances +* Trolling, insulting/derogatory comments, and personal or political attacks +* Public or private harassment +* Publishing others' private information, such as a physical or electronic address, without explicit permission +* Other conduct which could reasonably be considered inappropriate in a professional setting + +## Our Responsibilities + +Project maintainers are responsible for clarifying the standards of acceptable behavior and are expected to take appropriate and fair corrective action in response to any instances of unacceptable behavior. + +Project maintainers have the right and responsibility to remove, edit, or reject comments, commits, code, wiki edits, issues, and other contributions that are not aligned to this Code of Conduct, or to ban temporarily or permanently any contributor for other behaviors that they deem inappropriate, threatening, offensive, or harmful. + +## Scope + +This Code of Conduct applies both within project spaces and in public spaces when an individual is representing the project or its community. Examples of representing a project or community include using an official project e-mail address, posting via an official social media account, or acting as an appointed representative at an online or offline event. Representation of a project may be further defined and clarified by project maintainers. + +## Enforcement + +Instances of abusive, harassing, or otherwise unacceptable behavior may be reported by contacting the project team at paddle-dev@baidu.com. The project team will review and investigate all complaints, and will respond in a way that it deems appropriate to the circumstances. The project team is obligated to maintain confidentiality with regard to the reporter of an incident. Further details of specific enforcement policies may be posted separately. + +Project maintainers who do not follow or enforce the Code of Conduct in good faith may face temporary or permanent repercussions as determined by other members of the project's leadership. + +## Attribution + +This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4, available at [http://contributor-covenant.org/version/1/4][version] + +[homepage]: http://contributor-covenant.org +[version]: http://contributor-covenant.org/version/1/4/ diff --git a/CODE_OF_CONDUCT_cn.md b/CODE_OF_CONDUCT_cn.md new file mode 100644 index 0000000000000000000000000000000000000000..2be794f1f324cf9b6bc304d4e5812076b56f4551 --- /dev/null +++ b/CODE_OF_CONDUCT_cn.md @@ -0,0 +1,50 @@ +# 参与者公约 + +## 我们的保证 + +为了促进一个开放透明且友好的环境，我们作为贡献者和维护者保证：无论年龄、种族、民族、性别认同和表达（方式）、体型、身体健全与否、经验水平、国籍、个人表现、宗教或性别取向，参与者在我们项目和社区中都免于骚扰。 + +## 我们的标准 + +有助于创造正面环境的行为包括但不限于： +* 使用友好和包容性语言 +* 尊重不同的观点和经历 +* 耐心地接受建设性批评 +* 关注对社区最有利的事情 +* 友善对待其他社区成员 + +身为参与者不能接受的行为包括但不限于： +* 使用与性有关的言语或是图像，以及不受欢迎的性骚扰 +* 捣乱/煽动/造谣的行为或进行侮辱/贬损的评论，人身攻击及政治攻击 +* 公开或私下的骚扰 +* 未经许可地发布他人的个人资料，例如住址或是电子地址 +* 其他可以被合理地认定为不恰当或者违反职业操守的行为 + +## 我们的责任 + +项目维护者有责任为「可接受的行为」标准做出诠释，以及对已发生的不被接受的行为采取恰当且公平的纠正措施。 + +项目维护者有权利及责任去删除、编辑、拒绝与本行为标准有所违背的评论(comments)、提交(commits)、代码、wiki 编辑、问题(issues)和其他贡献，以及项目维护者可暂时或永久性的禁止任何他们认为有不适当、威胁、冒犯、有害行为的贡献者。 + +## 使用范围 + +当一个人代表该项目或是其社区时，本行为标准适用于其项目平台和公共平台。 + +代表项目或是社区的情况，举例来说包括使用官方项目的电子邮件地址、通过官方的社区媒体账号发布或线上或线下事件中担任指定代表。 + +该项目的呈现方式可由其项目维护者进行进一步的定义及解释。 + +## 强制执行 + +可以通过paddle-dev@baidu.com，来联系项目团队来举报滥用、骚扰或其他不被接受的行为。 + +任何维护团队认为有必要且适合的所有投诉都将进行审查及调查，并做出相对应的回应。项目小组有对事件回报者有保密的义务。具体执行的方针近一步细节可能会单独公布。 + +没有切实地遵守或是执行本行为标准的项目维护人员，可能会因项目领导人或是其他成员的决定，暂时或是永久地取消其参与资格。 + +## 来源 + +本行为标准改编自[贡献者公约][主页]，版本 1.4 +可在此观看https://www.contributor-covenant.org/zh-cn/version/1/4/code-of-conduct.html + +[主页]: https://www.contributor-covenant.org diff --git a/Dockerfile b/Dockerfile index 857d3f3e5f64791146741ffb29feabfcb2ecbb84..6ac9901ac6cea12e97047efdfb6272c957f166ae 100644 --- a/Dockerfile +++ b/Dockerfile @@ -27,7 +27,7 @@ RUN apt-get update && \ curl sed grep graphviz libjpeg-dev zlib1g-dev \ python-matplotlib gcc-4.8 g++-4.8 \ automake locales clang-format swig doxygen cmake \ - liblapack-dev liblapacke-dev libboost-dev \ + liblapack-dev liblapacke-dev \ clang-3.8 llvm-3.8 libclang-3.8-dev \ net-tools libtool && \ apt-get clean -y diff --git a/adversarial/README.md b/adversarial/README.md deleted file mode 100644 index 51da21918a9d6e2192a2e03eabef4fde97896bc5..0000000000000000000000000000000000000000 --- a/adversarial/README.md +++ /dev/null @@ -1,9 +0,0 @@ -# Advbox - -Advbox is a Python toolbox to create adversarial examples that fool neural networks. It requires Python and paddle. - -## How to use - -1. train a model and save it's parameters. (like fluid_mnist.py) -2. load the parameters which is trained in step1, then reconstruct the model.(like mnist_tutorial_fgsm.py) -3. use advbox to generate the adversarial sample. diff --git a/adversarial/advbox/__init__.py b/adversarial/advbox/__init__.py deleted file mode 100644 index f56f14f18dafdfe1e712cea178a63f09a087b587..0000000000000000000000000000000000000000 --- a/adversarial/advbox/__init__.py +++ /dev/null @@ -1,16 +0,0 @@ -# Copyright (c) 2017 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" - A set of tools for generating adversarial example on paddle platform -""" diff --git a/adversarial/advbox/attacks/base.py b/adversarial/advbox/attacks/base.py deleted file mode 100644 index 000baa48f626c7dddce49502d20c499f6424cd06..0000000000000000000000000000000000000000 --- a/adversarial/advbox/attacks/base.py +++ /dev/null @@ -1,52 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -""" -The base model of the model. -""" -from abc import ABCMeta, abstractmethod - - -class Attack(object): - """ - Abstract base class for adversarial attacks. `Attack` represent an adversarial attack - which search an adversarial example. subclass should implement the _apply() method. - - Args: - model(Model): an instance of the class advbox.base.Model. - - """ - __metaclass__ = ABCMeta - - def __init__(self, model): - self.model = model - - def __call__(self, image_label): - """ - Generate the adversarial sample. - - Args: - image_label(list): The image and label tuple list with one element. - """ - adv_img = self._apply(image_label) - return adv_img - - @abstractmethod - def _apply(self, image_label): - """ - Search an adversarial example. - - Args: - image_batch(list): The image and label tuple list with one element. - """ - raise NotImplementedError diff --git a/adversarial/advbox/attacks/gradientsign.py b/adversarial/advbox/attacks/gradientsign.py deleted file mode 100644 index cc26ffb69020a87f559c537f03de84f7c2bea2de..0000000000000000000000000000000000000000 --- a/adversarial/advbox/attacks/gradientsign.py +++ /dev/null @@ -1,87 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -""" -This module provide the attack method for FGSM's implement. -""" -from __future__ import division -import numpy as np -from collections import Iterable -from .base import Attack - - -class GradientSignAttack(Attack): - """ - This attack was originally implemented by Goodfellow et al. (2015) with the - infinity norm (and is known as the "Fast Gradient Sign Method"). This is therefore called - the Fast Gradient Method. - Paper link: https://arxiv.org/abs/1412.6572 - """ - - def _apply(self, image_label, epsilons=1000): - assert len(image_label) == 1 - pre_label = np.argmax(self.model.predict(image_label)) - - min_, max_ = self.model.bounds() - gradient = self.model.gradient(image_label) - gradient_sign = np.sign(gradient) * (max_ - min_) - - if not isinstance(epsilons, Iterable): - epsilons = np.linspace(0, 1, num=epsilons + 1) - - for epsilon in epsilons: - adv_img = image_label[0][0].reshape( - gradient_sign.shape) + epsilon * gradient_sign - adv_img = np.clip(adv_img, min_, max_) - adv_label = np.argmax(self.model.predict([(adv_img, 0)])) - if pre_label != adv_label: - return adv_img - - -FGSM = GradientSignAttack - - -class IteratorGradientSignAttack(Attack): - """ - This attack was originally implemented by Alexey Kurakin(Google Brain). - Paper link: https://arxiv.org/pdf/1607.02533.pdf - """ - - def _apply(self, image_label, epsilons=100, steps=10): - """ - Apply the iterative gradient sign attack. - Args: - image_label(list): The image and label tuple list of one element. - epsilons(list|tuple|int): The epsilon (input variation parameter). - steps(int): The number of iterator steps. - Return: - numpy.ndarray: The adversarail sample generated by the algorithm. - """ - assert len(image_label) == 1 - pre_label = np.argmax(self.model.predict(image_label)) - gradient = self.model.gradient(image_label) - min_, max_ = self.model.bounds() - - if not isinstance(epsilons, Iterable): - epsilons = np.linspace(0, 1, num=epsilons + 1) - - for epsilon in epsilons: - adv_img = image_label[0][0].reshape(gradient.shape) - for _ in range(steps): - gradient = self.model.gradient([(adv_img, image_label[0][1])]) - gradient_sign = np.sign(gradient) * (max_ - min_) - adv_img = adv_img + epsilon * gradient_sign - adv_img = np.clip(adv_img, min_, max_) - adv_label = np.argmax(self.model.predict([(adv_img, 0)])) - if pre_label != adv_label: - return adv_img diff --git a/adversarial/advbox/models/__init__.py b/adversarial/advbox/models/__init__.py deleted file mode 100644 index eee0f6efd4774b42fcd082eb06d1398d2ee51bc4..0000000000000000000000000000000000000000 --- a/adversarial/advbox/models/__init__.py +++ /dev/null @@ -1,16 +0,0 @@ -# Copyright (c) 2017 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" -Paddle model for target of attack -""" diff --git a/adversarial/advbox/models/base.py b/adversarial/advbox/models/base.py deleted file mode 100644 index 084e563f7b423f54f350b90649b04acc17b2db97..0000000000000000000000000000000000000000 --- a/adversarial/advbox/models/base.py +++ /dev/null @@ -1,103 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -""" -The base model of the model. -""" -from abc import ABCMeta -import abc - -abstractmethod = abc.abstractmethod - - -class Model(object): - """ - Base class of model to provide attack. - - - Args: - bounds(tuple): The lower and upper bound for the image pixel. - channel_axis(int): The index of the axis that represents the color channel. - preprocess(tuple): Two element tuple used to preprocess the input. First - substract the first element, then divide the second element. - """ - __metaclass__ = ABCMeta - - def __init__(self, bounds, channel_axis, preprocess=None): - assert len(bounds) == 2 - assert channel_axis in [0, 1, 2, 3] - - if preprocess is None: - preprocess = (0, 1) - self._bounds = bounds - self._channel_axis = channel_axis - self._preprocess = preprocess - - def bounds(self): - """ - Return the upper and lower bounds of the model. - """ - return self._bounds - - def channel_axis(self): - """ - Return the channel axis of the model. - """ - return self._channel_axis - - def _process_input(self, input_): - res = input_ - sub, div = self._preprocess - if sub != 0: - res = input_ - sub - assert div != 0 - if div != 1: - res /= div - return res - - @abstractmethod - def predict(self, image_batch): - """ - Calculate the prediction of the image batch. - - Args: - image_batch(numpy.ndarray): image batch of shape (batch_size, height, width, channels). - - Return: - numpy.ndarray: predictions of the images with shape (batch_size, num_of_classes). - """ - raise NotImplementedError - - @abstractmethod - def num_classes(self): - """ - Determine the number of the classes - - Return: - int: the number of the classes - """ - raise NotImplementedError - - @abstractmethod - def gradient(self, image_batch): - """ - Calculate the gradient of the cross-entropy loss w.r.t the image. - - Args: - image_batch(list): The image and label tuple list. - - Return: - numpy.ndarray: gradient of the cross-entropy loss w.r.t the image with - the shape (height, width, channel). - """ - raise NotImplementedError diff --git a/adversarial/advbox/models/paddle.py b/adversarial/advbox/models/paddle.py deleted file mode 100644 index 4048b47f897000c1b004cb05f8bbca985d5bbbb8..0000000000000000000000000000000000000000 --- a/adversarial/advbox/models/paddle.py +++ /dev/null @@ -1,114 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from __future__ import absolute_import - -import numpy as np -import paddle.v2 as paddle -import paddle.v2.fluid as fluid -from paddle.v2.fluid.framework import program_guard - -from .base import Model - - -class PaddleModel(Model): - """ - Create a PaddleModel instance. - When you need to generate a adversarial sample, you should construct an instance of PaddleModel. - - Args: - program(paddle.v2.fluid.framework.Program): The program of the model which generate the adversarial sample. - input_name(string): The name of the input. - logits_name(string): The name of the logits. - predict_name(string): The name of the predict. - cost_name(string): The name of the loss in the program. - """ - - def __init__(self, - program, - input_name, - logits_name, - predict_name, - cost_name, - bounds, - channel_axis=3, - preprocess=None): - super(PaddleModel, self).__init__( - bounds=bounds, channel_axis=channel_axis, preprocess=preprocess) - - if preprocess is None: - preprocess = (0, 1) - - self._program = program - self._place = fluid.CPUPlace() - self._exe = fluid.Executor(self._place) - - self._input_name = input_name - self._logits_name = logits_name - self._predict_name = predict_name - self._cost_name = cost_name - - # gradient - loss = self._program.block(0).var(self._cost_name) - param_grads = fluid.backward.append_backward( - loss, parameter_list=[self._input_name]) - self._gradient = dict(param_grads)[self._input_name] - - def predict(self, image_batch): - """ - Predict the label of the image_batch. - - Args: - image_batch(list): The image and label tuple list. - Return: - numpy.ndarray: predictions of the images with shape (batch_size, num_of_classes). - """ - feeder = fluid.DataFeeder( - feed_list=[self._input_name, self._logits_name], - place=self._place, - program=self._program) - predict_var = self._program.block(0).var(self._predict_name) - predict = self._exe.run(self._program, - feed=feeder.feed(image_batch), - fetch_list=[predict_var]) - return predict - - def num_classes(self): - """ - Calculate the number of classes of the output label. - - Return: - int: the number of classes - """ - predict_var = self._program.block(0).var(self._predict_name) - assert len(predict_var.shape) == 2 - return predict_var.shape[1] - - def gradient(self, image_batch): - """ - Calculate the gradient of the loss w.r.t the input. - - Args: - image_batch(list): The image and label tuple list. - Return: - list: The list of the gradient of the image. - """ - feeder = fluid.DataFeeder( - feed_list=[self._input_name, self._logits_name], - place=self._place, - program=self._program) - - grad, = self._exe.run(self._program, - feed=feeder.feed(image_batch), - fetch_list=[self._gradient]) - return grad diff --git a/adversarial/fluid_mnist.py b/adversarial/fluid_mnist.py deleted file mode 100644 index f8c7fe8d0ef6a6b0756ef73c14d8937b9cd1a738..0000000000000000000000000000000000000000 --- a/adversarial/fluid_mnist.py +++ /dev/null @@ -1,99 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -""" -CNN on mnist data using fluid api of paddlepaddle -""" -import paddle.v2 as paddle -import paddle.v2.fluid as fluid - - -def mnist_cnn_model(img): - """ - Mnist cnn model - - Args: - img(Varaible): the input image to be recognized - - Returns: - Variable: the label prediction - """ - conv_pool_1 = fluid.nets.simple_img_conv_pool( - input=img, - num_filters=20, - filter_size=5, - pool_size=2, - pool_stride=2, - act='relu') - - conv_pool_2 = fluid.nets.simple_img_conv_pool( - input=conv_pool_1, - num_filters=50, - filter_size=5, - pool_size=2, - pool_stride=2, - act='relu') - - logits = fluid.layers.fc(input=conv_pool_2, size=10, act='softmax') - return logits - - -def main(): - """ - Train the cnn model on mnist datasets - """ - img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32') - label = fluid.layers.data(name='label', shape=[1], dtype='int64') - logits = mnist_cnn_model(img) - cost = fluid.layers.cross_entropy(input=logits, label=label) - avg_cost = fluid.layers.mean(x=cost) - optimizer = fluid.optimizer.Adam(learning_rate=0.01) - optimizer.minimize(avg_cost) - - accuracy = fluid.evaluator.Accuracy(input=logits, label=label) - - BATCH_SIZE = 50 - PASS_NUM = 3 - ACC_THRESHOLD = 0.98 - LOSS_THRESHOLD = 10.0 - train_reader = paddle.batch( - paddle.reader.shuffle( - paddle.dataset.mnist.train(), buf_size=500), - batch_size=BATCH_SIZE) - - place = fluid.CPUPlace() - exe = fluid.Executor(place) - feeder = fluid.DataFeeder(feed_list=[img, label], place=place) - exe.run(fluid.default_startup_program()) - - for pass_id in range(PASS_NUM): - accuracy.reset(exe) - for data in train_reader(): - loss, acc = exe.run(fluid.default_main_program(), - feed=feeder.feed(data), - fetch_list=[avg_cost] + accuracy.metrics) - pass_acc = accuracy.eval(exe) - print("pass_id=" + str(pass_id) + " acc=" + str(acc) + " pass_acc=" - + str(pass_acc)) - if loss < LOSS_THRESHOLD and pass_acc > ACC_THRESHOLD: - break - - pass_acc = accuracy.eval(exe) - print("pass_id=" + str(pass_id) + " pass_acc=" + str(pass_acc)) - fluid.io.save_params( - exe, dirname='./mnist', main_program=fluid.default_main_program()) - print('train mnist done') - - -if __name__ == '__main__': - main() diff --git a/adversarial/mnist_tutorial_fgsm.py b/adversarial/mnist_tutorial_fgsm.py deleted file mode 100644 index c63e030cd826abe24eacab21394c612b7c2d9495..0000000000000000000000000000000000000000 --- a/adversarial/mnist_tutorial_fgsm.py +++ /dev/null @@ -1,100 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -""" -FGSM demos on mnist using advbox tool. -""" -import paddle.v2 as paddle -import paddle.v2.fluid as fluid -import matplotlib.pyplot as plt -import numpy as np - -from advbox.models.paddle import PaddleModel -from advbox.attacks.gradientsign import GradientSignAttack - - -def cnn_model(img): - """ - Mnist cnn model - Args: - img(Varaible): the input image to be recognized - Returns: - Variable: the label prediction - """ - #conv1 = fluid.nets.conv2d() - conv_pool_1 = fluid.nets.simple_img_conv_pool( - input=img, - num_filters=20, - filter_size=5, - pool_size=2, - pool_stride=2, - act='relu') - - conv_pool_2 = fluid.nets.simple_img_conv_pool( - input=conv_pool_1, - num_filters=50, - filter_size=5, - pool_size=2, - pool_stride=2, - act='relu') - - logits = fluid.layers.fc(input=conv_pool_2, size=10, act='softmax') - return logits - - -def main(): - """ - Advbox demo which demonstrate how to use advbox. - """ - IMG_NAME = 'img' - LABEL_NAME = 'label' - - img = fluid.layers.data(name=IMG_NAME, shape=[1, 28, 28], dtype='float32') - # gradient should flow - img.stop_gradient = False - label = fluid.layers.data(name=LABEL_NAME, shape=[1], dtype='int64') - logits = cnn_model(img) - cost = fluid.layers.cross_entropy(input=logits, label=label) - avg_cost = fluid.layers.mean(x=cost) - - place = fluid.CPUPlace() - exe = fluid.Executor(place) - - BATCH_SIZE = 1 - train_reader = paddle.batch( - paddle.reader.shuffle( - paddle.dataset.mnist.train(), buf_size=500), - batch_size=BATCH_SIZE) - feeder = fluid.DataFeeder( - feed_list=[IMG_NAME, LABEL_NAME], - place=place, - program=fluid.default_main_program()) - - fluid.io.load_params( - exe, "./mnist/", main_program=fluid.default_main_program()) - - # advbox demo - m = PaddleModel(fluid.default_main_program(), IMG_NAME, LABEL_NAME, - logits.name, avg_cost.name, (-1, 1)) - att = GradientSignAttack(m) - for data in train_reader(): - # fgsm attack - adv_img = att(data) - plt.imshow(n[0][0], cmap='Greys_r') - plt.show() - #np.save('adv_img', adv_img) - break - - -if __name__ == '__main__': - main() diff --git a/benchmark/paddle/image/alexnet.py b/benchmark/paddle/image/alexnet.py index 07f478d8fa4e1ac4c584d4c410f75555c7926d4b..70296081877588885a7a6d4d8491409b41a8d378 100644 --- a/benchmark/paddle/image/alexnet.py +++ b/benchmark/paddle/image/alexnet.py @@ -1,17 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -#!/usr/bin/env python +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/benchmark/paddle/image/googlenet.py b/benchmark/paddle/image/googlenet.py index 3241be9c5f56d3e3b422081e26878ae690a33268..2a850ccb7f2c75b467554181fc5f4aa8f2b97a09 100644 --- a/benchmark/paddle/image/googlenet.py +++ b/benchmark/paddle/image/googlenet.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from paddle.trainer_config_helpers import * diff --git a/benchmark/paddle/image/provider.py b/benchmark/paddle/image/provider.py index 220c4bee35c89e90d7ba5edbb9b92632a0215811..21e0d381aab24d21d91cdcc4aa630f6107405f9e 100644 --- a/benchmark/paddle/image/provider.py +++ b/benchmark/paddle/image/provider.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import io, os import random import numpy as np diff --git a/benchmark/paddle/image/resnet.py b/benchmark/paddle/image/resnet.py index acc6d31d4bb4eeca3e1d98e7b42ea8a2bd8bc90b..2846e4763f1cda4602f03af5ec649d57ee6cf0d8 100644 --- a/benchmark/paddle/image/resnet.py +++ b/benchmark/paddle/image/resnet.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from paddle.trainer_config_helpers import * diff --git a/benchmark/paddle/image/smallnet_mnist_cifar.py b/benchmark/paddle/image/smallnet_mnist_cifar.py index 64a5da3220bf3294b88c015fe133e7b04573d954..58879c454f37991405d83bbb593bb5d1e977ff53 100644 --- a/benchmark/paddle/image/smallnet_mnist_cifar.py +++ b/benchmark/paddle/image/smallnet_mnist_cifar.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from paddle.trainer_config_helpers import * diff --git a/benchmark/paddle/image/vgg.py b/benchmark/paddle/image/vgg.py index a357207a6282aa7864a7ecffa8c54e7b2360b45e..ca0a6798fb8c35b68cf84d263855955eb93ba0b0 100644 --- a/benchmark/paddle/image/vgg.py +++ b/benchmark/paddle/image/vgg.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from paddle.trainer_config_helpers import * diff --git a/benchmark/paddle/rnn/imdb.py b/benchmark/paddle/rnn/imdb.py index fc4ed4025f9ed2e0a32a1709ff8df4af53521196..c3b5faa19aaa325aaaf9cd9cc8f757d3f3b3bdcb 100755 --- a/benchmark/paddle/rnn/imdb.py +++ b/benchmark/paddle/rnn/imdb.py @@ -1,3 +1,17 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import six.moves.cPickle as pickle import gzip diff --git a/benchmark/paddle/rnn/provider.py b/benchmark/paddle/rnn/provider.py index c03df3a0026447d5ab239f90600975751d287799..f35cd5b079ff09ac35b171ff3032ecc5adabc947 100644 --- a/benchmark/paddle/rnn/provider.py +++ b/benchmark/paddle/rnn/provider.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import io, os import random import numpy as np diff --git a/benchmark/paddle/rnn/rnn.py b/benchmark/paddle/rnn/rnn.py index 97005f2c351ff2301a1b834e79931bb6ccc7abc8..83eb3e565473f7e7e91cddeaa3cd2aafb7e3df2c 100755 --- a/benchmark/paddle/rnn/rnn.py +++ b/benchmark/paddle/rnn/rnn.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from paddle.trainer_config_helpers import * diff --git a/benchmark/tensorflow/image/alexnet.py b/benchmark/tensorflow/image/alexnet.py index edf462e6a188f4a83d7276086fcacfb644b562a1..a37d7e7c62282891d67cee74cac1408eac1244f7 100644 --- a/benchmark/tensorflow/image/alexnet.py +++ b/benchmark/tensorflow/image/alexnet.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from six.moves import xrange # pylint: disable=redefined-builtin from datetime import datetime import math diff --git a/benchmark/tensorflow/image/alexnet_multi_gpu.py b/benchmark/tensorflow/image/alexnet_multi_gpu.py index 90b8f16bca027fd14a62c2e65bdb606c6b7d79d9..2ebab8fb60d471cd6de6d332d81608f2a992b9be 100644 --- a/benchmark/tensorflow/image/alexnet_multi_gpu.py +++ b/benchmark/tensorflow/image/alexnet_multi_gpu.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from six.moves import xrange # pylint: disable=redefined-builtin from datetime import datetime import math diff --git a/benchmark/tensorflow/image/googlenet.py b/benchmark/tensorflow/image/googlenet.py index 55431eceb3ce41c05fc5fb2d417135494adf3b2c..1202cbb171efd74dec1fc6690a82e1f5126685f0 100644 --- a/benchmark/tensorflow/image/googlenet.py +++ b/benchmark/tensorflow/image/googlenet.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from six.moves import xrange from datetime import datetime import math diff --git a/benchmark/tensorflow/image/googlenet_multi_gpu.py b/benchmark/tensorflow/image/googlenet_multi_gpu.py index 44de3800a8a42d90debae2c567795789f3eb0a7d..f06437eb6c82bf0dfbb9a76799e83182ec5ee888 100644 --- a/benchmark/tensorflow/image/googlenet_multi_gpu.py +++ b/benchmark/tensorflow/image/googlenet_multi_gpu.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from six.moves import xrange # pylint: disable=redefined-builtin from datetime import datetime import math diff --git a/benchmark/tensorflow/image/smallnet_mnist_cifar.py b/benchmark/tensorflow/image/smallnet_mnist_cifar.py index 0858b5f9c9c60264c0427c9e1fbdfcd167cf418e..558c68575f4ae3ceba7119ed081549ad63e208d8 100644 --- a/benchmark/tensorflow/image/smallnet_mnist_cifar.py +++ b/benchmark/tensorflow/image/smallnet_mnist_cifar.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from six.moves import xrange # pylint: disable=redefined-builtin from datetime import datetime import math diff --git a/benchmark/tensorflow/rnn/reader.py b/benchmark/tensorflow/rnn/reader.py index 710940c9ae24eb307ed864aa2b17d66c1d6ee29b..9660d3c22b3a16954b0a1d09d38cf033824f0a5f 100755 --- a/benchmark/tensorflow/rnn/reader.py +++ b/benchmark/tensorflow/rnn/reader.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import os.path import io import numpy as np diff --git a/benchmark/tensorflow/rnn/rnn.py b/benchmark/tensorflow/rnn/rnn.py index 507481b9ccd3a037d396f91bb860255a343905c5..f288083e13656563b511980553245142efec4e65 100755 --- a/benchmark/tensorflow/rnn/rnn.py +++ b/benchmark/tensorflow/rnn/rnn.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from six.moves import xrange # pylint: disable=redefined-builtin import math diff --git a/benchmark/tensorflow/rnn/rnn_multi_gpu.py b/benchmark/tensorflow/rnn/rnn_multi_gpu.py index f24cbaef62d94f2d94c243bd3c8ffcf1dcec7614..eabee4fa8fe6325212ace1c11be4862cd2720b08 100755 --- a/benchmark/tensorflow/rnn/rnn_multi_gpu.py +++ b/benchmark/tensorflow/rnn/rnn_multi_gpu.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python from six.moves import xrange # pylint: disable=redefined-builtin import re diff --git a/cmake/external/boost.cmake b/cmake/external/boost.cmake new file mode 100644 index 0000000000000000000000000000000000000000..c70d83b3f4bb24740ed67b4e2f98a3ced26d1648 --- /dev/null +++ b/cmake/external/boost.cmake @@ -0,0 +1,51 @@ +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +include(ExternalProject) + +set(BOOST_PROJECT "extern_boost") +set(BOOST_VER "1.41.0") +set(BOOST_TAR "boost_1_41_0") +set(BOOST_URL "http://sourceforge.net/projects/boost/files/boost/${BOOST_VER}/${BOOST_TAR}.tar.gz") +set(BOOST_SOURCES_DIR ${THIRD_PARTY_PATH}/boost) +set(BOOST_DOWNLOAD_DIR "${BOOST_SOURCES_DIR}/src/${BOOST_PROJECT}") +set(BOOST_INCLUDE_DIR "${BOOST_DOWNLOAD_DIR}/${BOOST_TAR}" CACHE PATH "boost include directory." FORCE) + +include_directories(${BOOST_INCLUDE_DIR}) + +ExternalProject_Add( + ${BOOST_PROJECT} + ${EXTERNAL_PROJECT_LOG_ARGS} + DOWNLOAD_DIR ${BOOST_DOWNLOAD_DIR} + DOWNLOAD_COMMAND wget --no-check-certificate ${BOOST_URL} -c -q -O ${BOOST_TAR}.tar.gz + && tar zxf ${BOOST_TAR}.tar.gz + DOWNLOAD_NO_PROGRESS 1 + PREFIX ${BOOST_SOURCES_DIR} + CONFIGURE_COMMAND "" + BUILD_COMMAND "" + INSTALL_COMMAND "" + UPDATE_COMMAND "" +) + +if (${CMAKE_VERSION} VERSION_LESS "3.3.0") + set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/boost_dummy.c) + file(WRITE ${dummyfile} "const char *dummy = \"${dummyfile}\";") + add_library(boost STATIC ${dummyfile}) +else() + add_library(boost INTERFACE) +endif() + +add_dependencies(boost ${BOOST_PROJECT}) +list(APPEND external_project_dependencies boost) +set(Boost_INCLUDE_DIR ${BOOST_INCLUDE_DIR}) diff --git a/cmake/external/eigen.cmake b/cmake/external/eigen.cmake index c4712f19eb80b34ffbf713d2b13fc0c775312af1..d49c8d601102cf865287c33349bff5eee6a90f6d 100644 --- a/cmake/external/eigen.cmake +++ b/cmake/external/eigen.cmake @@ -1,8 +1,8 @@ INCLUDE(ExternalProject) SET(EIGEN_SOURCE_DIR ${THIRD_PARTY_PATH}/eigen3) - -INCLUDE_DIRECTORIES(${EIGEN_SOURCE_DIR}/src/extern_eigen3) +SET(EIGEN_INCLUDE_DIR ${EIGEN_SOURCE_DIR}/src/extern_eigen3) +INCLUDE_DIRECTORIES(${EIGEN_INCLUDE_DIR}) ExternalProject_Add( extern_eigen3 @@ -28,3 +28,9 @@ endif() add_dependencies(eigen3 extern_eigen3) LIST(APPEND external_project_dependencies eigen3) + +IF(NOT WITH_C_API AND WITH_FLUID) + INSTALL(FILES ${EIGEN_INCLUDE_DIR}/Eigen/Core DESTINATION third_party/eigen3/Eigen) + INSTALL(DIRECTORY ${EIGEN_INCLUDE_DIR}/Eigen/src DESTINATION third_party/eigen3/Eigen) + INSTALL(DIRECTORY ${EIGEN_INCLUDE_DIR}/unsupported/Eigen DESTINATION third_party/eigen3/unsupported) +ENDIF() diff --git a/cmake/external/gflags.cmake b/cmake/external/gflags.cmake index d4f252bb9f64c8db82b841fedf0817f5d8596501..60946304541a20809276c3e665d8524baf209006 100644 --- a/cmake/external/gflags.cmake +++ b/cmake/external/gflags.cmake @@ -52,7 +52,7 @@ ADD_DEPENDENCIES(gflags extern_gflags) LIST(APPEND external_project_dependencies gflags) -IF(WITH_C_API) +IF(WITH_C_API OR WITH_FLUID) INSTALL(DIRECTORY ${GFLAGS_INCLUDE_DIR} DESTINATION third_party/gflags) IF(ANDROID) INSTALL(FILES ${GFLAGS_LIBRARIES} DESTINATION third_party/gflags/lib/${ANDROID_ABI}) diff --git a/cmake/external/glog.cmake b/cmake/external/glog.cmake index 0c6b3aafcb4e990b9d4549820137474e5968a7aa..382fbda3b5cfeba893f03871cf65498d20804f36 100644 --- a/cmake/external/glog.cmake +++ b/cmake/external/glog.cmake @@ -68,7 +68,7 @@ LINK_LIBRARIES(glog gflags) LIST(APPEND external_project_dependencies glog) -IF(WITH_C_API) +IF(WITH_C_API OR WITH_FLUID) INSTALL(DIRECTORY ${GLOG_INCLUDE_DIR} DESTINATION third_party/glog) IF(ANDROID) INSTALL(FILES ${GLOG_LIBRARIES} DESTINATION third_party/glog/lib/${ANDROID_ABI}) diff --git a/cmake/external/openblas.cmake b/cmake/external/openblas.cmake index 0e79c0cc7992060cbe3b668ec927936183389eb6..4012a164be1d20bba050fb09749b11afc7b99588 100644 --- a/cmake/external/openblas.cmake +++ b/cmake/external/openblas.cmake @@ -100,6 +100,11 @@ IF(NOT ${CBLAS_FOUND}) \"${CBLAS_INSTALL_DIR}/lib -> ${CMAKE_INSTALL_PREFIX}/${TMP_INSTALL_DIR}\" )" ) + INSTALL(CODE "execute_process( + COMMAND rm -r ${CMAKE_INSTALL_PREFIX}/${TMP_INSTALL_DIR}/cmake + ${CMAKE_INSTALL_PREFIX}/${TMP_INSTALL_DIR}/pkgconfig + )" + ) ENDIF() ENDIF(NOT ${CBLAS_FOUND}) diff --git a/cmake/external/protobuf.cmake b/cmake/external/protobuf.cmake index ff5855052dabaa0b63099cd219f3f04e22f1aa85..365a370a9cfb708379bcff18ae6aa0725d420ae1 100644 --- a/cmake/external/protobuf.cmake +++ b/cmake/external/protobuf.cmake @@ -250,7 +250,7 @@ IF(NOT PROTOBUF_FOUND) SET(PROTOBUF_PROTOC_LIBRARY ${extern_protobuf_PROTOC_LIBRARY} CACHE FILEPATH "protoc library." FORCE) - IF(WITH_C_API) + IF(WITH_C_API OR WITH_FLUID) INSTALL(DIRECTORY ${PROTOBUF_INCLUDE_DIR} DESTINATION third_party/protobuf) IF(ANDROID) INSTALL(FILES ${PROTOBUF_LITE_LIBRARY} DESTINATION third_party/protobuf/lib/${ANDROID_ABI}) diff --git a/cmake/generic.cmake b/cmake/generic.cmake index 585db019d521b1699baadfae31ef95b5059c71b4..18770fe2861380ea1320aef5cb7ec3432147d7ce 100644 --- a/cmake/generic.cmake +++ b/cmake/generic.cmake @@ -224,12 +224,18 @@ function(cc_test TARGET_NAME) if(WITH_TESTING) set(options "") set(oneValueArgs "") - set(multiValueArgs SRCS DEPS) + set(multiValueArgs SRCS DEPS ARGS) cmake_parse_arguments(cc_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) add_executable(${TARGET_NAME} ${cc_test_SRCS}) - target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main paddle_memory gtest gflags) + # Support linking flags: --whole-archive (Linux) / -force_load (MacOS) + target_circle_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main paddle_memory gtest gflags) + if("${cc_test_DEPS}" MATCHES "ARCHIVE_START") + list(REMOVE_ITEM cc_test_DEPS ARCHIVE_START ARCHIVE_END) + endif() add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main paddle_memory gtest gflags) - add_test(NAME ${TARGET_NAME} COMMAND ${TARGET_NAME} WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) + add_test(NAME ${TARGET_NAME} + COMMAND ${TARGET_NAME} ${cc_test_ARGS} + WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) endif() endfunction(cc_test) @@ -457,7 +463,7 @@ endfunction() function(py_test TARGET_NAME) if(WITH_TESTING) - set(options STATIC static SHARED shared) + set(options "") set(oneValueArgs "") set(multiValueArgs SRCS DEPS ARGS) cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) diff --git a/cmake/make_resource.py b/cmake/make_resource.py index d71e82eca2cd10179d5ec498f7fb2aa5da679c9f..4f9f5546b9d4176d1035311cb9e1acf0c0eeccfd 100644 --- a/cmake/make_resource.py +++ b/cmake/make_resource.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import os import re import sys diff --git a/doc/api/v1/data_provider/dataprovider_cn.rst b/doc/api/v1/data_provider/dataprovider_cn.rst deleted file mode 100644 index d08c6b3efacbc35ae274d5b207fe91e747124e79..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/dataprovider_cn.rst +++ /dev/null @@ -1,15 +0,0 @@ -.. _api_dataprovider: - -DataProvider的介绍 -================== - -DataProvider是PaddlePaddle负责提供数据的模块。其作用是将数据传入内存或显存，让神经网络可以进行训练或预测。用户可以通过简单使用Python接口 :ref:`api_pydataprovider2` ，来自定义传数据的过程。如果有更复杂的使用，或者需要更高的效率，用户也可以在C++端自定义一个 ``DataProvider`` 。 - -PaddlePaddle需要用户在网络配置（trainer_config.py）中定义使用哪种DataProvider，并且在DataProvider中实现如何访问训练文件列表（train.list）或测试文件列表（test.list）。 - -- train.list和test.list存放在本地（推荐直接存放到训练目录，以相对路径引用)。一般情况下，两者均为纯文本文件，其中每一行对应一个数据文件地址： - - - 如果数据文件存于本地磁盘，这个地址则为它的绝对路径或相对路径(相对于PaddlePaddle程序运行时的路径)。 - - 地址也可以为hdfs文件路径，或者数据库连接路径等。 - - 由于这个地址会被DataProvider使用，因此，如何解析该地址也是用户自定义DataProvider时需要考虑的地方。 -- 如果没有设置test.list，或设置为None，那么在训练过程中不会执行测试操作；否则，会根据命令行参数指定的测试方式，在训练过程中进行测试，从而防止过拟合。 diff --git a/doc/api/v1/data_provider/dataprovider_en.rst b/doc/api/v1/data_provider/dataprovider_en.rst deleted file mode 100644 index 96efbb1da959daec561009fdcc95d353b191dec8..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/dataprovider_en.rst +++ /dev/null @@ -1,34 +0,0 @@ -Introduction -============== -DataProvider is a module that loads training or testing data into cpu or gpu -memory for the following triaining or testing process. - -For simple use, users can use Python :code:`PyDataProvider` to dynamically reads -the original data in any format or in any form, and then transfer them into a -data format PaddlePaddle requires. The process is extremly flexible and highly -customized, with sacrificing the efficiency only a little. This is extremly -useful when you have to dynamically generate certain kinds of data according to, -for example, the training performance. - -Besides, users also can customize a C++ :code:`DataProvider` for a more -complex usage, or for a higher efficiency. - -The following parameters are required to define in the PaddlePaddle network -configuration file (trainer_config.py): which DataProvider is chosen to used, -and specific parameters for DataProvider, including training file list -(train.list) and testing file list (test.list). - -Train.list and test.list are simply two plain text files, which defines path -of training or testing data. It is recommended that directly placing them into -the training directory, and reference to them by using a relative path ( -relative to the PaddePaddle program). - -Testing or evaluating will not be performed during training if the test.list is -not set or set to None. Otherwise, PaddlePaddle will evaluate the trained model -by the specified tesing data while training, every testing period (a user -defined command line parameter in PaddlePaddle) to prevent over-fitting. - -Each line of train.list and test.list is an absolute or relative path (relative -to the PaddePaddle program runtime) of data file. Fascinatingly more, each line -can also be a HDFS file path or a SQL connection string. As long as the user -assures how to access each file in DataProvider. diff --git a/doc/api/v1/data_provider/pydataprovider2_cn.rst b/doc/api/v1/data_provider/pydataprovider2_cn.rst deleted file mode 100644 index 8f9db31cfb9946e1d2db3872718bd92787d861f0..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/pydataprovider2_cn.rst +++ /dev/null @@ -1,229 +0,0 @@ -.. _api_pydataprovider2: - -PyDataProvider2的使用 -===================== - -PyDataProvider2是PaddlePaddle使用Python提供数据的推荐接口。该接口使用多线程读取数据，并提供了简单的Cache功能；同时可以使用户只关注如何从文件中读取每一条数据，而不用关心数据如何传输，如何存储等等。 - -.. contents:: - -MNIST的使用场景 ---------------- - -我们以MNIST手写识别为例，来说明PyDataProvider2的简单使用场景。 - -样例数据 -++++++++ - -MNIST是一个包含有70,000张灰度图片的数字分类数据集。样例数据 ``mnist_train.txt`` 如下： - -.. literalinclude:: src/mnist_train.txt - -其中每行数据代表一张图片，行内使用 ``;`` 分成两部分。第一部分是图片的标签，为0-9中的一个数字；第二部分是28*28的图片像素灰度值。 对应的 ``train.list`` 即为这个数据文件的名字： - -.. literalinclude:: src/train.list - -dataprovider的使用 -++++++++++++++++++ - -.. literalinclude:: src/mnist_provider.dict.py - -- 首先，引入PaddlePaddle的PyDataProvider2包。 -- 其次，定义一个Python的 `Decorator `_ `@provider`_ 。用于将下一行的数据输入函数标记成一个PyDataProvider2，同时设置它的input_types属性。 - - - `input_types`_：设置这个PyDataProvider2返回什么样的数据。本例根据网络配置中 ``data_layer`` 的名字，显式指定返回的是一个28*28维的稠密浮点数向量和一个[0-9]的10维整数标签。 - - .. literalinclude:: src/mnist_config.py - :lines: 9-10 - - - 注意：如果用户不显示指定返回数据的对应关系，那么PaddlePaddle会根据layer的声明顺序，来确定对应关系。但这个关系可能不正确，所以推荐使用显式指定的方式来设置input_types。 -- 最后，实现数据输入函数（如本例的 ``process`` 函数）。 - - - 该函数的功能是：打开文本文件，读取每一行，将行中的数据转换成与input_types一致的格式，然后返回给PaddlePaddle进程。注意， - - - 返回的顺序需要和input_types中定义的顺序一致。 - - 返回时，必须使用Python关键词 ``yield`` ，相关概念是 ``generator`` 。 - - 一次yield调用，返回一条完整的样本。如果想为一个数据文件返回多条样本，只需要在函数中调用多次yield即可（本例中使用for循环进行多次调用）。 - - - 该函数具有两个参数： - - - settings：在本例中没有使用，具体可以参考 `init_hook`_ 中的说明。 - - filename：为 ``train.list`` 或 ``test.list`` 中的一行，即若干数据文件路径的某一个。 - -网络配置中的调用 -++++++++++++++++ - -在网络配置里，只需要一行代码就可以调用这个PyDataProvider2，如， - -.. literalinclude:: src/mnist_config.py - :lines: 1-7 - -训练数据是 ``train.list`` ，没有测试数据，调用的PyDataProvider2是 ``mnist_provider`` 模块中的 ``process`` 函数。 - -小结 -+++++ - -至此，简单的PyDataProvider2样例就说明完毕了。对用户来说，仅需要知道如何从 **一个文件** 中读取 **一条样本** ，就可以将数据传送给PaddlePaddle了。而PaddlePaddle则会帮用户做以下工作： - -* 将数据组合成Batch进行训练 -* 对训练数据进行Shuffle -* 多线程的数据读取 -* 缓存训练数据到内存(可选) -* CPU->GPU双缓存 - -是不是很简单呢？ - -时序模型的使用场景 ------------------- -样例数据 -++++++++ - -时序模型是指数据的某一维度是一个序列形式，即包含时间步信息。所谓时间步信息，不一定和时间有关系，只是说明数据的顺序是重要的。例如，文本信息就是一个序列数据。 - -本例采用英文情感分类的数据，即将一段英文文本数据，分类成正面情绪和负面情绪两类(用0和1表示)。样例数据 ``sentimental_train.txt`` 如下： - -.. literalinclude:: src/sentimental_train.txt - -dataprovider的使用 -++++++++++++++++++ - -相对MNIST而言，这个dataprovider较复杂，主要原因是增加了初始化机制 `init_hook`_。本例的 ``on_init`` 函数就是根据该机制配置的，它会在dataprovider创建的时候执行。 - -- 其中 ``input_types`` 和在 `@provider`_ 中配置的效果一致。本例中的输入特征是词ID的序列，因此使用 ``integer_value_sequence`` 类型来设置。 -- 将 ``dictionary`` 存入settings对象，在 ``process`` 函数中使用。 dictionary是从网络配置中传入的dict对象，即一个将单词字符串映射到单词ID的字典。 - -.. literalinclude:: src/sentimental_provider.py - -网络配置中的调用 -++++++++++++++++ - -调用这个PyDataProvider2的方法，基本上和MNIST样例一致，除了 - -* 在配置中需要读取外部字典。 -* 在声明DataProvider的时候传入dictionary作为参数。 - -.. literalinclude:: src/sentimental_config.py - :emphasize-lines: 12-14 - -参考(Reference) ---------------- - -@provider -+++++++++ - -``@provider`` 是一个Python的 `Decorator`_ ，可以将某一个函数标记成一个PyDataProvider2。如果不了解 `Decorator`_ 是什么也没关系，只需知道这是一个标记属性的方法就可以了。它包含的属性参数如下: - -* input_types：数据输入格式。具体的格式说明，请参考 `input_types`_ 。 -* should_shuffle：是不是要对数据做Shuffle。训练时默认shuffle，测试时默认不shuffle。 -* min_pool_size：设置内存中最小暂存的数据条数，也是PaddlePaddle所能够保证的shuffle粒度。如果为-1，则会预先读取全部数据到内存中。 -* pool_size： 设置内存中暂存的数据条数。如果为-1（默认），则不在乎内存暂存多少条数据。如果设置，则推荐大于训练时batch size的值，并且在内存足够的情况下越大越好。 -* can_over_batch_size：是否允许暂存略微多余pool_size的数据。由于这样做可以避免很多死锁问题，一般推荐设置成True。 -* calc_batch_size：可以传入一个函数，用于自定义每条数据的batch size（默认为1）。 -* cache： 数据缓存的策略，具体请参考 `cache`_ 。 -* init_hook：初始化时调用的函数，具体请参考 `init_hook`_ 。 -* check：如果为true，会根据input_types检查数据的合法性。 -* check_fail_continue：如果为true，那么当check出数据不合法时，会扔到这条数据，继续训练或预测。（对check=false的情况，没有作用） - -input_types -+++++++++++ - -PaddlePaddle的数据包括四种主要类型，和三种序列模式。 - -四种数据类型： - -* dense_vector：稠密的浮点数向量。 -* sparse_binary_vector：稀疏的01向量，即大部分值为0，但有值的地方必须为1。 -* sparse_float_vector：稀疏的向量，即大部分值为0，但有值的部分可以是任何浮点数。 -* integer：整数标签。 - -三种序列模式： - -* SequenceType.NO_SEQUENCE：不是一条序列 -* SequenceType.SEQUENCE：是一条时间序列 -* SequenceType.SUB_SEQUENCE： 是一条时间序列，且序列的每一个元素还是一个时间序列。 - -不同的数据类型和序列模式返回的格式不同，列表如下： - -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| | NO_SEQUENCE | SEQUENCE | SUB_SEQUENCE | -+======================+=====================+===================================+================================================+ -| dense_vector | [f, f, ...] | [[f, ...], [f, ...], ...] | [[[f, ...], ...], [[f, ...], ...],...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| sparse_binary_vector | [i, i, ...] | [[i, ...], [i, ...], ...] | [[[i, ...], ...], [[i, ...], ...],...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| sparse_float_vector | [(i,f), (i,f), ...] | [[(i,f), ...], [(i,f), ...], ...] | [[[(i,f), ...], ...], [[(i,f), ...], ...],...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| integer_value | i | [i, i, ...] | [[i, ...], [i, ...], ...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ - -其中，f代表一个浮点数，i代表一个整数。 - -注意：对sparse_binary_vector和sparse_float_vector，PaddlePaddle存的是有值位置的索引。例如， - -- 对一个5维非序列的稀疏01向量 ``[0, 1, 1, 0, 0]`` ，类型是sparse_binary_vector，返回的是 ``[1, 2]`` 。 -- 对一个5维非序列的稀疏浮点向量 ``[0, 0.5, 0.7, 0, 0]`` ，类型是sparse_float_vector，返回的是 ``[(1, 0.5), (2, 0.7)]`` 。 - -init_hook -+++++++++ - -init_hook可以传入一个函数。该函数在初始化的时候会被调用，其参数如下: - -* 第一个参数是settings对象，它和数据传入函数的第一个参数（如本例中 ``process`` 函数的 ``settings`` 参数）必须一致。该对象具有以下两个属性： - * settings.input_types：数据输入格式，具体请参考 `input_types`_ 。 - * settings.logger：一个logging对象。 -* 其他参数使用 ``kwargs`` （key word arguments）传入，包括以下两种： - * PaddlePaddle定义的参数: 1）is_train：bool型参数，表示用于训练或预测；2）file_list：所有文件列表。 - * 用户定义的参数：使用args在网络配置中设置。 - -注意：PaddlePaddle保留添加参数的权力，因此init_hook尽量使用 ``**kwargs`` 来接受不使用的函数以保证兼容性。 - -cache -+++++ - -PyDataProvider2提供了两种简单的Cache策略： - -* CacheType.NO_CACHE：不缓存任何数据，每次都会从python端读取数据 -* CacheType.CACHE_PASS_IN_MEM：第一个pass会从python端读取数据，剩下的pass会直接从内存里 - 读取数据。 - - -注意事项 --------- - -可能的内存泄露问题 -++++++++++++++++++ - -PaddlePaddle将train.list中的每一行都传递给process函数，从而生成多个generator。当训练数据非常多时，就会生成非常多的generator。 - -虽然每个generator在没有调用的时候，是几乎不占内存的；但当调用过一次后，generator便会存下当前的上下文(Context)，而这个Context可能会非常大。并且，generator至少需要调用两次才会知道是否停止。所以，即使process函数里面只有一个yield，也需要两次随机选择到相同generator的时候，才会释放该段内存。 - -.. code-block:: python - - def func(): - yield 0 - - f = func() # 创建generator - tmp = next(f) # 调用一次，返回0 - tmp = next(f) # 调用第二次的时候，才会Stop Iteration - -由于顺序调用这些generator不会出现上述问题，因此有两种解决方案： - -1. **最佳推荐**：将样本的地址放入另一个文本文件，train.list写入那个文本文件的地址。即不要将每一个样本都放入train.list。 -2. 在generator的上下文中尽量留下非常少的变量引用，例如 - -.. code-block:: python - - def real_process(fn): - # ... read from fn - return result # 当函数返回的时候，python可以解除掉内部变量的引用。 - - def process(fn): - yield real_process(fn) - -注意：这个问题是PyDataProvider读数据时候的逻辑问题，很难整体修正。 - -内存不够用的情况 -++++++++++++++++ - -PyDataProvider2会尽可能多的使用内存。因此，对于内存较小的机器，推荐使用 ``pool_size`` 变量来设置内存中暂存的数据条。具体请参考 `@provider`_ 中的说明。 - diff --git a/doc/api/v1/data_provider/pydataprovider2_en.rst b/doc/api/v1/data_provider/pydataprovider2_en.rst deleted file mode 100644 index e8fb6292779790765154502bff319ea10ab1e70b..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/pydataprovider2_en.rst +++ /dev/null @@ -1,249 +0,0 @@ -.. _api_pydataprovider2: - -PyDataProvider2 -=============== - -We highly recommand users to use PyDataProvider2 to provide training or testing -data to PaddlePaddle. The user only needs to focus on how to read a single -sample from the original data file by using PyDataProvider2, leaving all of the -trivial work, including, transfering data into cpu/gpu memory, shuffle, binary -serialization to PyDataProvider2. PyDataProvider2 uses multithreading and a -fanscinating but simple cache strategy to optimize the efficiency of the data -providing process. - -DataProvider for the non-sequential model ------------------------------------------ - -Here we use the MNIST handwriting recognition data as an example to illustrate -how to write a simple PyDataProvider. - -MNIST is a handwriting classification data set. It contains 70,000 digital -grayscale images. Labels of the training sample range from 0 to 9. All the -images have been size-normalized and centered into images with the same size -of 28 x 28 pixels. - -A small part of the original data as an example is shown as below: - -.. literalinclude:: src/mnist_train.txt - -Each line of the data contains two parts, separated by :code:`;`. The first part is -label of an image. The second part contains 28x28 pixel float values. - -Just write path of the above data into train.list. It looks like this: - -.. literalinclude:: src/train.list - -The corresponding dataprovider is shown as below: - -.. literalinclude:: src/mnist_provider.dict.py - -The first line imports PyDataProvider2 package. -The main function is the process function, that has two parameters. -The first parameter is the settings, which is not used in this example. -The second parameter is the filename, that is exactly each line of train.list. -This parameter is passed to the process function by PaddlePaddle. - -:code:`@provider` is a Python -`Decorator `_ . -It sets some properties to DataProvider, and constructs a real PaddlePaddle -DataProvider from a very simple user implemented python function. It does not -matter if you are not familiar with `Decorator`_. You can keep it simple by -just taking :code:`@provider` as a fixed mark above the provider function you -implemented. - -`input_types`_ defines the data format that a DataProvider returns. -In this example, it is set to a 28x28-dimensional dense vector and an integer -scalar, whose value ranges from 0 to 9. -`input_types`_ can be set to several kinds of input formats, please refer to the -document of `input_types`_ for more details. - - -The process method is the core part to construct a real DataProvider in -PaddlePaddle. It implements how to open the text file, how to read one sample -from the original text file, convert them into `input_types`_, and give them -back to PaddlePaddle process at line 23. -Note that data yielded by the process function must follow the same order that -`input_types`_ are defined. - - -With the help of PyDataProvider2, user can focus on how to generate ONE traning -sample by using keywords :code:`yield`. -:code:`yield` is a python keyword, and a concept related to it includes -:code:`generator`. - -Only a few lines of codes need to be added into the training configuration file, -you can take this as an example. - -.. literalinclude:: src/mnist_config.py - -Here we specify training data by :code:`train.list`, and no testing data is specified. -The method which actually provide data is :code:`process`. - -User also can use another style to provide data, which defines the -:code:`data_layer`'s name explicitly when `yield`. For example, -the :code:`dataprovider` is shown as below. - -.. literalinclude:: src/mnist_provider.dict.py - :linenos: - -If user did't give the :code:`data_layer`'s name, PaddlePaddle will use -the order of :code:`data_layer` definition roughly to determine which feature to -which :code:`data_layer`. This order may be not correct, so TO DEFINE THE -:code:`data_layer`'s NAMES EXPLICITLY IS THE RECOMMANDED WAY TO PROVIDER DATA. - -Now, this simple example of using PyDataProvider is finished. -The only thing that the user should know is how to generte **one sample** from -**one data file**. -And PaddlePadle will do all of the rest things\: - -* Form a training batch -* Shuffle the training data -* Read data with multithreading -* Cache the training data (Optional) -* CPU-> GPU double buffering. - -Is this cool? - -.. _api_pydataprovider2_sequential_model: - -DataProvider for the sequential model -------------------------------------- -A sequence model takes sequences as its input. A sequence is made up of several -timesteps. The so-called timestep, is not necessary to have something to do -with time. It can also be explained to that the order of data are taken into -consideration into model design and training. -For example, the sentence can be interpreted as a kind of sequence data in NLP -tasks. - -Here is an example on data proivider for English sentiment classification data. -The original input data are simple English text, labeled into positive or -negative sentiment (marked by 0 and 1 respectively). - -A small part of the original data as an example can be found in the path below: - -.. literalinclude:: src/sentimental_train.txt - -The corresponding data provider can be found in the path below: - -.. literalinclude:: src/sentimental_provider.py - -This data provider for sequential model is a little more complex than that -for MINST dataset. -A new initialization method is introduced here. -The method :code:`on_init` is configured to DataProvider by :code:`@provider`'s -:code:`init_hook` parameter, and it will be invoked once DataProvider is -initialized. The :code:`on_init` function has the following parameters: - -* The first parameter is the settings object. -* The rest parameters are passed by key word arguments. Some of them are passed - by PaddlePaddle, see reference for `init_hook`_. - The :code:`dictionary` object is a python dict object passed from the trainer - configuration file, and it maps word string to word id. - -To pass these parameters into DataProvider, the following lines should be added -into trainer configuration file. - -.. literalinclude:: src/sentimental_config.py - -The definition is basically same as MNIST example, except: -* Load dictionary in this configuration -* Pass it as a parameter to the DataProvider - -The `input_types` is configured in method :code:`on_init`. It has the same -effect to configure them by :code:`@provider`'s :code:`input_types` parameter. -However, the :code:`input_types` is set at runtime, so we can set it to -different types according to the input data. Input of the neural network is a -sequence of word id, so set :code:`seq_type` to :code:`integer_value_sequence`. - -Durning :code:`on_init`, we save :code:`dictionary` variable to -:code:`settings`, and it will be used in :code:`process`. Note the settings -parameter for the process function and for the on_init's function are a same -object. - -The basic processing logic is the same as MNIST's :code:`process` method. Each -sample in the data file is given back to PaddlePaddle process. - -Thus, the basic usage of PyDataProvider is here. -Please refer to the following section reference for details. - -Reference ---------- - -@provider -+++++++++ - -.. autofunction:: paddle.trainer.PyDataProvider2.provider - -input_types -+++++++++++ - -PaddlePaddle has four data types, and three sequence types. -The four data types are: - -* :code:`dense_vector`: dense float vector. -* :code:`sparse_binary_vector`: sparse binary vector, most of the value is 0, and - the non zero elements are fixed to 1. -* :code:`sparse_float_vector`: sparse float vector, most of the value is 0, and some - non zero elements can be any float value. They are given by the user. -* :code:`integer`: an integer scalar, that is especially used for label or word index. - -The three sequence types are: - -* :code:`SequenceType.NO_SEQUENCE` means the sample is not a sequence. -* :code:`SequenceType.SEQUENCE` means the sample is a sequence. -* :code:`SequenceType.SUB_SEQUENCE` means it is a nested sequence, that each timestep of - the input sequence is also a sequence. - -Different input type has a defferenct input format. Their formats are shown -in the above table. - -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| | NO_SEQUENCE | SEQUENCE | SUB_SEQUENCE | -+======================+=====================+===================================+================================================+ -| dense_vector | [f, f, ...] | [[f, ...], [f, ...], ...] | [[[f, ...], ...], [[f, ...], ...],...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| sparse_binary_vector | [i, i, ...] | [[i, ...], [i, ...], ...] | [[[i, ...], ...], [[i, ...], ...],...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| sparse_float_vector | [(i,f), (i,f), ...] | [[(i,f), ...], [(i,f), ...], ...] | [[[(i,f), ...], ...], [[(i,f), ...], ...],...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ -| integer_value | i | [i, i, ...] | [[i, ...], [i, ...], ...] | -+----------------------+---------------------+-----------------------------------+------------------------------------------------+ - -where f represents a float value, i represents an integer value. - -init_hook -+++++++++ - -init_hook is a function that is invoked once the data provoder is initialized. -Its parameters lists as follows: - -* The first parameter is a settings object, which is the same to :code:`settings` - in :code:`process` method. The object contains several attributes, including: - - * :code:`settings.input_types`: the input types. Reference `input_types`_. - * :code:`settings.logger`: a logging object. - -* The rest parameters are the key word arguments. It is made up of PaddpePaddle - pre-defined parameters and user defined parameters. - - * PaddlePaddle-defined parameters including: - - * :code:`is_train` is a bool parameter that indicates the DataProvider is used in - training or testing. - * :code:`file_list` is the list of all files. - - * User-defined parameters args can be set in training configuration. - -Note, PaddlePaddle reserves the right to add pre-defined parameter, so please -use :code:`**kwargs` in init_hook to ensure compatibility by accepting the -parameters which your init_hook does not use. - -cache -+++++ -DataProvider provides two simple cache strategy. They are: - -* :code:`CacheType.NO_CACHE` means do not cache any data, then data is read at runtime by - the user implemented python module every pass. -* :code:`CacheType.CACHE_PASS_IN_MEM` means the first pass reads data by the user - implemented python module, and the rest passes will directly read data from - memory. diff --git a/doc/api/v1/data_provider/src/mnist_config.py b/doc/api/v1/data_provider/src/mnist_config.py deleted file mode 100644 index 427e0465a68b630ff8a14337e326777f41b6481a..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/mnist_config.py +++ /dev/null @@ -1,23 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from paddle.trainer_config_helpers import * - -define_py_data_sources2( - train_list='train.list', - test_list=None, - module='mnist_provider', - obj='process') - -img = data_layer(name='pixel', size=784) -label = data_layer(name='label', size=10) diff --git a/doc/api/v1/data_provider/src/mnist_provider.dict.py b/doc/api/v1/data_provider/src/mnist_provider.dict.py deleted file mode 100644 index 3fbb783e2f66273ce79c6736a00d01ec58514bc9..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/mnist_provider.dict.py +++ /dev/null @@ -1,37 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from paddle.trainer.PyDataProvider2 import * - - -# Define a py data provider -@provider( - input_types={'pixel': dense_vector(28 * 28), - 'label': integer_value(10)}) -def process(settings, filename): # settings is not used currently. - f = open(filename, 'r') # open one of training file - - for line in f: # read each line - label, pixel = line.split(';') - - # get features and label - pixels_str = pixel.split(' ') - - pixels_float = [] - for each_pixel_str in pixels_str: - pixels_float.append(float(each_pixel_str)) - - # give data to paddle. - yield {"pixel": pixels_float, 'label': int(label)} - - f.close() # close file diff --git a/doc/api/v1/data_provider/src/mnist_train.txt b/doc/api/v1/data_provider/src/mnist_train.txt deleted file mode 100644 index 34be718ad9ea49e7d9aabc34ad70099479df3b31..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/mnist_train.txt +++ /dev/null @@ -1,3 +0,0 @@ -5;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.215686 0.533333 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.67451 0.992157 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.070588 0.886275 0.992157 0 0 0 0 0 0 0 0 0 0 0.192157 0.070588 0 0 0 0 0 0 0 0 0 0 0 0 0 0.670588 0.992157 0.992157 0 0 0 0 0 0 0 0 0 0.117647 0.933333 0.858824 0.313725 0 0 0 0 0 0 0 0 0 0 0 0.090196 0.858824 0.992157 0.831373 0 0 0 0 0 0 0 0 0 0.141176 0.992157 0.992157 0.611765 0.054902 0 0 0 0 0 0 0 0 0 0 0.258824 0.992157 0.992157 0.529412 0 0 0 0 0 0 0 0 0 0.368627 0.992157 0.992157 0.419608 0.003922 0 0 0 0 0 0 0 0 0 0.094118 0.835294 0.992157 0.992157 0.517647 0 0 0 0 0 0 0 0 0 0.603922 0.992157 0.992157 0.992157 0.603922 0.545098 0.043137 0 0 0 0 0 0 0 0.447059 0.992157 0.992157 0.956863 0.062745 0 0 0 0 0 0 0 0 0.011765 0.666667 0.992157 0.992157 0.992157 0.992157 0.992157 0.745098 0.137255 0 0 0 0 0 0.152941 0.866667 0.992157 0.992157 0.521569 0 0 0 0 0 0 0 0 0 0.070588 0.992157 0.992157 0.992157 0.803922 0.352941 0.745098 0.992157 0.945098 0.317647 0 0 0 0 0.580392 0.992157 0.992157 0.764706 0.043137 0 0 0 0 0 0 0 0 0 0.070588 0.992157 0.992157 0.776471 0.043137 0 0.007843 0.27451 0.882353 0.941176 0.176471 0 0 0.180392 0.898039 0.992157 0.992157 0.313725 0 0 0 0 0 0 0 0 0 0 0.070588 0.992157 0.992157 0.713725 0 0 0 0 0.627451 0.992157 0.729412 0.062745 0 0.509804 0.992157 0.992157 0.776471 0.035294 0 0 0 0 0 0 0 0 0 0 0.494118 0.992157 0.992157 0.968627 0.168627 0 0 0 0.423529 0.992157 0.992157 0.364706 0 0.717647 0.992157 0.992157 0.317647 0 0 0 0 0 0 0 0 0 0 0 0.533333 0.992157 0.984314 0.945098 0.603922 0 0 0 0.003922 0.466667 0.992157 0.988235 0.976471 0.992157 0.992157 0.788235 0.007843 0 0 0 0 0 0 0 0 0 0 0 0.686275 0.882353 0.364706 0 0 0 0 0 0 0.098039 0.588235 0.992157 0.992157 0.992157 0.980392 0.305882 0 0 0 0 0 0 0 0 0 0 0 0 0.101961 0.67451 0.321569 0 0 0 0 0 0 0 0.105882 0.733333 0.976471 0.811765 0.713725 0 0 0 0 0 0 0 0 0 0 0 0 0 0.65098 0.992157 0.321569 0 0 0 0 0 0 0 0 0 0.25098 0.007843 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0.94902 0.219608 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.968627 0.764706 0.152941 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.498039 0.25098 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0; -0;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.298039 0.333333 0.333333 0.333333 0.337255 0.333333 0.333333 0.109804 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.027451 0.223529 0.776471 0.964706 0.988235 0.988235 0.988235 0.992157 0.988235 0.988235 0.780392 0.098039 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.14902 0.698039 0.988235 0.992157 0.988235 0.901961 0.87451 0.568627 0.882353 0.976471 0.988235 0.988235 0.501961 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.188235 0.647059 0.988235 0.988235 0.745098 0.439216 0.098039 0 0 0 0.572549 0.988235 0.988235 0.988235 0 0 0 0 0 0 0 0 0 0 0 0 0 0.2 0.933333 0.992157 0.941176 0.247059 0 0 0 0 0 0 0.188235 0.898039 0.992157 0.992157 0 0 0 0 0 0 0 0 0 0 0 0.039216 0.639216 0.933333 0.988235 0.913725 0.278431 0 0 0 0 0 0 0 0.113725 0.843137 0.988235 0.988235 0 0 0 0 0 0 0 0 0 0 0 0.235294 0.988235 0.992157 0.988235 0.815686 0.07451 0 0 0 0 0 0 0 0.333333 0.988235 0.988235 0.552941 0 0 0 0 0 0 0 0 0 0 0.211765 0.878431 0.988235 0.992157 0.701961 0.329412 0.109804 0 0 0 0 0 0 0 0.698039 0.988235 0.913725 0.145098 0 0 0 0 0 0 0 0 0 0.188235 0.890196 0.988235 0.988235 0.745098 0.047059 0 0 0 0 0 0 0 0 0 0.882353 0.988235 0.568627 0 0 0 0 0 0 0 0 0 0.2 0.933333 0.992157 0.992157 0.992157 0.447059 0.294118 0 0 0 0 0 0 0 0 0.447059 0.992157 0.768627 0 0 0 0 0 0 0 0 0 0 0.623529 0.988235 0.988235 0.988235 0.988235 0.992157 0.47451 0 0 0 0 0 0 0 0.188235 0.933333 0.87451 0.509804 0 0 0 0 0 0 0 0 0 0 0.992157 0.988235 0.937255 0.792157 0.988235 0.894118 0.082353 0 0 0 0 0 0 0.027451 0.647059 0.992157 0.654902 0 0 0 0 0 0 0 0 0 0 0 0.623529 0.988235 0.913725 0.329412 0.376471 0.184314 0 0 0 0 0 0 0.027451 0.513725 0.988235 0.635294 0.219608 0 0 0 0 0 0 0 0 0 0 0 0.196078 0.929412 0.988235 0.988235 0.741176 0.309804 0 0 0 0 0 0 0.529412 0.988235 0.678431 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.223529 0.992157 0.992157 1 0.992157 0.992157 0.992157 0.992157 1 0.992157 0.992157 0.882353 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.023529 0.478431 0.654902 0.658824 0.952941 0.988235 0.988235 0.988235 0.992157 0.988235 0.729412 0.278431 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.196078 0.647059 0.764706 0.764706 0.768627 0.580392 0.047059 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0; -4;0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.180392 0.470588 0.623529 0.623529 0.623529 0.588235 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.243137 0.494118 0.862745 0.870588 0.960784 0.996078 0.996078 0.996078 0.996078 0.992157 0.466667 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.317647 0.639216 0.639216 0.639216 0.639216 0.639216 0.470588 0.262745 0.333333 0.929412 0.694118 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.811765 0.694118 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.811765 0.694118 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.811765 0.694118 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.184314 0.992157 0.694118 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.192157 0.996078 0.384314 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.454902 0.980392 0.219608 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.564706 0.941176 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.588235 0.776471 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.945098 0.560784 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.054902 0.952941 0.356863 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.337255 0.917647 0.109804 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.698039 0.701961 0.019608 0.4 0.662745 0.662745 0.662745 0.662745 0.662745 0.662745 0.662745 0.376471 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.090196 0.639216 0.972549 0.945098 0.913725 0.996078 0.996078 0.996078 0.996078 1 0.996078 0.996078 1 0.996078 0 0 0 0 0 0 0 0 0 0 0.007843 0.105882 0.717647 0.776471 0.905882 0.996078 0.996078 0.988235 0.980392 0.862745 0.537255 0.223529 0.223529 0.368627 0.376471 0.6 0.6 0.6 0 0 0 0 0 0 0 0 0.262745 0.470588 0.6 0.996078 0.996078 0.996078 0.996078 0.847059 0.356863 0.156863 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.909804 0.705882 0.823529 0.635294 0.490196 0.219608 0.113725 0.062745 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.152941 0.152941 0.156863 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0; diff --git a/doc/api/v1/data_provider/src/sentimental_config.py b/doc/api/v1/data_provider/src/sentimental_config.py deleted file mode 100644 index edbf3cf1400f5fde5dfa225e5bdbb60400a0691c..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/sentimental_config.py +++ /dev/null @@ -1,27 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from paddle.trainer_config_helpers import * - -dictionary = dict() -... # read dictionary from outside - -define_py_data_sources2( - train_list='train.list', - test_list=None, - module='sentimental_provider', - obj='process', - # above codes same as mnist sample. - args={ # pass to provider. - 'dictionary': dictionary - }) diff --git a/doc/api/v1/data_provider/src/sentimental_provider.py b/doc/api/v1/data_provider/src/sentimental_provider.py deleted file mode 100644 index 03ad1fe7d8c66233b078f94aa303e27cffb8e83c..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/sentimental_provider.py +++ /dev/null @@ -1,56 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from paddle.trainer.PyDataProvider2 import * - - -def on_init(settings, dictionary, **kwargs): - # on_init will invoke when data provider is initialized. The dictionary - # is passed from trainer_config, and is a dict object with type - # (word string => word id). - - # set input types in runtime. It will do the same thing as - # @provider(input_types) will do, but it is set dynamically during runtime. - settings.input_types = { - # The text is a sequence of integer values, and each value is a word id. - # The whole sequence is the sentences that we want to predict its - # sentimental. - 'data': integer_value_sequence(len(dictionary)), # text input - 'label': integer_value(2) # label positive/negative - } - - # save dictionary as settings.dictionary. - # It will be used in process method. - settings.dictionary = dictionary - - -@provider(init_hook=on_init) -def process(settings, filename): - f = open(filename, 'r') - - for line in f: # read each line of file - label, sentence = line.split('\t') # get label and sentence - words = sentence.split(' ') # get words - - # convert word string to word id - # the word not in dictionary will be ignored. - word_ids = [] - - for each_word in words: - if each_word in settings.dictionary: - word_ids.append(settings.dictionary[each_word]) - - # give data to paddle. - yield word_ids, int(label) - - f.close() diff --git a/doc/api/v1/data_provider/src/sentimental_train.txt b/doc/api/v1/data_provider/src/sentimental_train.txt deleted file mode 100644 index 0060ac267c4bf884a8e19553bc4bdea48c89c9b2..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/sentimental_train.txt +++ /dev/null @@ -1,3 +0,0 @@ -0 I saw this movie at the AFI Dallas festival . It all takes place at a lake house and it looks wonderful . -1 This documentary makes you travel all around the globe . It contains rare and stunning sequels from the wilderness . -... diff --git a/doc/api/v1/data_provider/src/train.list b/doc/api/v1/data_provider/src/train.list deleted file mode 100644 index 92bdc0a8b4c21b3091341d93afc5c536c3cffe47..0000000000000000000000000000000000000000 --- a/doc/api/v1/data_provider/src/train.list +++ /dev/null @@ -1 +0,0 @@ -mnist_train.txt diff --git a/doc/api/v1/index_cn.rst b/doc/api/v1/index_cn.rst deleted file mode 100644 index cf146dc088e3905a751ff55c26fd82ef0ba02c89..0000000000000000000000000000000000000000 --- a/doc/api/v1/index_cn.rst +++ /dev/null @@ -1,37 +0,0 @@ -API中文手册 -============ - -DataProvider API ----------------- - -.. toctree:: - :maxdepth: 1 - - data_provider/dataprovider_cn.rst - data_provider/pydataprovider2_cn.rst - -.. _api_trainer_config: - -Model Config API ----------------- - -.. toctree:: - :maxdepth: 1 - - trainer_config_helpers/optimizers.rst - trainer_config_helpers/data_sources.rst - trainer_config_helpers/layers.rst - trainer_config_helpers/activations.rst - trainer_config_helpers/poolings.rst - trainer_config_helpers/networks.rst - trainer_config_helpers/evaluators.rst - trainer_config_helpers/attrs.rst - - -Applications API ----------------- - -.. toctree:: - :maxdepth: 1 - - predict/swig_py_paddle_cn.rst diff --git a/doc/api/v1/index_en.rst b/doc/api/v1/index_en.rst deleted file mode 100644 index 10c297a71d6988c002de868e804ed9ee2345fbd7..0000000000000000000000000000000000000000 --- a/doc/api/v1/index_en.rst +++ /dev/null @@ -1,37 +0,0 @@ -API -=== - -DataProvider API ----------------- - -.. toctree:: - :maxdepth: 1 - - data_provider/dataprovider_en.rst - data_provider/pydataprovider2_en.rst - -.. _api_trainer_config: - -Model Config API ----------------- - -.. toctree:: - :maxdepth: 1 - - trainer_config_helpers/optimizers.rst - trainer_config_helpers/data_sources.rst - trainer_config_helpers/layers.rst - trainer_config_helpers/activations.rst - trainer_config_helpers/poolings.rst - trainer_config_helpers/networks.rst - trainer_config_helpers/evaluators.rst - trainer_config_helpers/attrs.rst - - -Applications API ----------------- - -.. toctree:: - :maxdepth: 1 - - predict/swig_py_paddle_en.rst diff --git a/doc/api/v1/predict/src/predict_sample.py b/doc/api/v1/predict/src/predict_sample.py deleted file mode 100644 index 51349250e80ce11e476d952991f0d046f65286b4..0000000000000000000000000000000000000000 --- a/doc/api/v1/predict/src/predict_sample.py +++ /dev/null @@ -1,135 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from py_paddle import swig_paddle, DataProviderConverter -from paddle.trainer.PyDataProvider2 import dense_vector -from paddle.trainer.config_parser import parse_config - -TEST_DATA = [[[ - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.215686, 0.533333, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.67451, 0.992157, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0.070588, 0.886275, 0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.192157, - 0.070588, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.670588, 0.992157, - 0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.117647, 0.933333, 0.858824, 0.313725, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.090196, 0.858824, 0.992157, 0.831373, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0.141176, 0.992157, 0.992157, 0.611765, 0.054902, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.258824, 0.992157, 0.992157, 0.529412, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0.368627, 0.992157, 0.992157, 0.419608, 0.003922, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0.094118, 0.835294, 0.992157, 0.992157, 0.517647, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0.603922, 0.992157, 0.992157, 0.992157, 0.603922, - 0.545098, 0.043137, 0, 0, 0, 0, 0, 0, 0, 0.447059, 0.992157, 0.992157, - 0.956863, 0.062745, 0, 0, 0, 0, 0, 0, 0, 0, 0.011765, 0.666667, 0.992157, - 0.992157, 0.992157, 0.992157, 0.992157, 0.745098, 0.137255, 0, 0, 0, 0, 0, - 0.152941, 0.866667, 0.992157, 0.992157, 0.521569, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0.070588, 0.992157, 0.992157, 0.992157, 0.803922, 0.352941, 0.745098, - 0.992157, 0.945098, 0.317647, 0, 0, 0, 0, 0.580392, 0.992157, 0.992157, - 0.764706, 0.043137, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.070588, 0.992157, 0.992157, - 0.776471, 0.043137, 0, 0.007843, 0.27451, 0.882353, 0.941176, 0.176471, 0, - 0, 0.180392, 0.898039, 0.992157, 0.992157, 0.313725, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0.070588, 0.992157, 0.992157, 0.713725, 0, 0, 0, 0, 0.627451, - 0.992157, 0.729412, 0.062745, 0, 0.509804, 0.992157, 0.992157, 0.776471, - 0.035294, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.494118, 0.992157, 0.992157, - 0.968627, 0.168627, 0, 0, 0, 0.423529, 0.992157, 0.992157, 0.364706, 0, - 0.717647, 0.992157, 0.992157, 0.317647, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0.533333, 0.992157, 0.984314, 0.945098, 0.603922, 0, 0, 0, 0.003922, - 0.466667, 0.992157, 0.988235, 0.976471, 0.992157, 0.992157, 0.788235, - 0.007843, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.686275, 0.882353, 0.364706, 0, - 0, 0, 0, 0, 0, 0.098039, 0.588235, 0.992157, 0.992157, 0.992157, 0.980392, - 0.305882, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.101961, 0.67451, 0.321569, - 0, 0, 0, 0, 0, 0, 0, 0.105882, 0.733333, 0.976471, 0.811765, 0.713725, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.65098, 0.992157, 0.321569, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0.25098, 0.007843, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, - 0.94902, 0.219608, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0.968627, 0.764706, 0.152941, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.498039, 0.25098, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0 -]], [[ - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0.298039, 0.333333, 0.333333, 0.333333, 0.337255, - 0.333333, 0.333333, 0.109804, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0.027451, 0.223529, 0.776471, 0.964706, 0.988235, 0.988235, 0.988235, - 0.992157, 0.988235, 0.988235, 0.780392, 0.098039, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0.14902, 0.698039, 0.988235, 0.992157, 0.988235, 0.901961, - 0.87451, 0.568627, 0.882353, 0.976471, 0.988235, 0.988235, 0.501961, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.188235, 0.647059, 0.988235, 0.988235, - 0.745098, 0.439216, 0.098039, 0, 0, 0, 0.572549, 0.988235, 0.988235, - 0.988235, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.2, 0.933333, 0.992157, - 0.941176, 0.247059, 0, 0, 0, 0, 0, 0, 0.188235, 0.898039, 0.992157, - 0.992157, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.039216, 0.639216, 0.933333, - 0.988235, 0.913725, 0.278431, 0, 0, 0, 0, 0, 0, 0, 0.113725, 0.843137, - 0.988235, 0.988235, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.235294, 0.988235, - 0.992157, 0.988235, 0.815686, 0.07451, 0, 0, 0, 0, 0, 0, 0, 0.333333, - 0.988235, 0.988235, 0.552941, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.211765, - 0.878431, 0.988235, 0.992157, 0.701961, 0.329412, 0.109804, 0, 0, 0, 0, 0, - 0, 0, 0.698039, 0.988235, 0.913725, 0.145098, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0.188235, 0.890196, 0.988235, 0.988235, 0.745098, 0.047059, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0.882353, 0.988235, 0.568627, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.2, - 0.933333, 0.992157, 0.992157, 0.992157, 0.447059, 0.294118, 0, 0, 0, 0, 0, - 0, 0, 0, 0.447059, 0.992157, 0.768627, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0.623529, 0.988235, 0.988235, 0.988235, 0.988235, 0.992157, 0.47451, 0, 0, - 0, 0, 0, 0, 0, 0.188235, 0.933333, 0.87451, 0.509804, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0.992157, 0.988235, 0.937255, 0.792157, 0.988235, 0.894118, - 0.082353, 0, 0, 0, 0, 0, 0, 0.027451, 0.647059, 0.992157, 0.654902, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0.623529, 0.988235, 0.913725, 0.329412, 0.376471, - 0.184314, 0, 0, 0, 0, 0, 0, 0.027451, 0.513725, 0.988235, 0.635294, - 0.219608, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.196078, 0.929412, 0.988235, - 0.988235, 0.741176, 0.309804, 0, 0, 0, 0, 0, 0, 0.529412, 0.988235, - 0.678431, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.223529, 0.992157, - 0.992157, 1, 0.992157, 0.992157, 0.992157, 0.992157, 1, 0.992157, 0.992157, - 0.882353, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.023529, - 0.478431, 0.654902, 0.658824, 0.952941, 0.988235, 0.988235, 0.988235, - 0.992157, 0.988235, 0.729412, 0.278431, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0.196078, 0.647059, 0.764706, 0.764706, 0.768627, - 0.580392, 0.047059, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0 -]]] - - -def main(): - conf = parse_config("./mnist_model/trainer_config.py", "") - print conf.data_config.load_data_args - network = swig_paddle.GradientMachine.createFromConfigProto( - conf.model_config) - assert isinstance(network, swig_paddle.GradientMachine) # For code hint. - network.loadParameters("./mnist_model/") - converter = DataProviderConverter([dense_vector(784)]) - inArg = converter(TEST_DATA) - print network.forwardTest(inArg) - - -if __name__ == '__main__': - swig_paddle.initPaddle("--use_gpu=0") - main() diff --git a/doc/api/v1/predict/swig_py_paddle_cn.rst b/doc/api/v1/predict/swig_py_paddle_cn.rst deleted file mode 100644 index 42f333dba2e996e70572b3cda085b83e402ede8e..0000000000000000000000000000000000000000 --- a/doc/api/v1/predict/swig_py_paddle_cn.rst +++ /dev/null @@ -1,58 +0,0 @@ -.. _api_swig_py_paddle: - -基于Python的预测 -================ - -预测流程 --------- - -PaddlePaddle使用swig对常用的预测接口进行了封装，通过编译会生成py_paddle软件包，安装该软件包就可以在python环境下实现模型预测。可以使用python的 ``help()`` 函数查询软件包相关API说明。 - -基于Python的模型预测，主要包括以下五个步骤。 - -1. 初始化PaddlePaddle环境 - - 在程序开始阶段，通过调用 ``swig_paddle.initPaddle()`` 并传入相应的命令行参数初始化PaddlePaddle。 - -2. 解析模型配置文件 - - 初始化之后，可以通过调用 ``parse_config()`` 解析训练模型时用的配置文件。注意预测数据通常不包含label, 同时预测网络通常直接输出最后一层的结果而不是像训练网络一样再接一层cost layer，所以一般需要对训练用的模型配置文件稍作相应修改才能在预测时使用。 - -3. 构造paddle.GradientMachine - - 通过调用 ``swig_paddle.GradientMachine.createFromConfigproto()`` 传入上一步解析出来的模型配置就可以创建一个 ``GradientMachine``。 - -4. 准备预测数据 - - swig_paddle中的预测接口的参数是自定义的C++数据类型，py_paddle里面提供了一个工具类 ``DataProviderConverter`` 可以用于接收和PyDataProvider2一样的输入数据并转换成预测接口所需的数据类型。 - -5. 模型预测 - - 通过调用 ``forwardTest()`` 传入预测数据，直接返回计算结果。 - - -预测Demo --------- - -如下是一段使用mnist model来实现手写识别的预测代码。完整的代码见 ``src_root/doc/ui/predict/predict_sample.py`` 。mnist model可以通过 ``src_root\demo\mnist`` 目录下的demo训练出来。 - -.. literalinclude:: src/predict_sample.py - :language: python - :lines: 15-18,121-136 - - -Demo预测输出如下，其中value即为softmax层的输出。由于TEST_DATA包含两条预测数据，所以输出的value包含两个向量 。 - -.. code-block:: text - - [{'id': None, 'value': array( - [[ 5.53018653e-09, 1.12194102e-05, 1.96644767e-09, - 1.43630644e-02, 1.51111044e-13, 9.85625684e-01, - 2.08823112e-10, 2.32777140e-08, 2.00186201e-09, - 1.15501715e-08], - [ 9.99982715e-01, 1.27787406e-10, 1.72296313e-05, - 1.49316648e-09, 1.36540484e-11, 6.93137714e-10, - 2.70634608e-08, 3.48565123e-08, 5.25639710e-09, - 4.48684503e-08]], dtype=float32)}] - - diff --git a/doc/api/v1/predict/swig_py_paddle_en.rst b/doc/api/v1/predict/swig_py_paddle_en.rst deleted file mode 100644 index 1c628e6971fa5643e6a9ca629488049957686193..0000000000000000000000000000000000000000 --- a/doc/api/v1/predict/swig_py_paddle_en.rst +++ /dev/null @@ -1,59 +0,0 @@ -Python Prediction -================== - -PaddlePaddle offers a set of clean prediction interfaces for python with the help of -SWIG. The main steps of predict values in python are: - -* Parse training configurations -* Construct GradientMachine -* Prepare data -* Predict - -Here is a sample python script that shows the typical prediction process for the -MNIST classification problem. A complete sample code could be found at -:code:`src_root/doc/ui/predict/predict_sample.py`. - -.. literalinclude:: src/predict_sample.py - :language: python - :lines: 15-18,90-100,101-104 - -The module that does the most of the job is py_paddle.swig_paddle, it's -generated by SWIG and has complete documents, for more details you can use -python's :code:`help()` function. Let's walk through the above python script: - -* At the beginning, use :code:`swig_paddle.initPaddle()` to initialize - PaddlePaddle with command line arguments, for more about command line arguments - see :ref:`cmd_detail_introduction` . -* Parse the configuration file that is used in training with :code:`parse_config()`. - Because data to predict with always have no label, and output of prediction work - normally is the output layer rather than the cost layer, so you should modify - the configuration file accordingly before using it in the prediction work. -* Create a neural network with - :code:`swig_paddle.GradientMachine.createFromConfigproto()`, which takes the - parsed configuration :code:`conf.model_config` as argument. Then load the - trained parameters from the model with :code:`network.loadParameters()`. -* Create a data converter object of utility class :code:`DataProviderConverter`. - - Note: As swig_paddle can only accept C++ matrices, we offer a utility - class DataProviderConverter that can accept the same input data with - PyDataProvider2, for more information please refer to document - of :ref:`api_pydataprovider2` . -* Do the prediction with :code:`forwardTest()`, which takes the converted - input data and outputs the activations of the output layer. - -Here is a typical output: - -.. code-block:: text - - [{'id': None, 'value': array([[ 5.53018653e-09, 1.12194102e-05, 1.96644767e-09, - 1.43630644e-02, 1.51111044e-13, 9.85625684e-01, - 2.08823112e-10, 2.32777140e-08, 2.00186201e-09, - 1.15501715e-08], - [ 9.99982715e-01, 1.27787406e-10, 1.72296313e-05, - 1.49316648e-09, 1.36540484e-11, 6.93137714e-10, - 2.70634608e-08, 3.48565123e-08, 5.25639710e-09, - 4.48684503e-08]], dtype=float32)}] - -:code:`value` is the output of the output layer, each row represents result of -the corresponding row in the input data, each element represents activation of -the corresponding neuron in the output layer. - diff --git a/doc/api/v2/fluid/data_feeder.rst b/doc/api/v2/fluid/data_feeder.rst index 0fa78f7dfb04c13be7eb83b7fd35cb03f2f4a7fa..a591c7334fd31c98a94b50a4344f251560a0f2f9 100644 --- a/doc/api/v2/fluid/data_feeder.rst +++ b/doc/api/v2/fluid/data_feeder.rst @@ -1,9 +1,14 @@ +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + =========== -DataFeeder +data_feeder =========== DataFeeder ------------ -.. automodule:: paddle.v2.fluid.data_feeder - :members: DataFeeder +---------- + +.. autoclass:: paddle.v2.fluid.data_feeder.DataFeeder + :members: :noindex: + diff --git a/doc/api/v2/fluid/evaluator.rst b/doc/api/v2/fluid/evaluator.rst index a23f3301d0331e0ea3733f06444515eb4680cd31..00dcecfd628a35d83d1c596bf0aea819a1705862 100644 --- a/doc/api/v2/fluid/evaluator.rst +++ b/doc/api/v2/fluid/evaluator.rst @@ -1,9 +1,21 @@ -=========== -Evaluator -=========== - -Evaluator ------------ -.. automodule:: paddle.v2.fluid.evaluator - :members: Evaluator +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + +========= +evaluator +========= + +Accuracy +-------- + +.. autoclass:: paddle.v2.fluid.evaluator.Accuracy + :members: :noindex: + +ChunkEvaluator +-------------- + +.. autoclass:: paddle.v2.fluid.evaluator.ChunkEvaluator + :members: + :noindex: + diff --git a/doc/api/v2/fluid/executor.rst b/doc/api/v2/fluid/executor.rst index 3a283538c120cfa1ef646c390bb71c6251c23675..a028f6283f2ca333bdf6c9857a98661c0222b41e 100644 --- a/doc/api/v2/fluid/executor.rst +++ b/doc/api/v2/fluid/executor.rst @@ -1,9 +1,32 @@ -=========== -Executor -=========== +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + +======== +executor +======== Executor +-------- + +.. autoclass:: paddle.v2.fluid.executor.Executor + :members: + :noindex: + +global_scope +------------ + +.. autofunction:: paddle.v2.fluid.executor.global_scope + :noindex: + +scope_guard ----------- -.. automodule:: paddle.v2.fluid.executor - :members: Executor + +.. autofunction:: paddle.v2.fluid.executor.scope_guard + :noindex: + +switch_scope +------------ + +.. autofunction:: paddle.v2.fluid.executor.switch_scope :noindex: + diff --git a/doc/api/v2/fluid/gen_doc.py b/doc/api/v2/fluid/gen_doc.py new file mode 100644 index 0000000000000000000000000000000000000000..a2147fd3f7ea635d8f14210fbcd1a568ee2230ee --- /dev/null +++ b/doc/api/v2/fluid/gen_doc.py @@ -0,0 +1,109 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from __future__ import print_function +import argparse +import sys +import types + +import paddle.v2.fluid as fluid + + +def parse_arg(): + parser = argparse.ArgumentParser() + parser.add_argument('--submodules', nargs="*") + parser.add_argument( + 'module', type=str, help='Generate the documentation of which module') + return parser.parse_args() + + +class DocGenerator(object): + def __init__(self, module_name, stream=sys.stdout): + self.stream = stream + self.module_name = module_name + if not hasattr(fluid, module_name): + raise ValueError("Cannot find fluid.{0}".format(module_name)) + else: + self.module = getattr(fluid, module_name) + self.stream.write('''.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + +''') + + self._print_header_(module_name, dot='=', is_title=True) + + def print_submodule(self, submodule_name): + submodule = getattr(self.module, submodule_name) + if submodule is None: + raise ValueError("Cannot find submodule {0}".format(submodule_name)) + self.print_section(submodule_name) + + for item in submodule.__all__: + self.print_item(item) + + def print_current_module(self): + for item in self.module.__all__: + self.print_item(item) + + def print_section(self, name): + self._print_header_(name, dot='=', is_title=False) + + def print_item(self, name): + item = getattr(self.module, name) + if isinstance(item, types.TypeType): + self.print_class(name) + elif isinstance(item, types.FunctionType): + self.print_method(name) + else: + raise RuntimeError("Unsupported item {0}".format(name)) + + def print_class(self, name): + self._print_header_(name, dot='-', is_title=False) + self.stream.write('''.. autoclass:: paddle.v2.fluid.{0}.{1} + :members: + :noindex: + +'''.format(self.module_name, name)) + + def print_method(self, name): + self._print_header_(name, dot='-', is_title=False) + self.stream.write('''.. autofunction:: paddle.v2.fluid.{0}.{1} + :noindex: + +'''.format(self.module_name, name)) + + def _print_header_(self, name, dot, is_title): + dot_line = dot * len(name) + if is_title: + self.stream.write(dot_line) + self.stream.write('\n') + self.stream.write(name) + self.stream.write('\n') + self.stream.write(dot_line) + self.stream.write('\n') + self.stream.write('\n') + + +def main(): + args = parse_arg() + gen = DocGenerator(args.module) + if args.submodules is None: + gen.print_current_module() + else: + for submodule_name in args.submodules: + gen.print_submodule(submodule_name) + + +if __name__ == '__main__': + main() diff --git a/doc/api/v2/fluid/gen_doc.sh b/doc/api/v2/fluid/gen_doc.sh new file mode 100755 index 0000000000000000000000000000000000000000..ba7b7ba8e51399deb852b0a7c8ddd3128f521e85 --- /dev/null +++ b/doc/api/v2/fluid/gen_doc.sh @@ -0,0 +1,7 @@ +#!/bin/bash +python gen_doc.py layers --submodules control_flow device io nn ops tensor > layers.rst + +for module in io data_feeder evaluator executor initializer io nets optimizer param_attr profiler regularizer +do + python gen_doc.py ${module} > ${module}.rst +done diff --git a/doc/api/v2/fluid/initializer.rst b/doc/api/v2/fluid/initializer.rst index 8f587837e9873370722062404f511654a9460587..c38be033fff2997930525f51c93995db09daa2b6 100644 --- a/doc/api/v2/fluid/initializer.rst +++ b/doc/api/v2/fluid/initializer.rst @@ -1,50 +1,35 @@ +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + =========== -Initializer +initializer =========== +Constant +-------- - -Initializer ------------ -.. automodule:: paddle.v2.fluid.initializer - :members: Initializer - :noindex: - - - -ConstantInitializer -------------------- -.. automodule:: paddle.v2.fluid.initializer - :members: ConstantInitializer +.. autoclass:: paddle.v2.fluid.initializer.Constant + :members: :noindex: +Uniform +------- - -UniformInitializer ------------------- -.. automodule:: paddle.v2.fluid.initializer - :members: UniformInitializer - :noindex: - - - -NormalInitializer ------------------ -.. automodule:: paddle.v2.fluid.initializer - :members: NormalInitializer +.. autoclass:: paddle.v2.fluid.initializer.Uniform + :members: :noindex: +Normal +------ -XavierInitializer ------------------ -.. automodule:: paddle.v2.fluid.initializer - :members: XavierInitializer +.. autoclass:: paddle.v2.fluid.initializer.Normal + :members: :noindex: +Xavier +------ -MSRAInitializer ---------------- -.. automodule:: paddle.v2.fluid.initializer - :members: MSRAInitializer +.. autoclass:: paddle.v2.fluid.initializer.Xavier + :members: :noindex: diff --git a/doc/api/v2/fluid/io.rst b/doc/api/v2/fluid/io.rst index 67f68c4e9e16b379207b8de114cdf769e056f78e..37c9c273e369532e8ff596e9649cb695a98a2505 100644 --- a/doc/api/v2/fluid/io.rst +++ b/doc/api/v2/fluid/io.rst @@ -1,10 +1,61 @@ -=========== -IO -=========== +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! +== +io +== +save_vars +--------- -is_parameter +.. autofunction:: paddle.v2.fluid.io.save_vars + :noindex: + +save_params ----------- -.. autofunction:: paddle.v2.fluid.io.is_parameter + +.. autofunction:: paddle.v2.fluid.io.save_params + :noindex: + +save_persistables +----------------- + +.. autofunction:: paddle.v2.fluid.io.save_persistables + :noindex: + +load_vars +--------- + +.. autofunction:: paddle.v2.fluid.io.load_vars + :noindex: + +load_params +----------- + +.. autofunction:: paddle.v2.fluid.io.load_params :noindex: + +load_persistables +----------------- + +.. autofunction:: paddle.v2.fluid.io.load_persistables + :noindex: + +save_inference_model +-------------------- + +.. autofunction:: paddle.v2.fluid.io.save_inference_model + :noindex: + +load_inference_model +-------------------- + +.. autofunction:: paddle.v2.fluid.io.load_inference_model + :noindex: + +get_inference_program +--------------------- + +.. autofunction:: paddle.v2.fluid.io.get_inference_program + :noindex: + diff --git a/doc/api/v2/fluid/layers.rst b/doc/api/v2/fluid/layers.rst index 62c154e65dcff1bdfb00109bf1b724c34731652e..e24613b94b422b7cdf9c6383c359fa92a4faf6ff 100644 --- a/doc/api/v2/fluid/layers.rst +++ b/doc/api/v2/fluid/layers.rst @@ -1,495 +1,799 @@ -========== -Layers -========== +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! +====== +layers +====== -fc ---- -.. autofunction:: paddle.v2.fluid.layers.fc - :noindex: +control_flow +============ -embedding ---------- -.. autofunction:: paddle.v2.fluid.layers.embedding +split_lod_tensor +---------------- + +.. autofunction:: paddle.v2.fluid.layers.split_lod_tensor :noindex: -dynamic_lstm ------------- -.. autofunction:: paddle.v2.fluid.layers.dynamic_lstm +merge_lod_tensor +---------------- + +.. autofunction:: paddle.v2.fluid.layers.merge_lod_tensor :noindex: -data ----- -.. autofunction:: paddle.v2.fluid.layers.data +BlockGuard +---------- + +.. autoclass:: paddle.v2.fluid.layers.BlockGuard + :members: :noindex: -mean ----- -.. autofunction:: paddle.v2.fluid.layers.mean +BlockGuardWithCompletion +------------------------ + +.. autoclass:: paddle.v2.fluid.layers.BlockGuardWithCompletion + :members: :noindex: -mul ---- -.. autofunction:: paddle.v2.fluid.layers.mul +StaticRNNMemoryLink +------------------- + +.. autoclass:: paddle.v2.fluid.layers.StaticRNNMemoryLink + :members: :noindex: -elementwise_add ---------------- -.. autofunction:: paddle.v2.fluid.layers.elementwise_add +WhileGuard +---------- + +.. autoclass:: paddle.v2.fluid.layers.WhileGuard + :members: :noindex: -elementwise_sub ---------------- -.. autofunction:: paddle.v2.fluid.layers.elementwise_sub +While +----- + +.. autoclass:: paddle.v2.fluid.layers.While + :members: :noindex: -elementwise_mul ---------------- -.. autofunction:: paddle.v2.fluid.layers.elementwise_mul +lod_rank_table +-------------- + +.. autofunction:: paddle.v2.fluid.layers.lod_rank_table :noindex: -elementwise_div ---------------- -.. autofunction:: paddle.v2.fluid.layers.elementwise_div +max_sequence_len +---------------- + +.. autofunction:: paddle.v2.fluid.layers.max_sequence_len :noindex: +topk +---- -dropout -------- -.. autofunction:: paddle.v2.fluid.layers.dropout +.. autofunction:: paddle.v2.fluid.layers.topk :noindex: +lod_tensor_to_array +------------------- -reshape --------- -.. autofunction:: paddle.v2.fluid.layers.reshape +.. autofunction:: paddle.v2.fluid.layers.lod_tensor_to_array :noindex: +array_to_lod_tensor +------------------- -sigmoid +.. autofunction:: paddle.v2.fluid.layers.array_to_lod_tensor + :noindex: + +increment --------- -.. autofunction:: paddle.v2.fluid.layers.sigmoid + +.. autofunction:: paddle.v2.fluid.layers.increment :noindex: +array_write +----------- -scale +.. autofunction:: paddle.v2.fluid.layers.array_write + :noindex: + +create_array +------------ + +.. autofunction:: paddle.v2.fluid.layers.create_array + :noindex: + +less_than --------- -.. autofunction:: paddle.v2.fluid.layers.scale + +.. autofunction:: paddle.v2.fluid.layers.less_than :noindex: +array_read +---------- -transpose +.. autofunction:: paddle.v2.fluid.layers.array_read + :noindex: + +shrink_memory +------------- + +.. autofunction:: paddle.v2.fluid.layers.shrink_memory + :noindex: + +array_length +------------ + +.. autofunction:: paddle.v2.fluid.layers.array_length + :noindex: + +IfElse +------ + +.. autoclass:: paddle.v2.fluid.layers.IfElse + :members: + :noindex: + +DynamicRNN +---------- + +.. autoclass:: paddle.v2.fluid.layers.DynamicRNN + :members: + :noindex: + +ConditionalBlock +---------------- + +.. autoclass:: paddle.v2.fluid.layers.ConditionalBlock + :members: + :noindex: + +StaticRNN --------- -.. autofunction:: paddle.v2.fluid.layers.transpose + +.. autoclass:: paddle.v2.fluid.layers.StaticRNN + :members: :noindex: +reorder_lod_tensor_by_rank +-------------------------- -sigmoid_cross_entropy_with_logits ---------------------------------- -.. autofunction:: paddle.v2.fluid.layers.esigmoid_cross_entropy_with_logits +.. autofunction:: paddle.v2.fluid.layers.reorder_lod_tensor_by_rank :noindex: +ParallelDo +---------- -cast +.. autoclass:: paddle.v2.fluid.layers.ParallelDo + :members: + :noindex: + +Print +----- + +.. autofunction:: paddle.v2.fluid.layers.Print + :noindex: + +device +====== + +get_places +---------- + +.. autofunction:: paddle.v2.fluid.layers.get_places + :noindex: + +io +== + +data ---- -.. autofunction:: paddle.v2.fluid.layers.cast + +.. autofunction:: paddle.v2.fluid.layers.data :noindex: +BlockGuardServ +-------------- -concat -------- -.. autofunction:: paddle.v2.fluid.layers.concat +.. autoclass:: paddle.v2.fluid.layers.BlockGuardServ + :members: :noindex: +ListenAndServ +------------- -sums +.. autoclass:: paddle.v2.fluid.layers.ListenAndServ + :members: + :noindex: + +Send ---- -.. autofunction:: paddle.v2.fluid.layers.sums + +.. autofunction:: paddle.v2.fluid.layers.Send :noindex: +nn +== -linear_chain_crf ----------------- -.. autofunction:: paddle.v2.fluid.layers.linear_chain_crf +fc +-- + +.. autofunction:: paddle.v2.fluid.layers.fc :noindex: +embedding +--------- -assign -------- .. autofunction:: paddle.v2.fluid.layers.embedding :noindex: +dynamic_lstm +------------ -split_lod_tensor ----------------- -.. autofunction:: paddle.v2.fluid.layers.split_lod_tensor +.. autofunction:: paddle.v2.fluid.layers.dynamic_lstm :noindex: +dynamic_lstmp +------------- -merge_lod_tensor +.. autofunction:: paddle.v2.fluid.layers.dynamic_lstmp + :noindex: + +dynamic_gru +----------- + +.. autofunction:: paddle.v2.fluid.layers.dynamic_gru + :noindex: + +gru_unit +-------- + +.. autofunction:: paddle.v2.fluid.layers.gru_unit + :noindex: + +linear_chain_crf ---------------- -.. autofunction:: paddle.v2.fluid.layers.merge_lod_tensor + +.. autofunction:: paddle.v2.fluid.layers.linear_chain_crf + :noindex: + +crf_decoding +------------ + +.. autofunction:: paddle.v2.fluid.layers.crf_decoding :noindex: cos_sim --------- +------- + .. autofunction:: paddle.v2.fluid.layers.cos_sim :noindex: - cross_entropy ------------- + .. autofunction:: paddle.v2.fluid.layers.cross_entropy :noindex: - - square_error_cost ----------------- + .. autofunction:: paddle.v2.fluid.layers.square_error_cost :noindex: - accuracy ---------- +-------- + .. autofunction:: paddle.v2.fluid.layers.accuracy :noindex: +chunk_eval +---------- + +.. autofunction:: paddle.v2.fluid.layers.chunk_eval + :noindex: sequence_conv ------------- + .. autofunction:: paddle.v2.fluid.layers.sequence_conv :noindex: - conv2d ------ + .. autofunction:: paddle.v2.fluid.layers.conv2d :noindex: - sequence_pool ------------- + .. autofunction:: paddle.v2.fluid.layers.sequence_pool :noindex: - +pool2d +------ + +.. autofunction:: paddle.v2.fluid.layers.pool2d + :noindex: + +batch_norm +---------- + +.. autofunction:: paddle.v2.fluid.layers.batch_norm + :noindex: + +beam_search_decode +------------------ + +.. autofunction:: paddle.v2.fluid.layers.beam_search_decode + :noindex: + +conv2d_transpose +---------------- + +.. autofunction:: paddle.v2.fluid.layers.conv2d_transpose + :noindex: + +sequence_expand +--------------- + +.. autofunction:: paddle.v2.fluid.layers.sequence_expand + :noindex: + +lstm_unit +--------- + +.. autofunction:: paddle.v2.fluid.layers.lstm_unit + :noindex: + +reduce_sum +---------- + +.. autofunction:: paddle.v2.fluid.layers.reduce_sum + :noindex: + +reduce_mean +----------- + +.. autofunction:: paddle.v2.fluid.layers.reduce_mean + :noindex: + +reduce_max +---------- + +.. autofunction:: paddle.v2.fluid.layers.reduce_max + :noindex: + +reduce_min +---------- + +.. autofunction:: paddle.v2.fluid.layers.reduce_min + :noindex: + sequence_first_step ------------------- + .. autofunction:: paddle.v2.fluid.layers.sequence_first_step :noindex: - sequence_last_step ------------------ + .. autofunction:: paddle.v2.fluid.layers.sequence_last_step :noindex: +dropout +------- -pool2d ------- -.. autofunction:: paddle.v2.fluid.layers.pool2d +.. autofunction:: paddle.v2.fluid.layers.dropout :noindex: +split +----- -batch_norm ----------- -.. autofunction:: paddle.v2.fluid.layers.batch_norm +.. autofunction:: paddle.v2.fluid.layers.split :noindex: - -beam_search_decode +ctc_greedy_decoder ------------------ -.. autofunction:: paddle.v2.fluid.layers.beam_search_decode - :noindex: - -lod_rank_table --------------- -.. autofunction:: paddle.v2.fluid.layers.lod_rank_table +.. autofunction:: paddle.v2.fluid.layers.ctc_greedy_decoder :noindex: +edit_distance +------------- -max_sequence_len ----------------- -.. autofunction:: paddle.v2.fluid.layers.max_sequence_len +.. autofunction:: paddle.v2.fluid.layers.edit_distance :noindex: +l2_normalize +------------ -topk ------ -.. autofunction:: paddle.v2.fluid.layers.topk +.. autofunction:: paddle.v2.fluid.layers.l2_normalize :noindex: +matmul +------ -lod_tensor_to_array -------------------- -.. autofunction:: paddle.v2.fluid.layers.lod_tensor_to_array +.. autofunction:: paddle.v2.fluid.layers.matmul :noindex: +warpctc +------- - -array_to_lod_tensor -------------------- -.. autofunction:: paddle.v2.fluid.layers.array_to_lod_tensor +.. autofunction:: paddle.v2.fluid.layers.warpctc :noindex: +sequence_reshape +---------------- +.. autofunction:: paddle.v2.fluid.layers.sequence_reshape + :noindex: +transpose +--------- -fill_constant -------------- -.. autofunction:: paddle.v2.fluid.layers.fill_constant +.. autofunction:: paddle.v2.fluid.layers.transpose :noindex: +im2sequence +----------- - -fill_constant_batch_size_like ------------------------------ -.. autofunction:: paddle.v2.fluid.layers.fill_constant_batch_size_like +.. autofunction:: paddle.v2.fluid.layers.im2sequence :noindex: +nce +--- -ones ----- -.. autofunction:: paddle.v2.fluid.layers.ones +.. autofunction:: paddle.v2.fluid.layers.nce :noindex: +beam_search +----------- -zeros ------ -.. autofunction:: paddle.v2.fluid.layers.zeros +.. autofunction:: paddle.v2.fluid.layers.beam_search :noindex: +row_conv +-------- -increment ---------- -.. autofunction:: paddle.v2.fluid.layers.increment +.. autofunction:: paddle.v2.fluid.layers.row_conv :noindex: +multiplex +--------- -array_write ------------ -.. autofunction:: paddle.v2.fluid.layers.array_write +.. autofunction:: paddle.v2.fluid.layers.multiplex :noindex: +ops +=== +mean +---- -create_array ------------- -.. autofunction:: paddle.v2.fluid.layers.create_array +.. autofunction:: paddle.v2.fluid.layers.mean :noindex: +mul +--- -less_than ---------- -.. autofunction:: paddle.v2.fluid.layers.less_than +.. autofunction:: paddle.v2.fluid.layers.mul :noindex: +reshape +------- -array_read ----------- -.. autofunction:: paddle.v2.fluid.layers.array_read +.. autofunction:: paddle.v2.fluid.layers.reshape :noindex: +scale +----- -shrink_memory --------------- -.. autofunction:: paddle.v2.fluid.layers.shrink_memory +.. autofunction:: paddle.v2.fluid.layers.scale :noindex: +sigmoid_cross_entropy_with_logits +--------------------------------- -array_length -------------- -.. autofunction:: paddle.v2.fluid.layers.array_length +.. autofunction:: paddle.v2.fluid.layers.sigmoid_cross_entropy_with_logits :noindex: +elementwise_add +--------------- -conv2d_transpose ----------------- -.. autofunction:: paddle.v2.fluid.layers.conv2d_transpose +.. autofunction:: paddle.v2.fluid.layers.elementwise_add :noindex: +elementwise_div +--------------- -sequence_expand +.. autofunction:: paddle.v2.fluid.layers.elementwise_div + :noindex: + +elementwise_sub --------------- -.. autofunction:: paddle.v2.fluid.layers.sequence_expand + +.. autofunction:: paddle.v2.fluid.layers.elementwise_sub :noindex: +elementwise_mul +--------------- -gru_unit --------- -.. autofunction:: paddle.v2.fluid.layers.gru_unit +.. autofunction:: paddle.v2.fluid.layers.elementwise_mul :noindex: +elementwise_max +--------------- -lstm_unit ---------- -.. autofunction:: paddle.v2.fluid.layers.lstm_unit +.. autofunction:: paddle.v2.fluid.layers.elementwise_max :noindex: +elementwise_min +--------------- -sequence_softmax ----------------- -.. autofunction:: paddle.v2.fluid.layers.sequence_softmax +.. autofunction:: paddle.v2.fluid.layers.elementwise_min :noindex: +elementwise_pow +--------------- -reduce_sum ----------- -.. autofunction:: paddle.v2.fluid.layers.reduce_sum +.. autofunction:: paddle.v2.fluid.layers.elementwise_pow :noindex: +clip +---- -reduce_mean ------------ -.. autofunction:: paddle.v2.fluid.layers.reduce_mean +.. autofunction:: paddle.v2.fluid.layers.clip :noindex: +clip_by_norm +------------ -reduce_max ----------- -.. autofunction:: paddle.v2.fluid.layers.reduce_max +.. autofunction:: paddle.v2.fluid.layers.clip_by_norm :noindex: +sequence_softmax +---------------- -reduce_min ----------- -.. autofunction:: paddle.v2.fluid.layers.reduce_min +.. autofunction:: paddle.v2.fluid.layers.sequence_softmax :noindex: +sigmoid +------- -split ------ -.. autofunction:: paddle.v2.fluid.layers.split +.. autofunction:: paddle.v2.fluid.layers.sigmoid :noindex: logsigmoid ---------- + .. autofunction:: paddle.v2.fluid.layers.logsigmoid :noindex: exp --- + .. autofunction:: paddle.v2.fluid.layers.exp :noindex: relu ---- + .. autofunction:: paddle.v2.fluid.layers.relu :noindex: tanh ---- + .. autofunction:: paddle.v2.fluid.layers.tanh :noindex: tanh_shrink ----------- + .. autofunction:: paddle.v2.fluid.layers.tanh_shrink :noindex: softshrink ---------- + .. autofunction:: paddle.v2.fluid.layers.softshrink :noindex: sqrt ---- + .. autofunction:: paddle.v2.fluid.layers.sqrt :noindex: abs ----- +--- + .. autofunction:: paddle.v2.fluid.layers.abs :noindex: ceil ---- + .. autofunction:: paddle.v2.fluid.layers.ceil :noindex: floor ----- + .. autofunction:: paddle.v2.fluid.layers.floor :noindex: round ----- + .. autofunction:: paddle.v2.fluid.layers.round :noindex: reciprocal ---------- + .. autofunction:: paddle.v2.fluid.layers.reciprocal :noindex: log --- + .. autofunction:: paddle.v2.fluid.layers.log :noindex: square ------ + .. autofunction:: paddle.v2.fluid.layers.square :noindex: softplus -------- + .. autofunction:: paddle.v2.fluid.layers.softplus :noindex: softsign ---------- +-------- + .. autofunction:: paddle.v2.fluid.layers.softsign :noindex: brelu ----- + .. autofunction:: paddle.v2.fluid.layers.brelu :noindex: leaky_relu ---------- + .. autofunction:: paddle.v2.fluid.layers.leaky_relu :noindex: soft_relu --------- + .. autofunction:: paddle.v2.fluid.layers.soft_relu :noindex: elu ----- +--- + .. autofunction:: paddle.v2.fluid.layers.elu :noindex: relu6 ----- + .. autofunction:: paddle.v2.fluid.layers.relu6 :noindex: pow ----- +--- + .. autofunction:: paddle.v2.fluid.layers.pow :noindex: +stanh +----- + +.. autofunction:: paddle.v2.fluid.layers.stanh + :noindex: + hard_shrink ----------- + .. autofunction:: paddle.v2.fluid.layers.hard_shrink :noindex: thresholded_relu ---------------- + .. autofunction:: paddle.v2.fluid.layers.thresholded_relu :noindex: hard_sigmoid -------------- +------------ + .. autofunction:: paddle.v2.fluid.layers.hard_sigmoid :noindex: swish ------- +----- + .. autofunction:: paddle.v2.fluid.layers.swish :noindex: + +tensor +====== + +create_tensor +------------- + +.. autofunction:: paddle.v2.fluid.layers.create_tensor + :noindex: + +create_parameter +---------------- + +.. autofunction:: paddle.v2.fluid.layers.create_parameter + :noindex: + +create_global_var +----------------- + +.. autofunction:: paddle.v2.fluid.layers.create_global_var + :noindex: + +cast +---- + +.. autofunction:: paddle.v2.fluid.layers.cast + :noindex: + +concat +------ + +.. autofunction:: paddle.v2.fluid.layers.concat + :noindex: + +sums +---- + +.. autofunction:: paddle.v2.fluid.layers.sums + :noindex: + +assign +------ + +.. autofunction:: paddle.v2.fluid.layers.assign + :noindex: + +fill_constant_batch_size_like +----------------------------- + +.. autofunction:: paddle.v2.fluid.layers.fill_constant_batch_size_like + :noindex: + +fill_constant +------------- + +.. autofunction:: paddle.v2.fluid.layers.fill_constant + :noindex: + +ones +---- + +.. autofunction:: paddle.v2.fluid.layers.ones + :noindex: + +zeros +----- + +.. autofunction:: paddle.v2.fluid.layers.zeros + :noindex: + diff --git a/doc/api/v2/fluid/nets.rst b/doc/api/v2/fluid/nets.rst index cca0dcdf082c0d809fab1aebba2c0b6c7b8efa2a..015581b7660848bdb0845fafe2d3fc05405e6ae6 100644 --- a/doc/api/v2/fluid/nets.rst +++ b/doc/api/v2/fluid/nets.rst @@ -1,27 +1,31 @@ -=========== -Nets -=========== +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + +==== +nets +==== simple_img_conv_pool -------------------- -.. autofunction:: paddle.v2.fluid.nets.simple_img_conv_pool - :noindex: - -img_conv_group ---------------- -.. autofunction:: paddle.v2.fluid.nets.img_conv_group +.. autofunction:: paddle.v2.fluid.nets.simple_img_conv_pool :noindex: - sequence_conv_pool ------------------ + .. autofunction:: paddle.v2.fluid.nets.sequence_conv_pool :noindex: - glu --- + .. autofunction:: paddle.v2.fluid.nets.glu :noindex: +scaled_dot_product_attention +---------------------------- + +.. autofunction:: paddle.v2.fluid.nets.scaled_dot_product_attention + :noindex: + diff --git a/doc/api/v2/fluid/optimizer.rst b/doc/api/v2/fluid/optimizer.rst index 19b4940f08de3e2f7dc177f2961e538946d10a78..1691ebb9a7cb16da96e04147d0adea322374f529 100644 --- a/doc/api/v2/fluid/optimizer.rst +++ b/doc/api/v2/fluid/optimizer.rst @@ -1,54 +1,49 @@ -=========== -Optimizer -=========== - -Optimizer ------------ -.. automodule:: paddle.v2.fluid.optimizer - :members: Optimizer - :noindex: +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! +========= +optimizer +========= -SGDOptimizer ------------ -.. automodule:: paddle.v2.fluid.optimizer - :members: SGDOptimizer - :noindex: +SGD +--- +.. autoclass:: paddle.v2.fluid.optimizer.SGD + :members: + :noindex: +Momentum +-------- -MomentumOptimizer ------------------ -.. automodule:: paddle.v2.fluid.optimizer - :members: MomentumOptimizer +.. autoclass:: paddle.v2.fluid.optimizer.Momentum + :members: :noindex: +Adagrad +------- - -AdagradOptimizer ----------------- -.. automodule:: paddle.v2.fluid.optimizer - :members: AdagradOptimizer +.. autoclass:: paddle.v2.fluid.optimizer.Adagrad + :members: :noindex: +Adam +---- -AdamOptimizer -------------- -.. automodule:: paddle.v2.fluid.optimizer - :members: AdamOptimizer +.. autoclass:: paddle.v2.fluid.optimizer.Adam + :members: :noindex: +Adamax +------ -AdamaxOptimizer ------------ -.. automodule:: paddle.v2.fluid.optimizer - :members: AdamaxOptimizer +.. autoclass:: paddle.v2.fluid.optimizer.Adamax + :members: :noindex: +DecayedAdagrad +-------------- -DecayedAdagradOptimizer ------------------------ -.. automodule:: paddle.v2.fluid.optimizer - :members: DecayedAdagradOptimizer +.. autoclass:: paddle.v2.fluid.optimizer.DecayedAdagrad + :members: :noindex: diff --git a/doc/api/v2/fluid/param_attr.rst b/doc/api/v2/fluid/param_attr.rst index ca0c8af9e8c4f2271de7a131ad0d27c0e8635f50..8083d0d858dafcd275eaddb9b475875ee42ef724 100644 --- a/doc/api/v2/fluid/param_attr.rst +++ b/doc/api/v2/fluid/param_attr.rst @@ -1,11 +1,21 @@ -=========== +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + +========== +param_attr +========== + ParamAttr -=========== +--------- +.. autoclass:: paddle.v2.fluid.param_attr.ParamAttr + :members: + :noindex: +WeightNormParamAttr +------------------- -ParamAttr ------------ -.. automodule:: paddle.v2.fluid.param_attr - :members: ParamAttr +.. autoclass:: paddle.v2.fluid.param_attr.WeightNormParamAttr + :members: :noindex: + diff --git a/doc/api/v2/fluid/profiler.rst b/doc/api/v2/fluid/profiler.rst index 7d4042d1f41c12c4a551ba6576559d612116872a..4a1ff7cb6976e0054f77428b699ea679aa91394f 100644 --- a/doc/api/v2/fluid/profiler.rst +++ b/doc/api/v2/fluid/profiler.rst @@ -1,10 +1,25 @@ -=========== -Profiler -=========== +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! +======== +profiler +======== +cuda_profiler +------------- -Profiler ------------ .. autofunction:: paddle.v2.fluid.profiler.cuda_profiler :noindex: + +reset_profiler +-------------- + +.. autofunction:: paddle.v2.fluid.profiler.reset_profiler + :noindex: + +profiler +-------- + +.. autofunction:: paddle.v2.fluid.profiler.profiler + :noindex: + diff --git a/doc/api/v2/fluid/regularizer.rst b/doc/api/v2/fluid/regularizer.rst index 868e225ed3d59e79aeb217fb88081ea25f80fa2c..2c17d15599baa1d02eb87c7b6c40034769ebb3a4 100644 --- a/doc/api/v2/fluid/regularizer.rst +++ b/doc/api/v2/fluid/regularizer.rst @@ -1,25 +1,27 @@ +.. THIS FILE IS GENERATED BY `gen_doc.{py|sh}` + !DO NOT EDIT THIS FILE MANUALLY! + =========== -Regularizer +regularizer =========== -WeightDecayRegularizer ----------------------- -.. automodule:: paddle.v2.fluid.regularizer - :members: WeightDecayRegularizer - :noindex: - +append_regularization_ops +------------------------- -L2DecayRegularizer ------------------- -.. automodule:: paddle.v2.fluid.regularizer - :members: L2DecayRegularizer +.. autofunction:: paddle.v2.fluid.regularizer.append_regularization_ops :noindex: +L1Decay +------- +.. autoclass:: paddle.v2.fluid.regularizer.L1Decay + :members: + :noindex: -L1DecayRegularizer -------------------- -.. automodule:: paddle.v2.fluid.regularizer - :members: L1DecayRegularizer +L2Decay +------- +.. autoclass:: paddle.v2.fluid.regularizer.L2Decay + :members: + :noindex: diff --git a/doc/design/csp.md b/doc/design/csp.md new file mode 100644 index 0000000000000000000000000000000000000000..ba9cacfdea7dcf7c6499b562dfc58400d082f2c8 --- /dev/null +++ b/doc/design/csp.md @@ -0,0 +1,96 @@ +# Design Doc: CSP in PaddlePaddle Fluid + +## Motivation + +Concurrent programming is important for deep learning. Few example applications are: + +1. The main thread keeps reading the next mini-batch while another thread uses the GPU for computing. +2. The main thread performs the computation while another thread uploads the local gradients from each trainer to the parameter server. + +Most DL systems, including TensorFlow, Caffe2, and MxNet, can asynchronously execute operators in a graph. However, Fluid doesn't have the concept of a graph at all, as the design goal of Fluid is that of a programming language. + +## Concurrent Programming Models + +There were many concurrent programming models, implemented in various forms: + +| concurrent programming model | implementation | +|-----|-----| +| mutex | types and functions in standard libraries | +| semaphore | types and functions in standard libraries | +| communicating sequential processes (CSP) | Go programming language | +| actor model | Erlang programming language | +| message passing | MPI | +| bulk synchronous parallel (BSP) | Pregel distributed programming framework | + +Since Fluid was designed to be a programming language, we would like to implement CSP in Fluid. + +### CSP v.s. Actor Model + +A well-known implementation of Actor Model is the Erlang programming language. In Actor Model, *processes* could send messages to another process and receive messages from another process given the process IDs. We can find the three ingredients, process with ID, send, and recv, in MPI too. Indeed, we can rewrite Erlang programs in Python + MPI with possibly fewer lines of code. Our concern with Actor Model is that it doesn't seem reasonable to implement process management in a programming language's runtime library; instead, it should be the operating systems' responsibility to manage processes and libraries like MPI for send/recv. + +## CSP in Fluid + +Fluid has two fundamental control-flows: *if-else* and *while*. If we are to implement CSP, we need the following: + +1. a new data type: *channel* and operators *send* and *recv*, +1. *goroutine* or thread, and +1. a new control-flow: select. + +We also need Python wrappers for the above components. + +The type *channel* is conceptually the blocking queue. In Go, its implemented is a [blocking circular queue](https://github.com/golang/go/blob/68ce117cf17b8debf5754bfd476345779b5b6616/src/runtime/chan.go#L31-L50), which supports send and recv. + +The `select` operation has been in OS kernels long before Go language. All Unix kernels implement system calls *poll* and *select*. They monitor multiple file descriptors to see if I/O is possible on any of them. This takes O(N) time. Since Linux 2.6, a new system call, *epoll*, can do the same in O(1) time. In BSD systems, there is a similar system call *kqueue*. Go's Linux implementation uses epoll. + +It might be a good idea to implement Fluid's select using epoll too. In this design doc, we start from the O(N) way, so we could focus on Python binding and the syntax. + +### Type Channel + +Fluid supports many data types: + +1. Tensor, +1. Row-sparse Tensor +1. LoD Tensor, +1. Tensor array, etc + +Each data type is registered in the [`framework.proto`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/framework.proto#L117-L127) as an enum value. To add a new type channel, we need to add a new type enum. + +To expose a C++ type to Python, we need to edit the [`pybind.cc`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/pybind/pybind.cc) file. [Here](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/pybind/pybind.cc#L120-L164) is an example how we expose C++ class LoDTensor. + +## Syntax Design + +### Create Channel + +In Go, we create a channel by specifying the element type and buffer size: + +```go +ch := make(chan int) // a channel without buffer +ch1 := make(chan int, 100) // a channel that can buffer 100 ints. +``` + +In Fluid, we should be able to do the same: + +```python +ch = fluid.make_chan(dtype=INT) +ch1 = fluid.make_chan(dtype=INT, 100) +``` + +In addition to that, we want channels that can hold more complex element types, e.g., Tensors of float16: + +```python +ch = fluid.make_chan(dtype=Tensor, etype=float16) +``` + +or Tensors of Tensors of float16 etc. + +The point here is that we need a consistent way to compose types, like in C++ we can have `Tensor...> >`. + +### Send and Recv + +### Select + +## Example Programs + +### 1. RPC between Trainers and Parameter Servers + +### 2. Concurrent Minibatch Loading diff --git a/doc/design/dist_refactor/distributed_architecture.md b/doc/design/dist_refactor/distributed_architecture.md index 3a741f95866fb6c301ca9097af7916281f2278cf..9368c5780dc922953f38bf0f86d9f797a4a8a6fe 100644 --- a/doc/design/dist_refactor/distributed_architecture.md +++ b/doc/design/dist_refactor/distributed_architecture.md @@ -152,12 +152,12 @@ for data in train_reader(): `JobDesc` object describe the distributed job resource specification to run on Cluster environment. - + `RemoteExecutor.run` sends the `ProgramDesc` and [TrainingJob](https://github.com/PaddlePaddle/cloud/blob/develop/doc/autoscale/README.md#training-job-resource) to a server in the cluster which executes `RemoteExecutor.listen`. This server is responsible -to start the final Kubernetes Jobs to run the different role of `ProgramDesc`. +to start the final Kubernetes Jobs to run the different role of `ProgramDesc` from `ConfigMap`. ### Placement Algorithm diff --git a/doc/design/dist_refactor/parameter_server.md b/doc/design/dist_refactor/parameter_server.md index 1094f06d461275a9ad4034d5e48b39856d967b71..805dd13048d41b995d2a01cda52b2ea33e4bbe1d 100644 --- a/doc/design/dist_refactor/parameter_server.md +++ b/doc/design/dist_refactor/parameter_server.md @@ -9,16 +9,16 @@ different purposes. ## Background -The previous implementations of the parameter server does not run a +The previous implementations of the parameter server do not run a fluid sub-program. Parameter initialization, optimizer computation, network communication and checkpointing are implemented twice on both the -trainer and the parameter server. +trainer as well as the parameter server. -It would be great if we can write code once and use them on both the -trainer and the parameter server: reduces code duplication and -improves extensibility. Given that after the current refactor, we are -representing everything as a computing graph on the -trainer. Representing everything as a computing graph on the parameter +It would be great if we can write code once and use them on both: the +trainer and the parameter server, since this reduces code duplication and +improves extensibility. Given that after the current refactoring, we are +representing everything as a computation graph on the +trainer. Representing everything as a computation graph on the parameter server becomes a natural extension. ## Design @@ -30,9 +30,9 @@ into sub-programs to be scheduled on different nodes with the following steps: 1. OP placement: the OPs will be placed on different nodes according - to heuristic that minimizes estimated total computation + to a heuristic that minimizes the estimated total computation time. Currently we will use a simple heuristic that puts parameter - varable on parameter server workers and everything else on trainer + variable on parameter server workers and everything else on trainer workers. 1. Add communication OPs to enable the communication between nodes. @@ -47,22 +47,22 @@ After converting: -1. The parameter variable W and it's optimizer program are placed on the parameter server. +1. The parameter variable W and its optimizer program are placed on the parameter server. 1. Operators are added to the program. - *Send* sends data to the connected *Recv* operator. The scheduler on the receive node will only schedule *Recv* operator to run when the *Send* operator has ran (the *Send* OP will mark the *Recv* OP runnable automatically). - - *Enueue* enqueues the input variable, it can block until space + - *Enqueue* enqueues the input variable, it can block until space become available in the queue. - *Dequeue* outputs configurable numbers of tensors from the - queue. It will block until the queue have the required number of + queue. It will block until the queue has the required number of tensors. ### Benefits -- Model parallelism become easier to implement: it's an extension to +- Model parallelism becomes easier to implement: it is an extension to the trainer - parameter server approach. We can have several "Transpilers" to achieve different goals. - User-defined optimizer is easier to add - user can now express it as @@ -72,22 +72,22 @@ After converting: ### Challenges -- It's important to balance the parameter shards of on multiple - parameter server. If a single parameter is very big (some +- It is important to balance the parameter shards on multiple + parameter servers. If a single parameter is very big (for example: some word-embedding, fully connected, softmax layer), we need to automatically partition the single parameter onto different parameter servers when possible (only element-wise optimizer depends on the parameter variable). -- In the "Aync SGD" figure, the "W" variable on the parameter server - could be read and wrote concurrently. See +- In the "Async SGD" figure, the "W" variable on the parameter server + could be read and written concurrently. See [here](https://github.com/PaddlePaddle/Paddle/pull/6394) for more - details about concurrent program in fluid. + details about concurrent program in Fluid. ### Discussion - Can the Enqueue OP be implemented under our current tensor design - (puts the input tensor into the queue tensor)? -- *Dequeue* OP will have variable numbers of output (depends on the + (put the input tensor into the queue tensor)? +- *Dequeue* OP will have variable numbers of output (depending on the `min_count` attribute), does our current design support it? (similar question for the *Add* OP) diff --git a/doc/design/dist_refactor/src/remote_executor.graffle b/doc/design/dist_refactor/src/remote_executor.graffle index ce2c18fee5687732053c48af9c8c290a994a8090..41b2067311694b56d211a4f32d1b76884eeffd2d 100644 Binary files a/doc/design/dist_refactor/src/remote_executor.graffle and b/doc/design/dist_refactor/src/remote_executor.graffle differ diff --git a/doc/design/dist_refactor/src/remote_executor.png b/doc/design/dist_refactor/src/remote_executor.png index 6be4b1841b99efdb59557975485d0387f422308c..744e2fb2e0f1bbe058e991ba7b2a09000965ee79 100644 Binary files a/doc/design/dist_refactor/src/remote_executor.png and b/doc/design/dist_refactor/src/remote_executor.png differ diff --git a/doc/design/error_clip.md b/doc/design/error_clip.md index 8e845462cce2a29556bcb6010b08f00fbc3d99d7..58aa73b8cd38d01e2426278a3479714e4fb6a3b0 100644 --- a/doc/design/error_clip.md +++ b/doc/design/error_clip.md @@ -46,12 +46,12 @@ class ErrorClipByValue(BaseErrorClipAttr): self.min = min def append_clip_op(self, block, grad_name): - block.append_op( - type="clip", - inputs={"X": grad_name}, - outputs={"Out": grad_name}, - attrs={"min": self.min, - "max": self.max}) + clip_op_desc = block.desc.append_op() + clip_op_desc.set_type("clip") + clip_op_desc.set_input("X", [grad_name]) + clip_op_desc.set_output("Out", [grad_name]) + clip_op_desc.set_attr("min", self.min) + clip_op_desc.set_attr("max", self.max) ``` The `BaseErrorClipAttr` have one main member functions: `append_clip_op(self, block, grad_name)`. @@ -80,6 +80,11 @@ def error_clip_callback(block, context): op_desc.output_arg_names()): fwd_var = block.var_recursive(grad_to_var[grad_n]) error_clip = getattr(fwd_var, "error_clip", None) + if not (error_clip is None or isinstance(error_clip, + BaseErrorClipAttr)): + raise TypeError( + "Variable's error_clip should be an instance of BaseErrorClipAttr or None." + ) if error_clip is not None: error_clip.append_clip_op(block, grad_n) ``` diff --git a/doc/design/fluid.md b/doc/design/fluid.md index 585dc8ef39c0cfb30f470d79f7b27a59ceb5e940..2acc168007d25a083f588b48f84e12e29baf4f47 100644 --- a/doc/design/fluid.md +++ b/doc/design/fluid.md @@ -105,18 +105,10 @@ There are two ways to execute a Fluid program. When a program is executed, it c There is a C++ class [`Executor`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/executor.h), which runs a `ProgramDesc`, similar to how an interpreter runs a Python program. -Fluid is moving towards the direction of a compiler, which is explain in more detail later in this article. +Fluid is moving towards the direction of a compiler, which is explain in [fluid_compiler.md](fluid_compiler.md). ## Backward Compatibility of Fluid Given all the advantages from the removal of the concept of a *model*, hardware manufacturers might still prefer the existence of the concept of a model, so it would be easier for them to support multiple frameworks all at once and could run a trained model during inference. For example, Nervana, a startup company acquired by Intel, has been working on an XPU that reads the models in the format known as [n-graph](https://github.com/NervanaSystems/ngraph). Similarly, [Movidius](https://www.movidius.com/) is producing a mobile deep learning chip that reads and runs graphs of operators. The well-known [ONNX](https://github.com/onnx/onnx) is also a file format of graphs of operators. For Fluid, we can write a converter that extracts the parts in the `ProgramDesc` protobuf message, converts them into a graph of operators, and exports the graph into the ONNX or n-graph format. - -## Towards a Deep Learning Language and the Compiler - -We can change the `if-then-else` and loop structure a little bit in the above Fluid example programs, to make it into a new programming language, different than Python. - -Even if we do not invent a new language, as long as we get the `ProgramDesc` message filled in, we can write a transpiler, which translates each invocation to an operator, into a C++ call to a kernel function of that operator. For example, a transpiler that weaves the CUDA kernels outputs an NVIDIA-friendly C++ program, which can be built using `nvcc`. Another transpiler could generate MKL-friendly code that should be built using `icc` from Intel. More interestingly, we can translate a Fluid program into its distributed version of two `ProgramDesc` messages, one for running on the trainer process, and the other one for the parameter server. For more details of the last example, the [concurrent programming design](concurrent_programming.md) document would be a good pointer. The following figure explains the proposed two-stage process: - -![](fluid-compiler.png) diff --git a/doc/design/fluid_compiler.md b/doc/design/fluid_compiler.md new file mode 100644 index 0000000000000000000000000000000000000000..2a6beafc52e815fa067b273bb5887ddcf6ab15ae --- /dev/null +++ b/doc/design/fluid_compiler.md @@ -0,0 +1,110 @@ +# PaddlePaddle Fluid: Towards a Compiled Programming Language + +As described in [fluid.md](fluid.md), when a Fluid application program +runs, it generates a `ProgramDesc` protobuf message as an intermediate +representation of itself. The C++ class `Executor` can run this +protobuf message as an interpreter. This article describes the Fluid +compiler. + +![](fluid-compiler.png) + +## ProgramDesc + +Before we go deeper into the idea of compiled language, let us take a +look at a simple example Fluid application. + +```python +import "fluid" + +func paddlepaddle() { + X = fluid.read(...) + W = fluid.Tensor(...) + Y = fluid.mult(X, W) +} +``` + +This program consists of a [block](block.md) of three operators -- +`read`, `assign`, and `mult`. Its `ProgramDesc` message looks like +the following + +```protobuf +message ProgramDesc { + block[0] = Block { + vars = [X, W, Y], + ops = [ + read(output = X) + assign(input = ..., output = W) + mult(input = {X, W}, output = Y) + ], + } +} +``` + +## Transpilers + +We can write a transpiler program that takes a `ProgramDesc`, e.g., +the above one, and outputs another `ProgramDesc`. Let us take some +examples: + +1. *Memory optimization transpiler*: We can write a transpiler that + inserts some `FreeMemoryOp`s in the above example `ProgramDesc` so + to free memory early, before the end of an iteration, so to keep a + small memory footprint. + +1. *Distributed training transpiler*: We can write a transpiler that + converts a`ProgramDesc` into its distributed version of two + `ProgramDesc`s -- one for running by the trainer processes and the + other for the parameter server. + +In the rest of this article, we talk about a special kind of +transpiler, *Native code generator*, which takes a `ProgramDesc` and +generates a `.cu` (or `.cc`) file, which could be built by C++ +compilers (gcc, nvcc, icc) into binaries. + +## Native Code Generator + +For the above example, the native code generator transpiler, say, the +CUDA code generator, should generate a `main` function: + +```c++ +void main() { + auto X = fluid_cuda_read(...); + auto W = fluid_cuda_create_tensor(...); + auto Y = fluid_cuda_mult(X, W); +} +``` + +and the definitions of functions `fluid_cuda_read`, +`fluid_cuda_create_tensor`, and `fluid_cuda_mult`. Please be aware +that each function could just define a C++ instance of an operator and +run it. For example + +```c++ +paddle::Tensor fluid_cuda_read(...) { + paddle::Tensor t; + paddle::operator::Read r(&t, ...); + r.Run(); + return t; +} +``` + +For computational operators that have multiple *kernels*, each for a +specific hardware platform, for example, the `mult` operator, the +generated code should call its CUDA kernel: + +```c++ +paddle::Tensor fluid_cuda_mult(const paddle::Tensor& a, + const paddle::Tensor& b) { + paddle::Tensor t; + paddle::operator::Mult m(a, b, ...); + Mult.Run(cuda_context); +} +``` + +where `cuda_context` could be a global variable of type +`paddle::CUDADeviceContext`. + +## Multi-Block Code Generation + +Most Fluid application programs may have more than one blocks. To +execute them, we need to trace [scopes](scope.md). diff --git a/doc/design/ops/sequence_decoder.md b/doc/design/ops/sequence_decoder.md index 9db5fb8e9a9f89b004bf71ddc064cd976c0d0bee..c4a9bbeeefca0e05c335dd60233691e8bac33015 100644 --- a/doc/design/ops/sequence_decoder.md +++ b/doc/design/ops/sequence_decoder.md @@ -22,7 +22,7 @@ The current `LoDTensor` is designed to store levels of variable-length sequences The integers in each level represent the begin and end (not inclusive) offset of a sequence **in the underlying tensor**, let's call this format the **absolute-offset LoD** for clarity. -The relative-offset LoD can retrieve any sequence very quickly but fails to represent empty sequences, for example, a two-level LoD is as follows +The absolute-offset LoD can retrieve any sequence very quickly but fails to represent empty sequences, for example, a two-level LoD is as follows ```python [[0, 3, 9] [0, 2, 3, 3, 3, 9]] @@ -119,7 +119,7 @@ def generate(): encoder_ctx_expanded = pd.lod_expand(encoder_ctx, target_word) decoder_input = pd.fc( act=pd.activation.Linear(), - input=[target_word, encoder_ctx], + input=[target_word, encoder_ctx_expanded], size=3 * decoder_dim) gru_out, cur_mem = pd.gru_step( decoder_input, mem=decoder_mem, size=decoder_dim) diff --git a/doc/design/speech/README.MD b/doc/design/speech/deep_speech_2.md similarity index 85% rename from doc/design/speech/README.MD rename to doc/design/speech/deep_speech_2.md index 7304650e628dba210488cd2dc4836318b5383b2a..cfdc4d6df04344c70d3334626bd38eca997c31ff 100644 --- a/doc/design/speech/README.MD +++ b/doc/design/speech/deep_speech_2.md @@ -140,7 +140,19 @@ TODO by Assignees ### Beam Search with CTC and LM -TODO by Assignees +
+
+Figure 2. Algorithm for CTC Beam Search Decoder. +
+ +- The **Beam Search Decoder** for DS2 CTC-trained network follows the similar approach in \[[3](#references)\] as shown in Figure 2, with two important modifications for the ambiguous parts: + - 1) in the iterative computation of probabilities, the assignment operation is changed to accumulation for one prefix may comes from different paths; + - 2) the if condition ```if l^+ not in A_prev then``` after probabilities' computation is deprecated for it is hard to understand and seems unnecessary. +- An **external scorer** would be passed into the decoder to evaluate a candidate prefix during decoding whenever a white space appended in English decoding and any character appended in Mandarin decoding. +- Such external scorer consists of language model, word count or any other custom scorers. +- The **language model** is built from Task 5, with parameters should be carefully tuned to achieve minimum WER/CER (c.f. Task 7) +- This decoder needs to perform with **high efficiency** for the convenience of parameters tuning and speech recognition in reality. + ## Future Work @@ -153,3 +165,4 @@ TODO by Assignees 1. Dario Amodei, etc., [Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin](http://proceedings.mlr.press/v48/amodei16.pdf). ICML 2016. 2. Dario Amodei, etc., [Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin](https://arxiv.org/abs/1512.02595). arXiv:1512.02595. +3. Awni Y. Hannun, etc. [First-Pass Large Vocabulary Continuous Speech Recognition using Bi-Directional Recurrent DNNs](https://arxiv.org/abs/1408.2873). arXiv:1408.2873 diff --git a/doc/design/speech/image/beam_search.png b/doc/design/speech/image/beam_search.png new file mode 100644 index 0000000000000000000000000000000000000000..7f7e35f34223162d0f7f0ed97375909c43b830ae Binary files /dev/null and b/doc/design/speech/image/beam_search.png differ diff --git a/doc/design/support_new_device.md b/doc/design/support_new_device.md index 4c5f10e2ecb9ec09b78926ca27552741d02d7cc9..8983df900460127fc130043c52373dab505363ba 100644 --- a/doc/design/support_new_device.md +++ b/doc/design/support_new_device.md @@ -2,9 +2,9 @@ ## Background -Deep learning has a high demand for computing resources. New high-performance devices and computing libraries are appearing very frequently. Deep learning frameworks have to integrate these high-performance devices and computing libraries flexibly and efficiently. +Deep learning has a high demand for computing resources. New high-performance devices and computing libraries are appearing very frequently. Deep learning frameworks have to integrate these high-performance devices and computing libraries in a flexible and efficient manner. -On one hand, hardware and computing libraries usually do not have a one-to-one correspondence. For example,Intel CPUs support Eigen and MKL computing libraries while Nvidia GPUs support Eigen and cuDNN computing libraries. We have to implement operator specific kernels for each computing library. +On one hand, hardware and computing libraries usually do not have a one-to-one correspondence. For example, Intel CPUs support Eigen and MKL computing libraries while Nvidia GPUs support Eigen and cuDNN computing libraries. We have to implement operator specific kernels for each computing library. On the other hand, users usually do not want to care about the low-level hardware and computing libraries when writing a neural network configuration. In Fluid, `Layer` is exposed in `Python`, and `Operator` is exposed in `C++`. Both `Layer` and `Operator` are hardware independent. @@ -17,7 +17,7 @@ For a general overview of fluid, please refer to the [overview doc](https://gith There are mainly three parts that we have to consider while integrating a new device/library: -- Place and DeviceContext: indicates the device id and manages hardware resources +- Place and DeviceContext: indicate the device id and manage hardware resources - Memory and Tensor: malloc/free data on certain device @@ -25,10 +25,10 @@ There are mainly three parts that we have to consider while integrating a new de ### Place and DeviceContext -Please remind that device and computing library are not one-to-one corresponding. A device can have a lot of computing libraries and a computing library can also support several devices. +Please note that device and computing library are not one-to-one corresponding. A device can have a lot of computing libraries and a computing library can also support several devices. #### Place -Fluid uses class [Place](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/place.h#L55) to represent the device memory where data is located. If we add another device, we have to add corresponding `DevicePlace`. +Fluid uses class [Place](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/place.h#L55) to represent the device memory where data is located. If we add another device, we have to add the corresponding `DevicePlace`. ``` | CPUPlace @@ -144,7 +144,7 @@ class Tensor { }; ``` -`Placeholder` is used to delay memory allocation; that is, we can first define a tensor, using `Resize` to configure its shape, and then call `mutuable_data` to allocate the actual memory. +`Placeholder` is used to delay memory allocation; that is, we can first define a tensor, using `Resize` to configurate its shape, and then call `mutuable_data` to allocate the actual memory. ```cpp paddle::framework::Tensor t; @@ -163,7 +163,7 @@ Fluid implements computing units based on different DeviceContexts. Some computi Let's take [MaxOutFunctor](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/math/maxouting.h#L27) as an example: -The interface is defined in header file. +The interface is defined in the header file. ``` template @@ -174,7 +174,7 @@ class MaxOutFunctor { }; ``` -CPU implemention is in .cc file +CPU implementation is in .cc file ``` template @@ -188,7 +188,7 @@ class MaxOutFunctor { }; ``` -CUDA implemention is in .cu file +CUDA implementation is in .cu file ``` template @@ -203,9 +203,9 @@ class MaxOutFunctor { ``` -We get computing handle from a concrete DeviceContext, and make compution on tensors. +We first obtain the computing handle from a concrete DeviceContext and then compute on tensors. -The implemention of `OpKernel` is similar to math functors, the extra thing we need to do is to register the OpKernel in a global map. +The implementation of `OpKernel` is similar to math functors, the extra thing we need to do is to register the OpKernel in a global map. Fluid provides different register interfaces in op_registry.h @@ -231,7 +231,7 @@ REGISTER_OP_CUDA_KERNEL( ## Advanced topics: How to switch between different Device/Library -Generally, we will impelement OpKernel for all Device/Library of an Operator. We can easily train a Convolutional Neural Network in GPU. However, some OpKernel is not sutibale on a specific Device. For example, crf operator can only run on CPU, whereas most other operators can run at GPU. To achieve high performance in such circumstance, we have to switch between different Device/Library. +Generally, we will implement OpKernel for all Device/Library of an Operator. We can easily train a Convolutional Neural Network in GPU. However, some OpKernel is not suitable on a specific Device. For example, crf operator can only run on CPU, whereas most other operators can run on GPU. To achieve high performance in such circumstance, we have to switch between different Device/Library. For more details, please refer to following docs: diff --git a/doc/design/switch_kernel.md b/doc/design/switch_kernel.md index 1846e5d9f99dd433b44ac6b5ae52893ec8f0d451..9719e031c70979cd95400701efd30879662e19bc 100644 --- a/doc/design/switch_kernel.md +++ b/doc/design/switch_kernel.md @@ -1,21 +1,24 @@ ## Background -Every operator has many kernels because there are multiple data types, places, data layout that Fluid supports. We use the `KernelType` to describe kernel types that operators can hold. +Every operator has many kernels because there are multiple data types, places, data layout, library type that Fluid supports. We use the `OpKernelType ` to describe kernel types that operators can hold. -The `KernelType` is as follows. +The `OpKernelType ` is as follows: -``` -struct KernelType { +```cpp +struct OpKernelType { Place place_; DataType data_type_; - LayoutType layout_; + DataLayout data_layout_; + LibraryType library_type_; }; ``` -The `place_` is a descriptor of the device and the computational library, e.g., `MKLDNNPlace`, `CUDAPlace`. +- The `place_` is a descriptor of the device, e.g., CPUPlace, CUDAPlace. -The `data_type_` is the data type that this kernel performs on, e.g., `FP32`, `INT64`. Note that one kernel may have inputs with different data types. However, it will be a major `data_type`. For example, the `cross_entropy` takes `int64` as it label, and `double`/`float` as its input logit and output cost. The major `data_type` of `cross_entropy` is `float`/`double`. +- The `data_type_` is the data type that this kernel performs on, e.g., `FP32`, `INT64`. Note that one kernel may have inputs with different data types. However, it will be a major `data_type`. For example, the `cross_entropy` takes `int64` as it label, and `double`/`float` as its input logit and output cost. The major `data_type` of `cross_entropy` is `float` or `double`. -The `layout` is useful for some computational library. One example is that MKLDNN uses many kinds of layout, such as `nChw8c`. Each kind of layout will invoke the different kernel. +- The `data_layout_ ` is useful for some computational library. One example is that MKLDNN uses many kinds of layout, such as `nChw8c`. Each kind of layout will invoke the different kernel. + +- The `library_type_` describes the computational library, e.g., `MKLDNN`, `CUDNN`. ## Problem @@ -25,42 +28,72 @@ We register a kernel for every operator and every kernel type ideally. However, 2. Some operators will take too many memory. It is better to force them into CPU. However, the rest of operators in this neural network will be performed on GPU, i.e., model parallel problem. 3. Some layout and place are particular. One example is that MKLDNN uses `nChw8` and there is no other library uses `nChw8c`. -Problems under these situations are similar. We can formalise this problem as follow. +Take one situation to give a detailed explanation, if we have two Operators: OP1 and OP2, OP1 has one output `op1_to_op2`, and `op1_to_op2` is the input of OP2. + +If OP1 and OP2 run on the same place(for example CPUPlace), then `op1_2_op2` can be used directly by OP2. + +``` +OP1(CPUPlace) + | + op1_2_op2 + | +OP2(CPUPlace) +``` + +If OP1 and OP2 run one different place, then OP2 cannot `use op1_2_op2` directly. + +Problems under these situations are similar. We can formalize this problem as follow. We register kernels with types $KT = \{kt_1, kt_2, kt_3, ...\}$ for one operator. The inputs of this operator should be run on kernel type $kt_{?}$, which the $kt_{?} \notin KT$. How to cast the input of this operator from $kt_{?}$ to any of kernel type in $KT$. -## Solution +## Solution: data transform -It is clearly that transforming inputs of an operator toadapt another kernel type is not related to the particular operator. So we should register these transformation methods as global methods. +It is clear that transforming inputs of an operator to adapt another kernel type is not related to the particular operator. So we should register these transformation methods as global methods. -We can infer a kernel type from the inputs of an operators. We let this kernel type as `actual kernel type`, which means this kernel type is the actually kernel type that operator should be performed. +We can infer kernel type for each input of an operator. We let this kernel type as `actual kernel type for var`, which means this kernel type is the kernel type that can process this input variable. We can get a kernel type by 1) The configuration of operator description. (Users may want to force use `MKL` for `conv` operator). 2) The place of the current executor. (Executor is running on GPU). This kernel type is what we expect the operator will be performed on. We let this kernel type as `expect kernel type`. -We transform the input data from `actual` to `expect` if the expect kernel type is not as same as actual kernel type. +We transform the input data from `actual` to `expect` if the actual kernel type is not as same as expect kernel type. -The algorithm is described as follow +The algorithm is described as following ```cpp -using DataTransformationFN = std::function; -using KernelTypePair = std::pair; - -map g_data_transformation_; - -void OpWithKernel::Run() { - vec inputs = ... - auto actual_kernel_type = GetActualKernelType(inputs); - - // The expected kernel type is related to actual kernel type. - // For the most operators, the expected kernel type is as same as - // actual kernel type. - // - // So we pass `actual_kernel_type` as a parameter of - // GetExpectedKernelType - auto expect_kernel_type = GetExpectedKernelType(actual_kernel_type); - - auto trans = g_data_transformation_[{actual_kernel_type, expect_kernel_type}]; - - kernel.run(trans(inputs)); +void OperatorWithKernel::Run( + const Scope& scope, + const platform::Place& place) const { + ExecutionContext ctx(...); + auto expected_kernel_key = this->GetExpectedKernelType(ctx); + + Scope& new_scope = scope.NewScope(); + + for (auto& var_name : this->Inputs()) { + auto* tensor_in = GetTensor(var_name); + auto kernel_type_for_var = this->GetKernelTypeForVar(...); + if (kernel_type_for_var.place_ != expected_kernel_key.place_) { + auto* trans_var = new_scope.Var(var_name); + auto* out = DataTransform(expected_kernel_key, + kernel_type_for_var, + *tensor_in); + CopyVariableWithTensor(...); + } + } + + auto kernel = kernels.find(expected_kernel_key); + kernel->Compute(ExecutionContext(...)); } ``` + +then the actual process for the multi-device above will be: + +``` +OP1(CPUPlace) + | +op1_2_op2(on CPU) + | +[transform](from CPU to GPU) + | +op1_2_op2(on GPU) + | +OP2(CUDAPlace) +``` diff --git a/doc/faq/local/index_cn.rst b/doc/faq/local/index_cn.rst index b331d9d36e6a279881c3b1a5586835e7186957fb..0306b1e5dd25a55545e464ce847291c33576575f 100644 --- a/doc/faq/local/index_cn.rst +++ b/doc/faq/local/index_cn.rst @@ -211,3 +211,49 @@ decoder_inputs = paddle.layer.fc( * list 中元素的个数等于网络中输出层的个数； * list 中每个元素是一个layer的输出结果矩阵，类型是numpy的ndarray； * 每一个layer输出矩阵的高度，在非序列输入时：等于样本数；序列输入时等于：输入序列中元素的总数；宽度等于配置中layer的size； + +6. 如何在训练过程中获得某一个layer的output +----------------------------------------------- + +可以在event_handler中，通过 :code:`event.gm.getLayerOutputs("layer_name")` 获得在模型配置中某一层的name :code:`layer_name` 在当前 +mini-batch forward的output的值。获得的值类型均为 :code:`numpy.ndarray` ，可以通过这个输出来完成自定义的评估指标计算等功能。例如下面代码： + +.. code-block:: python + + def score_diff(right_score, left_score): + return np.average(np.abs(right_score - left_score)) + + def event_handler(event): + if isinstance(event, paddle.event.EndIteration): + if event.batch_id % 25 == 0: + diff = score_diff( + event.gm.getLayerOutputs("right_score")["right_score"][ + "value"], + event.gm.getLayerOutputs("left_score")["left_score"][ + "value"]) + logger.info(("Pass %d Batch %d : Cost %.6f, " + "average absolute diff scores: %.6f") % + (event.pass_id, event.batch_id, event.cost, diff)) + +注意：此方法不能获取 :code:`paddle.layer.recurrent_group` 里step的内容，但可以获取 :code:`paddle.layer.recurrent_group` 的输出。 + +7. 如何在训练过程中获得参数的权重和梯度 +----------------------------------------------- + +在某些情况下，获得当前mini-batch的权重（或称作weights, parameters）有助于在训练时观察具体数值，方便排查以及快速定位问题。 +可以通过在 :code:`event_handler` 中打印其值（注意，需要使用 :code:`paddle.event.EndForwardBackward` 保证使用GPU训练时也可以获得）， +示例代码如下： + +.. code-block:: python + + ... + parameters = paddle.parameters.create(cost) + ... + def event_handler(event): + if isinstance(event, paddle.event.EndForwardBackward): + if event.batch_id % 25 == 0: + for p in parameters.keys(): + logger.info("Param %s, Grad %s", + parameters.get(p), parameters.get_grad(p)) + +注意：“在训练过程中获得某一个layer的output”和“在训练过程中获得参数的权重和梯度”都会造成训练中的数据从C++拷贝到numpy，会对训练性能造成影响。不要在注重性能的训练场景下使用。 \ No newline at end of file diff --git a/doc/faq/local/src/reduce_min_pool_size.py b/doc/faq/local/src/reduce_min_pool_size.py index 96073633d2b45bf83927b8cf446919fe916438c2..9efdb5707ac4a0cb701ec4fef6dfff399d6150cb 100644 --- a/doc/faq/local/src/reduce_min_pool_size.py +++ b/doc/faq/local/src/reduce_min_pool_size.py @@ -1,16 +1,18 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + + @provider(min_pool_size=0, ...) def process(settings, filename): os.system('shuf %s > %s.shuf' % (filename, filename)) # shuffle before. diff --git a/doc/faq/local/src/word2vec_config.py b/doc/faq/local/src/word2vec_config.py index 03619b2628ffca6166e8784222b7ea0196194b82..b4fcf0960eda37f98f9f7f2949148f8d51cd2008 100644 --- a/doc/faq/local/src/word2vec_config.py +++ b/doc/faq/local/src/word2vec_config.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + ... # the settings and define data provider is omitted. DICT_DIM = 3000 # dictionary dimension. word_ids = data_layer('word_ids', size=DICT_DIM) diff --git a/doc/faq/local/src/word2vec_dataprovider.py b/doc/faq/local/src/word2vec_dataprovider.py index a439a8f52ebc13ef281012e647834fd53a924d74..3b6273b0574fb54a7c374a1f19fdd993fff5c730 100644 --- a/doc/faq/local/src/word2vec_dataprovider.py +++ b/doc/faq/local/src/word2vec_dataprovider.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + DICT_DIM = 3000 diff --git a/doc/faq/model/index_cn.rst b/doc/faq/model/index_cn.rst index b47bbe05bdb39d1ade9434a7e54bf6ca88a91cc9..6947948bc79f4dba63954c459afb940e3242c405 100644 --- a/doc/faq/model/index_cn.rst +++ b/doc/faq/model/index_cn.rst @@ -67,3 +67,14 @@ * 不同于上述介绍的recurrent layer , :code:`paddle.networks.lstmemory_unit` 定义了LSTM单元在一个时间步内的计算过程，它并不是一个完整的recurrent layer，也不能接收序列数据作为输入； * :code:`paddle.networks.lstmemory_unit` 只能在recurrent_group中作为step function使用； + +5. PaddlePaddle的softmax能否指定计算的维度 +----------------------------------------- + +PaddlePaddle的softmax不能指定计算维度，只能按行计算。 +在图像任务中，对于NCHW，如果需要在C维度计算softmax，可以先使用 :code:`paddle.layer.switch_order` 改变维度顺序，即将NCHW转换成NHWC，再做一定的reshape，最后计算softmax。 + +6. PaddlePaddle是否支持维数可变的数据输入 +------------------------------------------ + +PaddlePaddle提供的 :code:`paddle.data_type.dense_array` 支持维数可变的数据输入。在使用时，将对应数据层的维数设置成一个大于输入数据维数的值用于占位即可。 diff --git a/doc/getstarted/build_and_install/build_from_source_cn.rst b/doc/getstarted/build_and_install/build_from_source_cn.rst index 71904dc41ed0d946867d890cc585e1b88450ca8c..ff904b1022a41612c9680dce92d3fc2c69ad7e93 100644 --- a/doc/getstarted/build_and_install/build_from_source_cn.rst +++ b/doc/getstarted/build_and_install/build_from_source_cn.rst @@ -115,7 +115,7 @@ PaddlePaddle的编译选项，包括生成CPU/GPU二进制文件、链接何种B "WITH_AVX", "是否编译含有AVX指令集的PaddlePaddle二进制文件", "ON" "WITH_PYTHON", "是否内嵌PYTHON解释器", "ON" "WITH_STYLE_CHECK", "是否编译时进行代码风格检查", "ON" - "WITH_TESTING", "是否开启单元测试", "ON" + "WITH_TESTING", "是否开启单元测试", "OFF" "WITH_DOC", "是否编译中英文文档", "OFF" "WITH_SWIG_PY", "是否编译PYTHON的SWIG接口，该接口可用于预测和定制化训练", "Auto" "WITH_GOLANG", "是否编译go语言的可容错parameter server", "ON" diff --git a/doc/getstarted/build_and_install/build_from_source_en.rst b/doc/getstarted/build_and_install/build_from_source_en.rst index 27f73b2e2c029b41d514e1612912ed1c335605b6..718fb869c23a1f7be82c87c726282bded9dad516 100644 --- a/doc/getstarted/build_and_install/build_from_source_en.rst +++ b/doc/getstarted/build_and_install/build_from_source_en.rst @@ -126,7 +126,7 @@ You can add :code:`-D` argument to pass such options, like: "WITH_AVX", "Build with AVX support", "ON" "WITH_PYTHON", "Build with integrated Python interpreter", "ON" "WITH_STYLE_CHECK", "Check code style when building", "ON" - "WITH_TESTING", "Build unit tests", "ON" + "WITH_TESTING", "Build unit tests", "OFF" "WITH_DOC", "Build documentations", "OFF" "WITH_SWIG_PY", "Build Python SWIG interface for V2 API", "Auto" "WITH_GOLANG", "Build fault-tolerant parameter server written in go", "ON" diff --git a/doc/getstarted/build_and_install/docker_install_cn.rst b/doc/getstarted/build_and_install/docker_install_cn.rst index bae42593ddc6f7a7eb47d603752ad6efa9820b45..79d214635a069a739060e0b79424729f6ff90387 100644 --- a/doc/getstarted/build_and_install/docker_install_cn.rst +++ b/doc/getstarted/build_and_install/docker_install_cn.rst @@ -25,14 +25,14 @@ .. code-block:: bash - docker pull docker.paddlepaddle.org/paddle + docker pull docker.paddlepaddlehub.com/paddle 下载GPU版本（cuda8.0_cudnn5_avx_mkl）的Docker镜像： .. code-block:: bash docker pull paddlepaddle/paddle:latest-gpu - docker pull docker.paddlepaddle.org/paddle:latest-gpu + docker pull docker.paddlepaddlehub.com/paddle:latest-gpu 选择下载使用不同的BLAS库的Docker镜像： @@ -49,7 +49,7 @@ docker pull paddlepaddle/paddle:[tag] # 比如： - docker pull docker.paddlepaddle.org/paddle:0.10.0-gpu + docker pull docker.paddlepaddlehub.com/paddle:0.11.0-gpu .. _docker_run: @@ -95,6 +95,12 @@ PaddlePaddle Book是为用户和开发者制作的一个交互式的Jupyter Note docker run -p 8888:8888 paddlepaddle/book +国内用户可以使用下面的镜像源来加速访问： + + .. code-block: bash + + docker run -p 8888:8888 docker.paddlepaddlehub.com/book + 然后在浏览器中输入以下网址： .. code-block:: text diff --git a/doc/getstarted/build_and_install/docker_install_en.rst b/doc/getstarted/build_and_install/docker_install_en.rst index 56a7c68e4d39c45249fa55a964dc48b7081596a6..e0e0559fb858a093db96a9b4ec1c5a45d6c71a38 100644 --- a/doc/getstarted/build_and_install/docker_install_en.rst +++ b/doc/getstarted/build_and_install/docker_install_en.rst @@ -26,14 +26,14 @@ For users in China, we provide a faster mirror: .. code-block:: bash - docker pull docker.paddlepaddle.org/paddle + docker pull docker.paddlepaddlehub.com/paddle Download GPU version (cuda8.0_cudnn5_avx_mkl) images: .. code-block:: bash docker pull paddlepaddle/paddle:latest-gpu - docker pull docker.paddlepaddle.org/paddle:latest-gpu + docker pull docker.paddlepaddlehub.com/paddle:latest-gpu Choose between different BLAS version: @@ -53,7 +53,7 @@ and run: docker pull paddlepaddle/paddle:[tag] # i.e. - docker pull docker.paddlepaddle.org/paddle:0.10.0-gpu + docker pull docker.paddlepaddlehub.com/paddle:0.11.0-gpu .. _docker_run: @@ -102,6 +102,12 @@ We provide a packaged book image, simply issue the command: docker run -p 8888:8888 paddlepaddle/book +For users in China, we provide a faster mirror: + + .. code-block: bash + + docker run -p 8888:8888 docker.paddlepaddlehub.com/book + Then, you would back and paste the address into the local browser: .. code-block:: text diff --git a/doc/getstarted/build_and_install/pip_install_cn.rst b/doc/getstarted/build_and_install/pip_install_cn.rst index 0c741e936b46eda5e7165e4ee54b545b14a28a19..8e4165da6b8135d083766c650f1092158f9d01c2 100644 --- a/doc/getstarted/build_and_install/pip_install_cn.rst +++ b/doc/getstarted/build_and_install/pip_install_cn.rst @@ -39,6 +39,7 @@ PaddlePaddle可以使用常用的Python包管理工具 "cpu_avx_mkl", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" "cpu_avx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "暂无" + "cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "暂无" "cuda7.5_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" diff --git a/doc/getstarted/build_and_install/pip_install_en.rst b/doc/getstarted/build_and_install/pip_install_en.rst index 285ed09805b09790beaef014f6813c227aff33ac..c1e806c0fe5f03139c0dff985f9ae0856eaa2e98 100644 --- a/doc/getstarted/build_and_install/pip_install_en.rst +++ b/doc/getstarted/build_and_install/pip_install_en.rst @@ -42,6 +42,7 @@ If the links below shows up the login form, just click "Log in as guest" to star "cpu_avx_mkl", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" "cpu_avx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "Not Available" + "cpu_noavx_openblas", "`paddlepaddle-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "Not Available" "cuda7.5_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" "cuda8.0_cudnn5_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" "cuda8.0_cudnn7_avx_mkl", "`paddlepaddle_gpu-0.11.0-cp27-cp27mu-linux_x86_64.whl `_", "`paddlepaddle_gpu-0.11.0-cp27-cp27m-linux_x86_64.whl `_", "`paddle.tgz `_" diff --git a/doc/getstarted/concepts/src/infer.py b/doc/getstarted/concepts/src/infer.py index ee71cd7a9a4fbddb93fa3aa2d9349f01f3673982..a1b60388c45ae83cbed3b5597e8b8aef5d69f814 100644 --- a/doc/getstarted/concepts/src/infer.py +++ b/doc/getstarted/concepts/src/infer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2 as paddle import numpy as np diff --git a/doc/getstarted/concepts/src/train.py b/doc/getstarted/concepts/src/train.py index d9c0c66b8a7bbb66d7b66cce38220a4c62fd6849..0e5bdb57bc95c513eb67d426741c860a34d37dd0 100644 --- a/doc/getstarted/concepts/src/train.py +++ b/doc/getstarted/concepts/src/train.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2 as paddle import numpy as np diff --git a/doc/howto/dev/new_op_en.md b/doc/howto/dev/new_op_en.md index 7175d8370d6ce08c6d502eb42b8e53252db89bbb..da8b1bdd1082e439456daf25e9b3a1e8eb534375 100644 --- a/doc/howto/dev/new_op_en.md +++ b/doc/howto/dev/new_op_en.md @@ -4,7 +4,8 @@ - [Implementing C++ Types](#implementing-c-types) - [Defining ProtoMaker](#defining-protomaker) - [Defining Operator](#defining-operator) - - [Registering Operator](#registering-operator) + - [Defining OpKernel](#defining-opkernel) + - [Registering Operator and OpKernel](#registering-operator-and-opkernel) - [Compilation](#compilation) - [Python Binding](#python-binding) - [Unit Tests](#unit-tests) @@ -16,12 +17,13 @@ Here are the base types needed. For details, please refer to the design docs. -- `framework::OperatorBase`: Operator (Op)base class. -- `framework::OpKernel`: Base class for Op computation. -- `framework::OperatorWithKernel`: Inherited from OperatorBase, describing an operator with computation. - `class OpProtoAndCheckerMaker`: Describes an Operator's input, output, attributes and description, mainly used to interface with Python API. +- `framework::OperatorBase`: Operator (Op)base class. +- `framework::OpKernel`: Base class for Op computation kernel. +- `framework::OperatorWithKernel`: Inherited from OperatorBase, describing an operator with computation kernels. + -An operator can be differentiated by whether in has kernel methods. An operator with kernel inherits from `OperatorWithKernel` while the ones without inherit from `OperatorBase`. This tutorial focuses on implementing operators with kernels. In short, an operator includes the following information: +Operators can be categorized into two groups: operator with kernel(s) and operator without kernel(s). An operator with kernel(s) inherits from `OperatorWithKernel` while the one without kernel(s) inherits from `OperatorBase`. This tutorial focuses on implementing operators with kernels. In short, an operator includes the following information: Information | Where is it defined @@ -32,7 +34,7 @@ Kernel implementation | The kernel methods shared between CPU and CUDA are Registering the Op | Ops are registered in `.cc` files; For Kernel registration, `.cc` files contain the CPU implementation, while `.cu` files contain the CUDA implementation. -New Operator implementations are added to the list [paddle/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators), with file names in the format `*_op.h` (if applicable), `*_op.cc`, `*_op.cu` (if applicable).** The system will use the naming scheme to automatically build operators and their corresponding Python extensions. ** +New Operator implementations are added to the list [paddle/operators](https://github.com/PaddlePaddle/Paddle/tree/develop/paddle/operators), with file names in the format `*_op.h` (if applicable), `*_op.cc`, `*_op.cu` (if applicable).** The system will use the naming scheme to automatically build operators and their corresponding Python extensions.** Let's take matrix multiplication operator, [MulOp](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/mul_op.cc), as an example to introduce the writing of an Operator with Kernel. @@ -156,7 +158,8 @@ Usually `OpProtoMaker` and `Op`'s type definitions are written in `.cc` files, w - `typename T` denotes data type, such as `float` or `double`. `MulKernel` types need to rewrite the interface for `Compute`. -- `Compute` takes one input variable `const framework::ExecutionContext& context`. + +- `Compute` takes one input parameter: `const framework::ExecutionContext& context`. - Compared with `InferShapeContext`, `ExecutionContext` includes device types, and can similarly extract input, output, and attribute variables. - `Compute` implements the computation logics of an `OpKernel`. @@ -177,7 +180,7 @@ Usually `OpProtoMaker` and `Op`'s type definitions are written in `.cc` files, w }; ``` -Note that **different devices (CPU, CUDA)share an Op definition; whether or not they share the same `OpKernel` depends on whether `Compute` calls functions that support both devices.** +Note that **different devices (CPU, CUDA)share one Op definition; whether or not they share the same `OpKernel` depends on whether `Compute` calls functions can support both devices.** `MulOp`'s CPU and CUDA share the same `Kernel`. A non-sharing `OpKernel` example can be seen in [`OnehotCrossEntropyOpKernel`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/cross_entropy_op.h#L43). @@ -188,13 +191,14 @@ This concludes the forward implementation of an operator. Next its operation and The definition of its corresponding backward operator, if applicable, is similar to that of an forward operator. **Note that a backward operator does not include a `ProtoMaker`**. -### Registering Operator +### Registering Operator and OpKernel - In `.cc` files, register forward and backward operator classes and the CPU kernel. ```cpp namespace ops = paddle::operators; REGISTER_OP(mul, ops::MulOp, ops::MulOpMaker, mul_grad, ops::MulOpGrad); + REGISTER_OP_CPU_KERNEL(mul, ops::MulKernel); REGISTER_OP_CPU_KERNEL(mul_grad, ops::MulGradKernel); @@ -204,6 +208,7 @@ The definition of its corresponding backward operator, if applicable, is similar - `REGISTER_OP` registers the `ops::MulOp` class, type named `mul`, its type `ProtoMaker` is `ops::MulOpMaker`, registering `ops::MulOpGrad` as `mul_grad`. - `REGISTER_OP_WITHOUT_GRADIENT` registers an operator without gradient. + - `REGISTER_OP_CPU_KERNEL` registers `ops::MulKernel` class and specialized template types `paddle::platform::CPUPlace` and `float`, which also registers `ops::MulGradKernel`. @@ -225,6 +230,7 @@ The definition of its corresponding backward operator, if applicable, is similar Run the following commands to compile. ``` +# maybe you need to rerun cmake make mul_op ``` diff --git a/doc/howto/dev/new_op_kernel_en.md b/doc/howto/dev/new_op_kernel_en.md new file mode 100644 index 0000000000000000000000000000000000000000..123df0a7ee4943c0b789ef9cfa6e0804d0fdd564 --- /dev/null +++ b/doc/howto/dev/new_op_kernel_en.md @@ -0,0 +1,121 @@ +## Add Kernels for a New Device + +### Background + +PaddlePaddle Fluid have hundreds of operators. Each operator could have one or more kernels. A kernel is an implementation of the operator for a certain device, which could be a hardware device, e.g., the CUDA GPU, or a library that utilizes a device, e.g., Intel MKL that makes full use of the Xeon CPU. + +[This document](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/howto/dev/new_op_en.md) explains how to add an operator, and its kernels. The kernels of an operator are indexed by a C++ type [`OpKernelType`](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/operator_kernel_type.md). An operator chooses the right kernel at runtime. This choosing mechanism is described [here](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/switch_kernel.md). + +### Write Kernels for A New Device + +#### Add A New Device + + For some historical reaons, we misuse the word *library* for *device*. For example, we call the deivce type by *library type*. An example is the header file [`library_type.h`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/library_type.h#L24). We will correct this ASAP. + +To register a new device, we need to add an enum value to `LibraryType`: + +``` +enum class LibraryType { + kPlain = 0, + kMKLDNN = 1, + kCUDNN = 2, +}; +``` + + +#### Add A New [Place](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/place.h#L53) + +If you have a new kind of Device, firstly you need to add a new kind of [`Place`](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/place.h#L53). For example `CUDAPlace`: + +```cpp +struct CUDAPlace { + CUDAPlace() : CUDAPlace(0) {} + explicit CUDAPlace(int d) : device(d) {} + + inline int GetDeviceId() const { return device; } + // needed for variant equality comparison + inline bool operator==(const CUDAPlace &o) const { + return device == o.device; + } + inline bool operator!=(const CUDAPlace &o) const { return !(*this == o); } + + int device; +}; + +typedef boost::variant Place; +``` + +#### Add [device context]((https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/device_context.h#L37)) +After a new kind of Device is added, you should add a corresponding [DeviceContext](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/platform/device_context.h#L37) for it. + +```cpp +class DeviceContext { + public: + virtual ~DeviceContext() {} + virtual Place GetPlace() const = 0; + + virtual void Wait() const {} +}; +``` + +#### Implement new [OpKernel](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/operator.h#L351) for your Device. + +A detailed documentation can be found in [`new_op_and_kernel`](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/howto/dev/new_op_en.md) + +```cpp +class OpKernelBase { + public: + /** + * ExecutionContext is the only parameter of Kernel Run function. + * Run will get input/output variables, state such as momentum and + * device resource such as CUDA stream, cublas handle, etc. from + * ExecutionContext. User should construct it before run the Operator. + */ + + virtual void Compute(const ExecutionContext& context) const = 0; + + virtual ~OpKernelBase() = default; +}; + +template +class OpKernel : public OpKernelBase { + public: + using ELEMENT_TYPE = T; +}; +``` + + +#### Register the OpKernel to framework + +After writing the components described above, we should register the kernel to the framework. + +We use `REGISTER_OP_KERNEL` to do the registration. + +```cpp +REGISTER_OP_KERNEL( + op_type, + library_type, + place_type, + kernel0, kernel1, ...) +``` + +kernel0, kernel1 are kernels that have the same `op_type`, `library_type`, `place_type` but different `data_types`. + +take [`conv2d`]((https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/operators/conv_cudnn_op.cu.cc#L318)) as an example: + + ```cpp + REGISTER_OP_KERNEL(conv2d, CPU, paddle::platform::CPUPlace, + paddle::operators::GemmConvKernel, + paddle::operators::GemmConvKernel); + + REGISTER_OP_KERNEL(conv2d, CUDNN, ::paddle::platform::CUDAPlace, + paddle::operators::CUDNNConvOpKernel, + paddle::operators::CUDNNConvOpKernel); + ``` + +In the code above: + + - `conv2d` is the type/name of the operator + - `CUDNN/CPU` is `library` + - `paddle::platform::CUDAPlace/CPUPlace` is `place` + - template parameter `float/double` on `CUDNNConvOpKernel` is `data_type`. diff --git a/doc/howto/optimization/cpu_profiling.md b/doc/howto/optimization/cpu_profiling.md index 1775374cf6e518586c28bbd8e04946c74df7e4c5..368af40cc7308cf6f4c609361078fe3ba02213ed 100644 --- a/doc/howto/optimization/cpu_profiling.md +++ b/doc/howto/optimization/cpu_profiling.md @@ -60,8 +60,7 @@ each column is as follows: | column | meaning | | --- | --- | | ncalls | the number of calls into a function | -| tottime | the total execution time of the function, not including the - execution time of other functions called by the function | +| tottime | the total execution time of the function, not including the execution time of other functions called by the function | | percall | tottime divided by ncalls | | cumtime | the total execution time of the function, including the execution time of other functions being called | | percall | cumtime divided by ncalls | diff --git a/doc/howto/usage/capi/organization_of_the_inputs_cn.md b/doc/howto/usage/capi/organization_of_the_inputs_cn.md index 563ec5ca21ec5d75800fa201943d65e6d6fe51ea..a889ae4ffab7be02468b4a5ac5a18e3cc77803c9 100644 --- a/doc/howto/usage/capi/organization_of_the_inputs_cn.md +++ b/doc/howto/usage/capi/organization_of_the_inputs_cn.md @@ -19,7 +19,7 @@ ### 基本使用概念 -- 在PaddlePaddle内部，神经网络中一个计算层的输入/输出被组织为一个 `Argument` 结构体，如果神经网络有多个输入或者多个输入，每一个输入/输入都会对应有自己的`Argument`。 +- 在PaddlePaddle内部，神经网络中一个计算层的输入/输出被组织为一个 `Argument` 结构体，如果神经网络有多个输入或者多个输出，每一个输入/输出都会对应有自己的`Argument`。 - `Argument` 并不真正“存储”数据，而是将输入/输出信息有机地组织在一起。 - 在`Argument`内部由`IVector`（对应着上文提到的一维整型数组）和`Matrix`（对应着上文提到的二维浮点型矩阵）来实际存储数据；由 `Sequence Start Positions` (下文详细解释) 来描述输入/输出的序列信息。 diff --git a/doc/howto/usage/cluster/fluid_cluster_train_en.md b/doc/howto/usage/cluster/fluid_cluster_train_en.md new file mode 100644 index 0000000000000000000000000000000000000000..ae825d9a517c7e9005d4e32f8f34b3f6a79be0c9 --- /dev/null +++ b/doc/howto/usage/cluster/fluid_cluster_train_en.md @@ -0,0 +1,153 @@ +# Fluid Distributed Training + +## Introduction + +In this article, we'll explain how to configure and run distributed training jobs with PaddlePaddle Fluid in a bare metal cluster. + +## Preparations + +### Getting the cluster ready + +Prepare the compute nodes in the cluster. Nodes in this cluster can be of any specification that runs PaddlePaddle, and with a unique IP address assigned to it. Make sure they can communicate to each other. + +### Have PaddlePaddle installed + +PaddlePaddle must be installed on all nodes. If you have GPU cards on your nodes, be sure to properly install drivers and CUDA libraries. + +PaddlePaddle build and installation guide can be found [here](http://www.paddlepaddle.org/docs/develop/documentation/en/getstarted/build_and_install/index_en.html). + +In addition to above, the `cmake` command should be run with the option `WITH_DISTRIBUTE` set to on. An example bare minimum `cmake` command would look as follows: + +``` bash +cmake .. -DWITH_DOC=OFF -DWITH_GPU=OFF -DWITH_DISTRIBUTE=ON -DWITH_SWIG_PY=ON -DWITH_PYTHON=ON +``` + +### Update the training script + +#### Non-cluster training script + +Let's take [Deep Learning 101](http://www.paddlepaddle.org/docs/develop/book/01.fit_a_line/index.html)'s first chapter: "fit a line" as an example. + +The non-cluster version of this demo with fluid API is as follows: + +``` python +import paddle.v2 as paddle +import paddle.v2.fluid as fluid + +x = fluid.layers.data(name='x', shape=[13], dtype='float32') +y_predict = fluid.layers.fc(input=x, size=1, act=None) +y = fluid.layers.data(name='y', shape=[1], dtype='float32') + +cost = fluid.layers.square_error_cost(input=y_predict, label=y) +avg_cost = fluid.layers.mean(x=cost) + +sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001) +sgd_optimizer.minimize(avg_cost) + +BATCH_SIZE = 20 + +train_reader = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.uci_housing.train(), buf_size=500), + batch_size=BATCH_SIZE) + +place = fluid.CPUPlace() +feeder = fluid.DataFeeder(place=place, feed_list=[x, y]) +exe = fluid.Executor(place) + +exe.run(fluid.default_startup_program()) + +PASS_NUM = 100 +for pass_id in range(PASS_NUM): + fluid.io.save_persistables(exe, "./fit_a_line.model/") + fluid.io.load_persistables(exe, "./fit_a_line.model/") + for data in train_reader(): + avg_loss_value, = exe.run(fluid.default_main_program(), + feed=feeder.feed(data), + fetch_list=[avg_cost]) + + if avg_loss_value[0] < 10.0: + exit(0) # if avg cost less than 10.0, we think our code is good. +exit(1) +``` + +We created a simple fully-connected neural network training program and handed it to the fluid executor to run for 100 passes. + +Now let's try to convert it to a distributed version to run on a cluster. + +#### Introducing parameter server + +As we can see from the non-cluster version of training script, there is only one role in the script: the trainer, that performs the computing as well as holds the parameters. In cluster training, since multi-trainers are working on the same task, they need one centralized place to hold and distribute parameters. This centralized place is called the Parameter Server in PaddlePaddle. + +![parameter server architecture](src/trainer.png) + +Parameter Server in fluid not only holds the parameters but is also assigned with a part of the program. Trainers communicate with parameter servers via send/receive OPs. For more technical details, please refer to [this document](https://github.com/PaddlePaddle/Paddle/blob/develop/doc/design/dist_refactor/distributed_architecture.md). + +Now we need to create programs for both: trainers and parameter servers, the question is how? + +#### Slice the program + +Fluid provides a tool called "Distributed Transpiler" that automatically converts the non-cluster program into cluster program. + +The idea behind this tool is to find the optimize OPs and gradient parameters, slice the program into 2 pieces and connect them with send/receive OP. + +Optimize OPs and gradient parameters can be found from the return values of optimizer's minimize function. + +To put them together: + +``` python +... #define the program, cost, and create sgd optimizer + +optimize_ops, params_grads = sgd_optimizer.minimize(avg_cost) #get optimize OPs and gradient parameters + +t = fluid.DistributeTranspiler() # create the transpiler instance +# slice the program into 2 pieces with optimizer_ops and gradient parameters list, as well as pserver_endpoints, which is a comma separated list of [IP:PORT] and number of trainers +t.transpile(optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) + +... #create executor + +# in pserver, run this +#current_endpoint here means current pserver IP:PORT you wish to run on +pserver_prog = t.get_pserver_program(current_endpoint) +pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) +exe.run(pserver_startup) +exe.run(pserver_prog) + +# in trainer, run this +... # define data reader +exe.run(fluid.default_startup_program()) +for pass_id in range(100): + for data in train_reader(): + exe.run(t.get_trainer_program()) + + +``` + +### E2E demo + +Please find the complete demo from [here](https://github.com/PaddlePaddle/Paddle/blob/develop/python/paddle/v2/fluid/tests/book_distribute/notest_dist_fit_a_line.py). +First `cd` into the folder that contains the `python` files. In this case: + +```bash +cd /paddle/python/paddle/v2/fluid/tests/book_distribute +``` + +In parameter server node run the following in the command line: + +``` bash +PSERVERS=192.168.1.2:6174 SERVER_ENDPOINT=192.168.1.2:6174 TRAINING_ROLE=PSERVER python notest_dist_fit_a_line.py +``` + +*please note we assume that your parameter server runs at 192.168.1.2:6174* + +Wait until the prompt `Server listening on 192.168.1.2:6174` + +Then in 2 of your trainer nodes run this: + +``` bash +PSERVERS=192.168.1.2:6174 SERVER_ENDPOINT=192.168.1.2:6174 TRAINING_ROLE=TRAINER python notest_dist_fit_a_line.py +``` + +*the reason you need to run this command twice in 2 nodes is because: in the script we set the trainer count to be 2. You can change this setting on line 50* + +Now you have 2 trainers and 1 parameter server up and running. diff --git a/doc/howto/usage/cluster/src/k8s_train/start_paddle.py b/doc/howto/usage/cluster/src/k8s_train/start_paddle.py index 1774f8b640c5a2dee036db36b40123b6de7bf68c..935c12bb67e1fe08bc135a7a2220fcd43c548482 100755 --- a/doc/howto/usage/cluster/src/k8s_train/start_paddle.py +++ b/doc/howto/usage/cluster/src/k8s_train/start_paddle.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/python # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved # diff --git a/doc/howto/usage/cluster/src/word2vec/api_train_v2.py b/doc/howto/usage/cluster/src/word2vec/api_train_v2.py index d449e02023f1ec48669ce734a24093aed031adc5..9a65f14628d8a0808dce25187b482354c72a838d 100644 --- a/doc/howto/usage/cluster/src/word2vec/api_train_v2.py +++ b/doc/howto/usage/cluster/src/word2vec/api_train_v2.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import gzip import math diff --git a/doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py b/doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py index a5dd347f0b594e6037182bfad39a6a736fdbab66..2afce9a66e521f9e3a8d566dd23762969f0594f5 100644 --- a/doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py +++ b/doc/howto/usage/cluster/src/word2vec/api_train_v2_cluster.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import math import os import paddle.v2 as paddle diff --git a/doc/howto/usage/cluster/src/word2vec/prepare.py b/doc/howto/usage/cluster/src/word2vec/prepare.py index 24f5c5b26d37ea03de3ab4dc2d967a4bd009eef0..ade01c378efced05787e1f62e6fa6c38f8ac5ad3 100644 --- a/doc/howto/usage/cluster/src/word2vec/prepare.py +++ b/doc/howto/usage/cluster/src/word2vec/prepare.py @@ -1,3 +1,17 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2 as paddle import tarfile import os diff --git a/doc/v1_api_tutorials/README.md b/doc/v1_api_tutorials/README.md deleted file mode 100644 index 071b8da61fbcab3e88819273008b4526546202ad..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/README.md +++ /dev/null @@ -1,5 +0,0 @@ -The tutorials in v1_api_tutorials are using v1_api currently, and will be upgraded to v2_api later. -Thus, v1_api_tutorials is a temporary directory. We decide not to maintain it and will delete it in future. - -Please go to [PaddlePaddle/book](https://github.com/PaddlePaddle/book) and -[PaddlePaddle/models](https://github.com/PaddlePaddle/models) to learn PaddlePaddle. diff --git a/doc/v1_api_tutorials/embedding_model/index_cn.md b/doc/v1_api_tutorials/embedding_model/index_cn.md deleted file mode 100644 index 2b4a79fbbfc0c4af74aa73c540919f5d9cf2635b..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/embedding_model/index_cn.md +++ /dev/null @@ -1,139 +0,0 @@ -# 中文词向量模型的使用 # ----------- -本文档介绍如何在PaddlePaddle平台上,使用预训练的标准格式词向量模型。 - -在此感谢 @lipeng 提出的代码需求，并给出的相关模型格式的定义。 - -## 介绍 ### -### 中文字典 ### -我们的字典使用内部的分词工具对百度知道和百度百科的语料进行分词后产生。分词风格如下： "《红楼梦》"将被分为 "《"，"红楼梦"，"》"，和 "《红楼梦》"。字典采用UTF8编码，输出有2列：词本身和词频。字典共包含 3206326个词和4个特殊标记： - - ``: 分词序列的开始 - - ``: 分词序列的结束 - - `PALCEHOLDER_JUST_IGNORE_THE_EMBEDDING`: 占位符，没有实际意义 - - ``: 未知词 - -### 中文词向量的预训练模型 ### -遵循文章 [A Neural Probabilistic Language Model](http://www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf)中介绍的方法，模型采用 n-gram 语言模型，结构如下图：6元上下文作为输入层->全连接层->softmax层 。对应于字典，我们预训练得到4种不同维度的词向量，分别为：32维、64维、128维和256维。 -
![](./neural-n-gram-model.png)
-
Figure 1. neural-n-gram-model
- -### 下载和数据抽取 ### -运行以下的命令下载和获取我们的字典和预训练模型： - - cd $PADDLE_ROOT/demo/model_zoo/embedding - ./pre_DictAndModel.sh - -## 中文短语改写的例子 ## -以下示范如何使用预训练的中文字典和词向量进行短语改写。 - -### 数据的准备和预处理 ### -首先，运行以下的命令下载数据集。该数据集（utf8编码）包含20个训练样例，5个测试样例和2个生成式样例。 - - cd $PADDLE_ROOT/demo/seqToseq/data - ./paraphrase_data.sh - -第二步，将数据处理成规范格式，在训练数集上训练生成词向量字典（数据将保存在 `$PADDLE_SOURCE_ROOT/demo/seqToseq/data/pre-paraphrase`）: - - cd $PADDLE_ROOT/demo/seqToseq/ - python preprocess.py -i data/paraphrase [--mergeDict] - -- 其中，如果使用`--mergeDict`选项，源语言短语和目标语言短语的字典将被合并（源语言和目标语言共享相同的编码字典）。本实例中，源语言和目标语言都是相同的语言，因此可以使用该选项。 - - -### 使用用户指定的词向量字典 ### -使用如下命令，从预训练模型中，根据用户指定的字典，抽取对应的词向量构成新的词表: - cd $PADDLE_ROOT/demo/model_zoo/embedding - python extract_para.py --preModel PREMODEL --preDict PREDICT --usrModel USRMODEL--usrDict USRDICT -d DIM - -- `--preModel PREMODEL`: 预训练词向量字典模型的路径 -- `--preDict PREDICT`: 预训练模型使用的字典的路径 -- `--usrModel USRMODEL`: 抽取出的新词表的保存路径 -- `--usrDict USRDICT`: 用户指定新的字典的路径，用于构成新的词表 -- `-d DIM`: 参数（词向量）的维度 - -此处，你也可以简单的运行以下的命令： - - cd $PADDLE_ROOT/demo/seqToseq/data/ - ./paraphrase_model.sh - -运行成功以后，你将会看到以下的模型结构： - - paraphrase_model - |--- _source_language_embedding - |--- _target_language_embedding - -### 在PaddlePaddle平台训练模型 ### -首先，配置模型文件，配置如下（可以参考保存在 `demo/seqToseq/paraphrase/train.conf`的配置）: - - from seqToseq_net import * - is_generating = False - - ################## Data Definition ##################### - train_conf = seq_to_seq_data(data_dir = "./data/pre-paraphrase", - job_mode = job_mode) - - ############## Algorithm Configuration ################## - settings( - learning_method = AdamOptimizer(), - batch_size = 50, - learning_rate = 5e-4) - - ################# Network configure ##################### - gru_encoder_decoder(train_conf, is_generating, word_vector_dim = 32) - -这个配置与`demo/seqToseq/translation/train.conf` 基本相同 - -然后，使用以下命令进行模型训练: - - cd $PADDLE_SOURCE_ROOT/demo/seqToseq/paraphrase - ./train.sh - -其中，`train.sh` 与`demo/seqToseq/translation/train.sh` 基本相同，只有2个配置不一样: - -- `--init_model_path`: 初始化模型的路径配置为`data/paraphrase_modeldata/paraphrase_model` -- `--load_missing_parameter_strategy`：如果参数模型文件缺失，除词向量模型外的参数将使用正态分布随机初始化 - -如果用户想要了解详细的数据集的格式、模型的结构和训练过程，请查看 [Text generation Tutorial](../text_generation/index_cn.md). - -## 可选功能 ## -### 观测词向量 -PaddlePaddle 平台为想观测词向量的用户提供了将二进制词向量模型转换为文本模型的功能: - - cd $PADDLE_ROOT/demo/model_zoo/embedding - python paraconvert.py --b2t -i INPUT -o OUTPUT -d DIM - -- `-i INPUT`: 输入的（二进制）词向量模型名称 -- `-o OUTPUT`: 输出的文本模型名称 -- `-d DIM`: （词向量）参数维度 - -运行完以上命令，用户可以在输出的文本模型中看到: - - 0,4,32156096 - -0.7845433,1.1937413,-0.1704215,0.4154715,0.9566584,-0.5558153,-0.2503305, ...... - 0.0000909,0.0009465,-0.0008813,-0.0008428,0.0007879,0.0000183,0.0001984, ...... - ...... - -- 其中，第一行是`PaddlePaddle` 输出文件的格式说明，包含3个属性：: - - `PaddlePaddle`的版本号，本例中为0 - - 浮点数占用的字节数，本例中为4 - - 总计的参数个数，本例中为32,156,096 -- 其余行是（词向量）参数行（假设词向量维度为32） - - 每行打印32个参数以','分隔 - - 共有32,156,096/32 = 1,004,877行，也就是说，模型共包含1,004,877个被向量化的词 - -### 词向量模型的修正 -`PaddlePaddle` 为想修正词向量模型的用户提供了将文本词向量模型转换为二进制模型的命令: - - cd $PADDLE_ROOT/demo/model_zoo/embedding - python paraconvert.py --t2b -i INPUT -o OUTPUT - -- `-i INPUT`: 输入的文本词向量模型名称 -- `-o OUTPUT`: 输出的二进制词向量模型名称 - -请注意，输入的文本格式如下: - - -0.7845433,1.1937413,-0.1704215,0.4154715,0.9566584,-0.5558153,-0.2503305, ...... - 0.0000909,0.0009465,-0.0008813,-0.0008428,0.0007879,0.0000183,0.0001984, ...... - ...... -- 输入文本中没有头部（格式说明）行 -- （输入文本）每行存储一个词，以逗号','分隔 diff --git a/doc/v1_api_tutorials/embedding_model/index_en.md b/doc/v1_api_tutorials/embedding_model/index_en.md deleted file mode 100644 index 9525f64f9b5384c8e44690fb0887fb2293108e0a..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/embedding_model/index_en.md +++ /dev/null @@ -1,140 +0,0 @@ -# Chinese Word Embedding Model Tutorial # ----------- -This tutorial is to guide you through the process of using a Pretrained Chinese Word Embedding Model in the PaddlePaddle standard format. - -We thank @lipeng for the pull request that defined the model schemas and pretrained the models. - -## Introduction ### -### Chinese Word Dictionary ### -Our Chinese-word dictionary is created on Baidu ZhiDao and Baidu Baike by using in-house word segmentor. For example, the participle of "《红楼梦》" is "《"，"红楼梦"，"》"，and "《红楼梦》". Our dictionary (using UTF-8 format) has has two columns: word and its frequency. The total word count is 3206326, including 4 special token: - - ``: the start of a sequence - - ``: the end of a sequence - - `PALCEHOLDER_JUST_IGNORE_THE_EMBEDDING`: a placeholder, just ignore it and its embedding - - ``: a word not included in dictionary - -### Pretrained Chinese Word Embedding Model ### -Inspired by paper [A Neural Probabilistic Language Model](http://www.jmlr.org/papers/volume3/bengio03a/bengio03a.pdf), our model architecture (**Embedding joint of six words->FullyConnect->SoftMax**) is as following graph. And for our dictionary, we pretrain four models with different word vector dimenstions, i.e 32, 64, 128, 256. -
![](./neural-n-gram-model.png)
-
Figure 1. neural-n-gram-model
- -### Download and Extract ### -To download and extract our dictionary and pretrained model, run the following commands. - - cd $PADDLE_ROOT/demo/model_zoo/embedding - ./pre_DictAndModel.sh - -## Chinese Paraphrasing Example ## -We provide a paraphrasing task to show the usage of pretrained Chinese Word Dictionary and Embedding Model. - -### Data Preparation and Preprocess ### - -First, run the following commands to download and extract the in-house dataset. The dataset (using UTF-8 format) has 20 training samples, 5 testing samples and 2 generating samples. - - cd $PADDLE_ROOT/demo/seqToseq/data - ./paraphrase_data.sh - -Second, preprocess data and build dictionary on train data by running the following commands, and the preprocessed dataset is stored in `$PADDLE_SOURCE_ROOT/demo/seqToseq/data/pre-paraphrase`: - - cd $PADDLE_ROOT/demo/seqToseq/ - python preprocess.py -i data/paraphrase [--mergeDict] - -- `--mergeDict`: if using this option, the source and target dictionary are merged, i.e, two dictionaries have the same context. Here, as source and target data are all chinese words, this option can be used. - -### User Specified Embedding Model ### -The general command of extracting desired parameters from the pretrained embedding model based on user dictionary is: - - cd $PADDLE_ROOT/demo/model_zoo/embedding - python extract_para.py --preModel PREMODEL --preDict PREDICT --usrModel USRMODEL--usrDict USRDICT -d DIM - -- `--preModel PREMODEL`: the name of pretrained embedding model -- `--preDict PREDICT`: the name of pretrained dictionary -- `--usrModel USRMODEL`: the name of extracted embedding model -- `--usrDict USRDICT`: the name of user specified dictionary -- `-d DIM`: dimension of parameter - -Here, you can simply run the command: - - cd $PADDLE_ROOT/demo/seqToseq/data/ - ./paraphrase_model.sh - -And you will see following embedding model structure: - - paraphrase_model - |--- _source_language_embedding - |--- _target_language_embedding - -### Training Model in PaddlePaddle ### -First, create a model config file, see example `demo/seqToseq/paraphrase/train.conf`: - - from seqToseq_net import * - is_generating = False - - ################## Data Definition ##################### - train_conf = seq_to_seq_data(data_dir = "./data/pre-paraphrase", - job_mode = job_mode) - - ############## Algorithm Configuration ################## - settings( - learning_method = AdamOptimizer(), - batch_size = 50, - learning_rate = 5e-4) - - ################# Network configure ##################### - gru_encoder_decoder(train_conf, is_generating, word_vector_dim = 32) - -This config is almost the same as `demo/seqToseq/translation/train.conf`. - -Then, train the model by running the command: - - cd $PADDLE_SOURCE_ROOT/demo/seqToseq/paraphrase - ./train.sh - -where `train.sh` is almost the same as `demo/seqToseq/translation/train.sh`, the only difference is following two command arguments: - -- `--init_model_path`: path of the initialization model, here is `data/paraphrase_model` -- `--load_missing_parameter_strategy`: operations when model file is missing, here use a normal distibution to initialize the other parameters except for the embedding layer - -For users who want to understand the dataset format, model architecture and training procedure in detail, please refer to [Text generation Tutorial](../text_generation/index_en.md). - -## Optional Function ## -### Embedding Parameters Observation -For users who want to observe the embedding parameters, this function can convert a PaddlePaddle binary embedding model to a text model by running the command: - - cd $PADDLE_ROOT/demo/model_zoo/embedding - python paraconvert.py --b2t -i INPUT -o OUTPUT -d DIM - -- `-i INPUT`: the name of input binary embedding model -- `-o OUTPUT`: the name of output text embedding model -- `-d DIM`: the dimension of parameter - -You will see parameters like this in output text model: - - 0,4,32156096 - -0.7845433,1.1937413,-0.1704215,0.4154715,0.9566584,-0.5558153,-0.2503305, ...... - 0.0000909,0.0009465,-0.0008813,-0.0008428,0.0007879,0.0000183,0.0001984, ...... - ...... - -- 1st line is **PaddlePaddle format file head**, it has 3 attributes: - - version of PaddlePaddle, here is 0 - - sizeof(float), here is 4 - - total number of parameter, here is 32156096 -- Other lines print the paramters (assume `` = 32) - - each line print 32 paramters splitted by ',' - - there is 32156096/32 = 1004877 lines, meaning there is 1004877 embedding words - -### Embedding Parameters Revision -For users who want to revise the embedding parameters, this function can convert a revised text embedding model to a PaddlePaddle binary model by running the command: - - cd $PADDLE_ROOT/demo/model_zoo/embedding - python paraconvert.py --t2b -i INPUT -o OUTPUT - -- `-i INPUT`: the name of input text embedding model. -- `-o OUTPUT`: the name of output binary embedding model - -Note that the format of input text model is as follows: - - -0.7845433,1.1937413,-0.1704215,0.4154715,0.9566584,-0.5558153,-0.2503305, ...... - 0.0000909,0.0009465,-0.0008813,-0.0008428,0.0007879,0.0000183,0.0001984, ...... - ...... -- there is no file header in 1st line -- each line stores parameters for one word, the separator is commas ',' diff --git a/doc/v1_api_tutorials/embedding_model/neural-n-gram-model.png b/doc/v1_api_tutorials/embedding_model/neural-n-gram-model.png deleted file mode 100644 index f70b765b3fd69816345a79fc59adfea46008dbfd..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/embedding_model/neural-n-gram-model.png and /dev/null differ diff --git a/doc/v1_api_tutorials/gan/gan.png b/doc/v1_api_tutorials/gan/gan.png deleted file mode 100644 index 0eafd7cb49b545f412f8e775804bcd0b22c42454..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/gan/gan.png and /dev/null differ diff --git a/doc/v1_api_tutorials/gan/index_en.md b/doc/v1_api_tutorials/gan/index_en.md deleted file mode 100644 index ac9ed37b2264778869f92c0910b1cb946fb4427f..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/gan/index_en.md +++ /dev/null @@ -1,137 +0,0 @@ -# Generative Adversarial Networks (GAN) - -This demo implements GAN training described in the original [GAN paper](https://arxiv.org/abs/1406.2661) and deep convolutional generative adversarial networks [DCGAN paper](https://arxiv.org/abs/1511.06434). - -The high-level structure of GAN is shown in Figure. 1 below. It is composed of two major parts: a generator and a discriminator, both of which are based on neural networks. The generator takes in some kind of noise with a known distribution and transforms it into an image. The discriminator takes in an image and determines whether it is artificially generated by the generator or a real image. So the generator and the discriminator are in a competitive game in which generator is trying to generate image to look as real as possible to fool the discriminator, while the discriminator is trying to distinguish between real and fake images. - -
![](./gan.png)
-

- Figure 1. GAN-Model-Structure - figure credit -

- -The generator and discriminator take turn to be trained using SGD. The objective function of the generator is for its generated images being classified as real by the discriminator, and the objective function of the discriminator is to correctly classify real and fake images. When the GAN model is trained to converge to the equilibrium state, the generator will transform the given noise distribution to the distribution of real images, and the discriminator will not be able to distinguish between real and fake images at all. - -## Implementation of GAN Model Structure -Since GAN model involves multiple neural networks, it requires to use paddle python API. So the code walk-through below can also partially serve as an introduction to the usage of Paddle Python API. - -There are three networks defined in gan_conf.py, namely **generator_training**, **discriminator_training** and **generator**. The relationship to the model structure we defined above is that **discriminator_training** is the discriminator, **generator** is the generator, and the **generator_training** combined the generator and discriminator since training generator would require the discriminator to provide loss function. This relationship is described in the following code: -```python -if is_generator_training: - noise = data_layer(name="noise", size=noise_dim) - sample = generator(noise) - -if is_discriminator_training: - sample = data_layer(name="sample", size=sample_dim) - -if is_generator_training or is_discriminator_training: - label = data_layer(name="label", size=1) - prob = discriminator(sample) - cost = cross_entropy(input=prob, label=label) - classification_error_evaluator( - input=prob, label=label, name=mode + '_error') - outputs(cost) - -if is_generator: - noise = data_layer(name="noise", size=noise_dim) - outputs(generator(noise)) -``` - -In order to train the networks defined in gan_conf.py, one first needs to initialize a Paddle environment, parse the config, create GradientMachine from the config and create trainer from GradientMachine as done in the code chunk below: -```python -import py_paddle.swig_paddle as api -# init paddle environment -api.initPaddle('--use_gpu=' + use_gpu, '--dot_period=10', - '--log_period=100', '--gpu_id=' + args.gpu_id, - '--save_dir=' + "./%s_params/" % data_source) - -# Parse config -gen_conf = parse_config(conf, "mode=generator_training,data=" + data_source) -dis_conf = parse_config(conf, "mode=discriminator_training,data=" + data_source) -generator_conf = parse_config(conf, "mode=generator,data=" + data_source) - -# Create GradientMachine -dis_training_machine = api.GradientMachine.createFromConfigProto( -dis_conf.model_config) -gen_training_machine = api.GradientMachine.createFromConfigProto( -gen_conf.model_config) -generator_machine = api.GradientMachine.createFromConfigProto( -generator_conf.model_config) - -# Create trainer -dis_trainer = api.Trainer.create(dis_conf, dis_training_machine) -gen_trainer = api.Trainer.create(gen_conf, gen_training_machine) -``` - -In order to balance the strength between generator and discriminator, we schedule to train whichever one is performing worse by comparing their loss function value. The loss function value can be calculated by a forward pass through the GradientMachine. -```python -def get_training_loss(training_machine, inputs): - outputs = api.Arguments.createArguments(0) - training_machine.forward(inputs, outputs, api.PASS_TEST) - loss = outputs.getSlotValue(0).copyToNumpyMat() - return numpy.mean(loss) -``` - -After training one network, one needs to sync the new parameters to the other networks. The code below demonstrates one example of such use case: -```python -# Train the gen_training -gen_trainer.trainOneDataBatch(batch_size, data_batch_gen) - -# Copy the parameters from gen_training to dis_training and generator -copy_shared_parameters(gen_training_machine, -dis_training_machine) -copy_shared_parameters(gen_training_machine, generator_machine) -``` - - -## A Toy Example -With the infrastructure explained above, we can now walk you through a toy example of generating two dimensional uniform distribution using 10 dimensional Gaussian noise. - -The Gaussian noises are generated using the code below: -```python -def get_noise(batch_size, noise_dim): - return numpy.random.normal(size=(batch_size, noise_dim)).astype('float32') -``` - -The real samples (2-D uniform) are generated using the code below: -```python -# synthesize 2-D uniform data in gan_trainer.py:114 -def load_uniform_data(): - data = numpy.random.rand(1000000, 2).astype('float32') - return data -``` - -The generator and discriminator network are built using fully-connected layer and batch_norm layer, and are defined in gan_conf.py. - -To train the GAN model, one can use the command below. The flag -d specifies the training data (cifar, mnist or uniform) and flag --useGpu specifies whether to use gpu for training (0 is cpu, 1 is gpu). -```bash -$python gan_trainer.py -d uniform --useGpu 1 -``` -The generated samples can be found in ./uniform_samples/ and one example is shown below as Figure 2. One can see that it roughly recovers the 2D uniform distribution. - -
![](./uniform_sample.png)
-

- Figure 2. Uniform Sample -

- -## MNIST Example -### Data preparation -To download the MNIST data, one can use the following commands: -```bash -$cd data/ -$./get_mnist_data.sh -``` - -### Model description -Following the DC-Gan paper (https://arxiv.org/abs/1511.06434), we use convolution/convolution-transpose layer in the discriminator/generator network to better deal with images. The details of the network structures are defined in gan_conf_image.py. - -### Training the model -To train the GAN model on mnist data, one can use the following command: -```bash -$python gan_trainer.py -d mnist --useGpu 1 -``` -The generated sample images can be found at ./mnist_samples/ and one example is shown below as Figure 3. -
![](./mnist_sample.png)
-

- Figure 3. MNIST Sample -

diff --git a/doc/v1_api_tutorials/gan/mnist_sample.png b/doc/v1_api_tutorials/gan/mnist_sample.png deleted file mode 100644 index f9c7bf7ddd7f148eac4fe347e9c38afaa8876760..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/gan/mnist_sample.png and /dev/null differ diff --git a/doc/v1_api_tutorials/gan/uniform_sample.png b/doc/v1_api_tutorials/gan/uniform_sample.png deleted file mode 100644 index e716c48e782019a757bed0cb443f2ed97386cbe2..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/gan/uniform_sample.png and /dev/null differ diff --git a/doc/v1_api_tutorials/imagenet_model/resnet_block.jpg b/doc/v1_api_tutorials/imagenet_model/resnet_block.jpg deleted file mode 100644 index e16bd3c624030c4c09b358a015b491141b42d8f1..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/imagenet_model/resnet_block.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/imagenet_model/resnet_model_cn.md b/doc/v1_api_tutorials/imagenet_model/resnet_model_cn.md deleted file mode 100644 index 82ec9d70b345c11aba3aa86f8206eedc8072bb88..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/imagenet_model/resnet_model_cn.md +++ /dev/null @@ -1,284 +0,0 @@ -# Model Zoo - ImageNet # - -[ImageNet](http://www.image-net.org/) 是通用物体分类领域一个众所周知的数据库。本教程提供了一个用于ImageNet上的卷积分类网络模型。 - -## ResNet 介绍 - -论文 [Deep Residual Learning for Image Recognition](http://arxiv.org/abs/1512.03385) 中提出的ResNet网络结构在2015年ImageNet大规模视觉识别竞赛(ILSVRC 2015)的分类任务中赢得了第一名。他们提出残差学习的框架来简化网络的训练，所构建网络结构的的深度比之前使用的网络有大幅度的提高。下图展示的是基于残差的连接方式。左图构造网络模块的方式被用于34层的网络中，而右图的瓶颈连接模块用于50层，101层和152层的网络结构中。 - -
![resnet_block](./resnet_block.jpg)
-
图 1. ResNet 网络模块
- -本教程中我们给出了三个ResNet模型，这些模型都是由原作者提供的模型转换过来的。我们使用PaddlePaddle在ILSVRC的验证集共50,000幅图像上测试了模型的分类错误率，其中输入图像的颜色通道顺序为**BGR**，保持宽高比缩放到短边为256，只截取中心方形的图像区域。分类错误率和模型大小由下表给出。 -
- ----- - - - - - - - - - - - - - - - - - - - - - - - - - - -
ResNetTop-1Model Size
ResNet-5024.9%99M
ResNet-10123.7%173M
ResNet-15223.2%234M
-
- -## ResNet 模型 - -50层，101层和152层的网络配置文件可参照```demo/model_zoo/resnet/resnet.py```。你也可以通过在命令行参数中增加一个参数如```--config_args=layer_num=50```来指定网络层的数目。 - -### 网络可视化 - -你可以通过执行下面的命令来得到ResNet网络的结构可视化图。该脚本会生成一个dot文件，然后可以转换为图片。需要安装graphviz来转换dot文件为图片。 - -``` -cd demo/model_zoo/resnet -./net_diagram.sh -``` - -### 模型下载 - -``` -cd demo/model_zoo/resnet -./get_model.sh -``` -你可以执行上述命令来下载所有的模型和均值文件，如果下载成功，这些文件将会被保存在```demo/model_zoo/resnet/model```路径下。 - -``` -mean_meta_224 resnet_101 resnet_152 resnet_50 -``` - * resnet_50: 50层网络模型。 - * resnet_101: 101层网络模型。 - * resnet_152: 152层网络模型。 - * mean\_meta\_224: 均值图像文件，图像大小为3 x 224 x 224，颜色通道顺序为**BGR**。你也可以使用这三个值: 103.939, 116.779, 123.68。 - -### 参数信息 - -* **卷积层权重** - - 由于每个卷积层后面连接的是batch normalization层，因此该层中没有偏置(bias)参数，并且只有一个权重。 - 形状: `(Co, ky, kx, Ci)` - * Co: 输出特征图的通道数目 - * ky: 滤波器核在垂直方向上的尺寸 - * kx: 滤波器核在水平方向上的尺寸 - * Ci: 输入特征图的通道数目 - - 二维矩阵: (Co * ky * kx, Ci), 行优先次序存储。 - -* **全连接层权重** - - 二维矩阵: (输入层尺寸, 本层尺寸), 行优先次序存储。 - -* **[Batch Normalization]() 层权重** - -本层有四个参数，实际上只有.w0和.wbias是需要学习的参数，另外两个分别是滑动均值和方差。在测试阶段它们将会被加载到模型中。下表展示了batch normalization层的参数。 -
- ----- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
参数名尺寸含义
_res2_1_branch1_bn.w0256gamma, 缩放参数
_res2_1_branch1_bn.w1256特征图均值
_res2_1_branch1_bn.w2256特征图方差
_res2_1_branch1_bn.wbias256beta, 偏置参数
-
- -### 参数读取 - -使用者可以使用下面的Python脚本来读取参数值: - -``` -import sys -import numpy as np - -def load(file_name): - with open(file_name, 'rb') as f: - f.read(16) # skip header for float type. - return np.fromfile(f, dtype=np.float32) - -if __name__=='__main__': - weight = load(sys.argv[1]) -``` - -或者直接使用下面的shell命令: - -``` -od -j 16 -f _res2_1_branch1_bn.w0 -``` - -## 特征提取 - -我们提供了C++和Python接口来提取特征。下面的例子使用了`demo/model_zoo/resnet/example`中的数据，详细地展示了整个特征提取的过程。 - -### C++接口 - -首先，在配置文件中的`define_py_data_sources2`里指定图像数据列表，具体请参照示例`demo/model_zoo/resnet/resnet.py`。 - -``` - train_list = 'train.list' if not is_test else None - # mean.meta is mean file of ImageNet dataset. - # mean.meta size : 3 x 224 x 224. - # If you use three mean value, set like: - # "mean_value:103.939,116.779,123.68;" - args={ - 'mean_meta': "model/mean_meta_224/mean.meta", - 'image_size': 224, 'crop_size': 224, - 'color': True,'swap_channel:': [2, 1, 0]} - define_py_data_sources2(train_list, - 'example/test.list', - module="example.image_list_provider", - obj="processData", - args=args) -``` - -第二步，在`resnet.py`文件中指定要提取特征的网络层的名字。例如， - -``` -Outputs("res5_3_branch2c_conv", "res5_3_branch2c_bn") -``` - -第三步，在`extract_fea_c++.sh`文件中指定模型路径和输出的目录，然后执行下面的命令。 - -``` -cd demo/model_zoo/resnet -./extract_fea_c++.sh -``` - -如果执行成功，特征将会存到`fea_output/rank-00000`文件中，如下所示。同时你可以使用`load_feature.py`文件中的`load_feature_c`接口来加载该文件。 - -``` --0.115318 -0.108358 ... -0.087884;-1.27664 ... -1.11516 -2.59123; --0.126383 -0.116248 ... -0.00534909;-1.42593 ... -1.04501 -1.40769; -``` - -* 每行存储的是一个样本的特征。其中，第一行存的是图像`example/dog.jpg`的特征，第二行存的是图像`example/cat.jpg`的特征。 -* 不同层的特征由分号`;`隔开，并且它们的顺序与`Outputs()`中指定的层顺序一致。这里，左边是`res5_3_branch2c_conv`层的特征，右边是`res5_3_branch2c_bn`层特征。 - -### Python接口 - -示例`demo/model_zoo/resnet/classify.py`中展示了如何使用Python来提取特征。下面的例子同样使用了`./example/test.list`中的数据。执行的命令如下： - -``` -cd demo/model_zoo/resnet -./extract_fea_py.sh -``` - -extract_fea_py.sh: - -``` -python classify.py \ - --job=extract \ - --conf=resnet.py\ - --use_gpu=1 \ - --mean=model/mean_meta_224/mean.meta \ - --model=model/resnet_50 \ - --data=./example/test.list \ - --output_layer="res5_3_branch2c_conv,res5_3_branch2c_bn" \ - --output_dir=features - -``` -* \--job=extract: 指定工作模式来提取特征。 -* \--conf=resnet.py: 网络配置文件。 -* \--use_gpu=1: 指定是否使用GPU。 -* \--model=model/resnet_50: 模型路径。 -* \--data=./example/test.list: 数据列表。 -* \--output_layer="xxx,xxx": 指定提取特征的层。 -* \--output_dir=features: 输出目录。 - -如果运行成功，你将会看到特征存储在`features/batch_0`文件中，该文件是由cPickle产生的。你可以使用`load_feature.py`中的`load_feature_py`接口来打开该文件，它将返回如下的字典： - -``` -{ -'cat.jpg': {'res5_3_branch2c_conv': array([[-0.12638293, -0.116248 , -0.11883899, ..., -0.00895038, 0.01994277, -0.00534909]], dtype=float32), 'res5_3_branch2c_bn': array([[-1.42593431, -1.28918779, -1.32414699, ..., -1.45933616, -1.04501402, -1.40769434]], dtype=float32)}, -'dog.jpg': {'res5_3_branch2c_conv': array([[-0.11531784, -0.10835785, -0.08809858, ...,0.0055237, 0.01505112, -0.08788397]], dtype=float32), 'res5_3_branch2c_bn': array([[-1.27663755, -1.18272924, -0.90937918, ..., -1.25178063, -1.11515927, -2.59122872]], dtype=float32)} -} -``` - -仔细观察，这些特征值与上述使用C++接口提取的结果是一致的。 - -## 预测 - -`classify.py`文件也可以用于对样本进行预测。我们提供了一个示例脚本`predict.sh`，它使用50层的ResNet模型来对`example/test.list`中的数据进行预测。 - -``` -cd demo/model_zoo/resnet -./predict.sh -``` - -predict.sh调用了`classify.py`: - -``` -python classify.py \ - --job=predict \ - --conf=resnet.py\ - --multi_crop \ - --model=model/resnet_50 \ - --use_gpu=1 \ - --data=./example/test.list -``` -* \--job=extract: 指定工作模型进行预测。 -* \--conf=resnet.py: 网络配置文件。network configure. -* \--multi_crop: 使用10个裁剪图像块，预测概率取平均。 -* \--use_gpu=1: 指定是否使用GPU。 -* \--model=model/resnet_50: 模型路径。 -* \--data=./example/test.list: 数据列表。 - -如果运行成功，你将会看到如下结果，其中156和285是这些图像的分类标签。 - -``` -Label of example/dog.jpg is: 156 -Label of example/cat.jpg is: 282 -``` diff --git a/doc/v1_api_tutorials/imagenet_model/resnet_model_en.md b/doc/v1_api_tutorials/imagenet_model/resnet_model_en.md deleted file mode 100644 index 478ad06193b14ba7fe02238df621db1f7b0804d4..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/imagenet_model/resnet_model_en.md +++ /dev/null @@ -1,284 +0,0 @@ -# Model Zoo - ImageNet # - -[ImageNet](http://www.image-net.org/) is a popular dataset for generic object classification. This tutorial provides convolutional neural network(CNN) models for ImageNet. - -## ResNet Introduction - -ResNets from paper [Deep Residual Learning for Image Recognition](http://arxiv.org/abs/1512.03385) won the 1st place on the ILSVRC 2015 classification task. They present residual learning framework to ease the training of networks that are substantially deeper than those used previously. The residual connections are shown in following figure. The left building block is used in network of 34 layers and the right bottleneck building block is used in network of 50, 101, 152 layers . - -
![resnet_block](./resnet_block.jpg)
-
Figure 1. ResNet Block
- -We present three ResNet models, which are converted from the models provided by the authors . The classfication errors tested in PaddlePaddle on 50,000 ILSVRC validation set with input images channel order of **BGR** by single scale with the shorter side of 256 and single crop as following table. -
- ----- - - - - - - - - - - - - - - - - - - - - - - - - - - -
ResNetTop-1Model Size
ResNet-5024.9%99M
ResNet-10123.7%173M
ResNet-15223.2%234M
-
- -## ResNet Model - -See ```demo/model_zoo/resnet/resnet.py```. This config contains network of 50, 101 and 152 layers. You can specify layer number by adding argument like ```--config_args=layer_num=50``` in command line arguments. - -### Network Visualization - -You can get a diagram of ResNet network by running the following commands. The script generates dot file and then converts dot file to PNG file, which needs to install graphviz to convert. - -``` -cd demo/model_zoo/resnet -./net_diagram.sh -``` - -### Model Download - -``` -cd demo/model_zoo/resnet -./get_model.sh -``` -You can run above command to download all models and mean file and save them in ```demo/model_zoo/resnet/model``` if downloading successfully. - -``` -mean_meta_224 resnet_101 resnet_152 resnet_50 -``` - * resnet_50: model of 50 layers. - * resnet_101: model of 101 layers. - * resnet_152: model of 152 layers. - * mean\_meta\_224: mean file with 3 x 224 x 224 size in **BGR** order. You also can use three mean values: 103.939, 116.779, 123.68. - -### Parameter Info - -* **Convolution Layer Weight** - - As batch normalization layer is connected after each convolution layer, there is no parameter of bias and only one weight in this layer. - shape: `(Co, ky, kx, Ci)` - * Co: channle number of output feature map. - * ky: filter size in vertical direction. - * kx: filter size in horizontal direction. - * Ci: channle number of input feature map. - - 2-Dim matrix: (Co * ky * kx, Ci), saved in row-major order. - -* **Fully connected Layer Weight** - - 2-Dim matrix: (input layer size, this layer size), saved in row-major order. - -* **[Batch Normalization]() Layer Weight** - -There are four parameters in this layer. In fact, only .w0 and .wbias are the learned parameters. The other two are therunning mean and variance respectively. They will be loaded in testing. Following table shows parameters of a batch normzalization layer. -
- ----- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Parameter NameNumberMeaning
_res2_1_branch1_bn.w0256gamma, scale parameter
_res2_1_branch1_bn.w1256mean value of feature map
_res2_1_branch1_bn.w2256variance of feature map
_res2_1_branch1_bn.wbias256beta, shift parameter
-
- -### Parameter Observation - -Users who want to observe the parameters can use Python to read: - -``` -import sys -import numpy as np - -def load(file_name): - with open(file_name, 'rb') as f: - f.read(16) # skip header for float type. - return np.fromfile(f, dtype=np.float32) - -if __name__=='__main__': - weight = load(sys.argv[1]) -``` - -or simply use following shell command: - -``` -od -j 16 -f _res2_1_branch1_bn.w0 -``` - -## Feature Extraction - -We provide both C++ and Python interfaces to extract features. The following examples use data in `demo/model_zoo/resnet/example` to show the extracting process in detail. - -### C++ Interface - -First, specify image data list in `define_py_data_sources2` in the config, see example `demo/model_zoo/resnet/resnet.py`. - -``` - train_list = 'train.list' if not is_test else None - # mean.meta is mean file of ImageNet dataset. - # mean.meta size : 3 x 224 x 224. - # If you use three mean value, set like: - # "mean_value:103.939,116.779,123.68;" - args={ - 'mean_meta': "model/mean_meta_224/mean.meta", - 'image_size': 224, 'crop_size': 224, - 'color': True,'swap_channel:': [2, 1, 0]} - define_py_data_sources2(train_list, - 'example/test.list', - module="example.image_list_provider", - obj="processData", - args=args) -``` - -Second, specify layers to extract features in `Outputs()` of `resnet.py`. For example, - -``` -Outputs("res5_3_branch2c_conv", "res5_3_branch2c_bn") -``` - -Third, specify model path and output directory in `extract_fea_c++.sh`, and then run the following commands. - -``` -cd demo/model_zoo/resnet -./extract_fea_c++.sh -``` - -If successful, features are saved in `fea_output/rank-00000` as follows. And you can use `load_feature_c` interface in `load_feature.py ` to load such a file. - -``` --0.115318 -0.108358 ... -0.087884;-1.27664 ... -1.11516 -2.59123; --0.126383 -0.116248 ... -0.00534909;-1.42593 ... -1.04501 -1.40769; -``` - -* Each line stores features of a sample. Here, the first line stores features of `example/dog.jpg` and second line stores features of `example/cat.jpg`. -* Features of different layers are splitted by `;`, and their order is consistent with the layer order in `Outputs()`. Here, the left features are `res5_3_branch2c_conv` layer and right features are `res5_3_branch2c_bn` layer. - -### Python Interface - -`demo/model_zoo/resnet/classify.py` is an example to show how to use Python to extract features. Following example still uses data of `./example/test.list`. Command is as follows: - -``` -cd demo/model_zoo/resnet -./extract_fea_py.sh -``` - -extract_fea_py.sh: - -``` -python classify.py \ - --job=extract \ - --conf=resnet.py\ - --use_gpu=1 \ - --mean=model/mean_meta_224/mean.meta \ - --model=model/resnet_50 \ - --data=./example/test.list \ - --output_layer="res5_3_branch2c_conv,res5_3_branch2c_bn" \ - --output_dir=features - -``` -* \--job=extract: specify job mode to extract feature. -* \--conf=resnet.py: network configure. -* \--use_gpu=1: speficy GPU mode. -* \--model=model/resnet_5: model path. -* \--data=./example/test.list: data list. -* \--output_layer="xxx,xxx": specify layers to extract features. -* \--output_dir=features: output diretcoty. - -If run successfully, you will see features saved in `features/batch_0`, this file is produced with cPickle. You can use `load_feature_py` interface in `load_feature.py` to open the file, and it returns a dictionary as follows: - -``` -{ -'cat.jpg': {'res5_3_branch2c_conv': array([[-0.12638293, -0.116248 , -0.11883899, ..., -0.00895038, 0.01994277, -0.00534909]], dtype=float32), 'res5_3_branch2c_bn': array([[-1.42593431, -1.28918779, -1.32414699, ..., -1.45933616, -1.04501402, -1.40769434]], dtype=float32)}, -'dog.jpg': {'res5_3_branch2c_conv': array([[-0.11531784, -0.10835785, -0.08809858, ...,0.0055237, 0.01505112, -0.08788397]], dtype=float32), 'res5_3_branch2c_bn': array([[-1.27663755, -1.18272924, -0.90937918, ..., -1.25178063, -1.11515927, -2.59122872]], dtype=float32)} -} -``` - -Observed carefully, these feature values are consistent with the above results extracted by C++ interface. - -## Prediction - -`classify.py` also can be used to predict. We provide an example script `predict.sh` to predict data in `example/test.list` using a ResNet model with 50 layers. - -``` -cd demo/model_zoo/resnet -./predict.sh -``` - -predict.sh calls the `classify.py`: - -``` -python classify.py \ - --job=predict \ - --conf=resnet.py\ - --multi_crop \ - --model=model/resnet_50 \ - --use_gpu=1 \ - --data=./example/test.list -``` -* \--job=extract: speficy job mode to predict. -* \--conf=resnet.py: network configure. -* \--multi_crop: use 10 crops and average predicting probability. -* \--use_gpu=1: speficy GPU mode. -* \--model=model/resnet_50: model path. -* \--data=./example/test.list: data list. - -If run successfully, you will see following results, where 156 and 285 are labels of the images. - -``` -Label of example/dog.jpg is: 156 -Label of example/cat.jpg is: 282 -``` diff --git a/doc/v1_api_tutorials/quick_start/index_cn.rst b/doc/v1_api_tutorials/quick_start/index_cn.rst deleted file mode 100644 index d565fcf95ef8489eb22a5a1b5a552b5336f4e371..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/quick_start/index_cn.rst +++ /dev/null @@ -1,397 +0,0 @@ -============= -快速入门教程 -============= - -我们将以 `文本分类问题 `_ 为例, -介绍PaddlePaddle的基本使用方法。 - -安装 -==== - -请参考 :ref:`install_steps` 安装PaddlePaddle。 - -使用概述 -======== - -**文本分类问题**：对于给定的一条文本，我们从提前给定的类别集合中选择其所属类别。 - -比如, 在购物网站上，通过查看买家对某个产品的评价反馈, 评估该产品的质量。 - -- 这个显示器很棒！ （好评） -- 用了两个月之后这个显示器屏幕碎了。（差评） - -使用PaddlePaddle, 每一个任务流程都可以被划分为如下五个步骤。 - - .. image:: src/Pipeline_cn.jpg - :align: center - :scale: 80% - -1. 数据格式准备 - - 本例每行保存一条样本，类别Id和文本信息用 ``Tab`` 间隔，文本中的单词用空格分隔（如果不切词，则字与字之间用空格分隔），例如：``类别Id '\t' 这 个 显 示 器 很 棒 ！`` -2. 向系统传送数据 - - PaddlePaddle可以执行用户的python脚本程序来读取各种格式的数据文件。 - - 本例的所有字符都将转换为连续整数表示的Id传给模型。 -3. 描述网络结构和优化算法 - - 本例由易到难展示4种不同的文本分类网络配置：逻辑回归模型，词向量模型，卷积模型，时序模型。 - - 常用优化算法包括Momentum, RMSProp，AdaDelta，AdaGrad，Adam，Adamax等，本例采用Adam优化方法，加了L2正则和梯度截断。 -4. 训练模型 -5. 应用模型 - -数据格式准备 ------------- - -接下来我们将展示如何用PaddlePaddle训练一个文本分类模型，将 `Amazon电子产品评论数据 `_ 分为好评(正样本)和差评(负样本)两种类别。 -`源代码 `_ 的 ``demo/quick_start`` 目录里提供了该数据的下载脚本和预处理脚本，你只需要在命令行输入以下命令，就能够很方便的完成数据下载和相应的预处理工作。 - -.. code-block:: bash - - cd demo/quick_start - ./data/get_data.sh - ./preprocess.sh - -数据预处理完成之后，通过配置类似于 ``dataprovider_*.py`` 的数据读取脚本和类似于 ``trainer_config.*.py`` 的训练模型脚本，PaddlePaddle将以设置参数的方式来设置 -相应的数据读取脚本和训练模型脚本。接下来，我们将对这两个步骤给出了详细的解释，你也可以先跳过本文的解释环节，直接进入训练模型章节, 使用 ``sh train.sh`` 开始训练模型， -查看`train.sh`内容，通过 **自底向上法** (bottom-up approach)来帮助你理解PaddlePaddle的内部运行机制。 - - -向系统传送数据 -============== - -Python脚本读取数据 ------------------- - -`DataProvider` 是PaddlePaddle负责提供数据的模块，主要职责在于将训练数据传入内存或者显存，让模型能够得到训练更新，其包括两个函数： - -* initializer：PaddlePaddle会在调用读取数据的Python脚本之前，先调用initializer函数。在下面例子里，我们在initialzier函数里初始化词表，并且在随后的读取数据过程中填充词表。 -* process：PaddlePaddle调用process函数来读取数据。每次读取一条数据后，process函数会用yield语句输出这条数据，从而能够被PaddlePaddle 捕获 (harvest)。 - -``dataprovider_bow.py`` 文件给出了完整例子： - -.. literalinclude:: ../../../demo/quick_start/dataprovider_bow.py - :language: python - :lines: 21-70 - :linenos: - :emphasize-lines: 8,33 - -详细内容请参见 :ref:`api_dataprovider` 。 - -配置中的数据加载定义 --------------------- - -在模型配置中通过 ``define_py_data_sources2`` 接口来加载数据： - -.. literalinclude:: ../../../demo/quick_start/trainer_config.emb.py - :language: python - :lines: 19-35 - :linenos: - :emphasize-lines: 12 - - -以下是对上述数据加载的解释： - -- data/train.list,data/test.list: 指定训练数据和测试数据 -- module="dataprovider_bow": 处理数据的Python脚本文件 -- obj="process": 指定生成数据的函数 -- args={"dictionary": word_dict}: 额外的参数，这里指定词典 - -更详细数据格式和用例请参考 :ref:`api_pydataprovider2` 。 - -模型网络结构 -============ - -本小节我们将介绍模型网络结构。 - - .. image:: src/PipelineNetwork_cn.jpg - :align: center - :scale: 80% - - -我们将以最基本的逻辑回归网络作为起点，并逐渐展示更加深入的功能。更详细的网络配置连接请参考 :ref:`api_trainer_config_helpers_layers` 。 -所有配置都能在 `源代码 `_ 的 ``demo/quick_start`` 目录下找到。 - -逻辑回归模型 ------------- - -具体流程如下: - - .. image:: src/NetLR_cn.jpg - :align: center - :scale: 80% - -- 获取利用 `one-hot vector `_ 表示的每个单词，维度是词典大小 - - .. code-block:: python - - word = data_layer(name="word", size=word_dim) - -- 获取该条样本类别Id，维度是类别个数。 - - .. code-block:: python - - label = data_layer(name="label", size=label_dim) - -- 利用逻辑回归模型对该向量进行分类，同时会计算分类准确率 - - .. code-block:: python - - # Define a fully connected layer with logistic activation (also called softmax activation). - output = fc_layer(input=word, - size=label_dim, - act_type=SoftmaxActivation()) - # Define cross-entropy classification loss and error. - classification_cost(input=output, label=label) - - - - input: 除去data层，每个层都有一个或多个input,多个input以list方式输入 - - size: 该层神经元个数 - - act_type: 激活函数类型 - -**效果总结**：我们将在后面介绍训练和预测流程的脚本。在此为方便对比不同网络结构，我们总结了各个网络的复杂度和效果。 - - ===================== =============================== ================= - 网络名称 参数数量 错误率 - ===================== =============================== ================= - 逻辑回归 252 KB 8.652 % - ===================== =============================== ================= - -词向量模型 ----------- - -embedding模型需要稍微改变提供数据的Python脚本，即 ``dataprovider_emb.py``，词向量模型、 -卷积模型、时序模型均使用该脚本。其中文本输入类型定义为整数时序类型integer_value_sequence。 - -.. code-block:: python - - def initializer(settings, dictionary, **kwargs): - settings.word_dict = dictionary - settings.input_types = [ - # Define the type of the first input as sequence of integer. - # The value of the integers range from 0 to len(dictrionary)-1 - integer_value_sequence(len(dictionary)), - # Define the second input for label id - integer_value(2)] - - @provider(init_hook=initializer) - def process(settings, file_name): - ... - # omitted, it is same as the data provider for LR model - -该模型依然使用逻辑回归分类网络的框架， 只是将句子用连续向量表示替换为用稀疏向量表示， 即对第三步进行替换。句子表示的计算更新为两步： - -.. image:: src/NetContinuous_cn.jpg - :align: center - :scale: 80% - -- 利用单词Id查找该单词对应的连续向量(维度为word_dim)， 输入N个单词，输出为N个word_dim维度向量 - - .. code-block:: python - - emb = embedding_layer(input=word, size=word_dim) - -- 将该句话包含的所有单词向量求平均, 得到句子的表示 - - .. code-block:: python - - avg = pooling_layer(input=emb, pooling_type=AvgPooling()) - -其它部分和逻辑回归网络结构一致。 - -**效果总结：** - - ===================== =============================== ================== - 网络名称 参数数量 错误率 - ===================== =============================== ================== - 词向量模型 15 MB 8.484 % - ===================== =============================== ================== - -卷积模型 ------------ - -卷积网络是一种特殊的从词向量表示到句子表示的方法， 也就是将词向量模型进一步演化为三个新步骤。 - -.. image:: src/NetConv_cn.jpg - :align: center - :scale: 80% - -文本卷积分可为三个步骤: - -1. 首先，从每个单词左右两端分别获取k个相邻的单词, 拼接成一个新的向量； - -2. 其次，对该向量进行非线性变换(例如Sigmoid变换), 使其转变为维度为hidden_dim的新向量； - -3. 最后，对整个新向量集合的每一个维度取最大值来表示最后的句子。 - -这三个步骤可配置为: - -.. code-block:: python - - text_conv = sequence_conv_pool(input=emb, - context_start=k, - context_len=2 * k + 1) - -**效果总结：** - - ===================== =============================== ======================== - 网络名称 参数数量 错误率 - ===================== =============================== ======================== - 卷积模型 16 MB 5.628 % - ===================== =============================== ======================== - -时序模型 ----------- - -.. image:: src/NetRNN_cn.jpg - :align: center - :scale: 80% - -时序模型，也称为RNN模型, 包括简单的 `RNN模型 `_, `GRU模型 `_ 和 `LSTM模型 `_ 等等。 - -- GRU模型配置： - - .. code-block:: python - - gru = simple_gru(input=emb, size=gru_size) - - -- LSTM模型配置： - - .. code-block:: python - - lstm = simple_lstm(input=emb, size=lstm_size) - -本次试验，我们采用单层LSTM模型，并使用了Dropout，**效果总结：** - - ===================== =============================== ========================= - 网络名称 参数数量 错误率 - ===================== =============================== ========================= - 时序模型 16 MB 4.812 % - ===================== =============================== ========================= - -优化算法 -========= - -`优化算法 `_ 包括 -Momentum, RMSProp，AdaDelta，AdaGrad，ADAM，Adamax等，这里采用Adam优化方法，同时使用了L2正则(L2 Regularization)和梯度截断(Gradient Clipping)。 - -.. code-block:: python - - settings(batch_size=128, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -训练模型 -========= - -在数据加载和网络配置完成之后， 我们就可以训练模型了。 - -.. image:: src/PipelineTrain_cn.jpg - :align: center - :scale: 80% - -训练模型，我们只需要运行 ``train.sh`` 训练脚本： - - .. code-block:: bash - - ./train.sh - -``train.sh`` 中包含了训练模型的基本命令。训练时所需设置的主要参数如下： - - .. code-block:: bash - - paddle train \ - --config=trainer_config.py \ - --log_period=20 \ - --save_dir=./output \ - --num_passes=15 \ - --use_gpu=false - -这里只简单介绍了单机训练，如何进行分布式训练，请参考 :ref:`cluster_train` 。 - -预测 -===== - -当模型训练好了之后，我们就可以进行预测了。 - -.. image:: src/PipelineTest_cn.jpg - :align: center - :scale: 80% - -之前配置文件中 ``test.list`` 指定的数据将会被测试，这里直接通过预测脚本 ``predict.sh`` 进行预测, -更详细的说明，请参考 :ref:`api_swig_py_paddle` 。 - - .. code-block:: bash - - model="output/pass-00003" - paddle train \ - --config=trainer_config.lstm.py \ - --use_gpu=false \ - --job=test \ - --init_model_path=$model \ - --config_args=is_predict=1 \ - --predict_output_dir=. \ - - mv rank-00000 result.txt - -这里以 ``output/pass-00003`` 为例进行预测，用户可以根据训练日志，选择测试结果最好的模型来预测。 - -预测结果以文本的形式保存在 ``result.txt`` 中，一行为一个样本，格式如下： - - .. code-block:: bash - - 预测ID;ID为0的概率 ID为1的概率 - 预测ID;ID为0的概率 ID为1的概率 - -总体效果总结 -============== - -在 ``/demo/quick_start`` 目录下，能够找到这里使用的所有数据, 网络配置, 训练脚本等等。 -对于Amazon-Elec测试集(25k), 如下表格，展示了上述网络模型的训练效果: - - ===================== =============================== ============= ================================== - 网络名称 参数数量 错误率 配置文件 - ===================== =============================== ============= ================================== - 逻辑回归模型 252 KB 8.652% trainer_config.lr.py - 词向量模型 15 MB 8.484% trainer_config.emb.py - 卷积模型 16 MB 5.628% trainer_config.cnn.py - 时序模型 16 MB 4.812% trainer_config.lstm.py - ===================== =============================== ============= ================================== - - -附录 -===== - -命令行参数 ----------- - -* \--config：网络配置 -* \--save_dir：模型存储路径 -* \--log_period：每隔多少batch打印一次日志 -* \--num_passes：训练轮次，一个pass表示过一遍所有训练样本 -* \--config_args：命令指定的参数会传入网络配置中。 -* \--init_model_path：指定初始化模型路径，可用在测试或训练时指定初始化模型。 - -默认一个pass保存一次模型，也可以通过saving_period_by_batches设置每隔多少batch保存一次模型。 -可以通过show_parameter_stats_period设置打印参数信息等。 -其他参数请参考 命令行参数文档（链接待补充）。 - -输出日志 ---------- - -.. code-block:: bash - - TrainerInternal.cpp:160] Batch=20 samples=2560 AvgCost=0.628761 CurrentCost=0.628761 Eval: classification_error_evaluator=0.304297 CurrentEval: classification_error_evaluator=0.304297 - -模型训练会看到类似上面这样的日志信息，详细的参数解释，请参考如下表格： - - =========================================== ============================================================== - 名称 解释 - =========================================== ============================================================== - Batch=20 表示过了20个batch - samples=2560 表示过了2560个样本 - AvgCost 每个pass的第0个batch到当前batch所有样本的平均cost - CurrentCost 当前log_period个batch所有样本的平均cost - Eval: classification_error_evaluator 每个pass的第0个batch到当前batch所有样本的平均分类错误率 - CurrentEval: classification_error_evaluator 当前log_period个batch所有样本的平均分类错误率 - =========================================== ============================================================== diff --git a/doc/v1_api_tutorials/quick_start/index_en.md b/doc/v1_api_tutorials/quick_start/index_en.md deleted file mode 100644 index ca110431cf921ae0480d3fb2b17c58f90a84cc0e..0000000000000000000000000000000000000000 --- a/doc/v1_api_tutorials/quick_start/index_en.md +++ /dev/null @@ -1,562 +0,0 @@ -# Quick Start - -This tutorial will teach the basics of deep learning (DL), including how to implement many different models in PaddlePaddle. You will learn how to: - - Prepare data into the standardized format that PaddlePaddle accepts. - - Write data providers that read data into PaddlePaddle. - - Configure neural networks in PaddlePaddle layer by layer. - - Train models. - - Perform inference with trained models. - - -## Install - -To get started, please install PaddlePaddle on your computer. Throughout this tutorial, you will learn by implementing different DL models for text classification. - -To install PaddlePaddle, please follow the instructions here: Build and Install. - -## Overview -For the first step, you will use PaddlePaddle to build a **text classification** system. For example, suppose you run an e-commence website, and you want to analyze the sentiment of user reviews to evaluate product quality. - -For example, given the input - -``` -This monitor is fantastic. -``` - -Your classifier should output “positive”, since this text snippet shows that the user is satisfied with the product. Given this input: - -``` -The monitor breaks down two months after purchase. -``` - -the classifier should output “negative“. - -To build your text classification system, your code will need to perform five steps: -
![](./src/Pipeline_en.jpg)
- - - Preprocess data into a standardized format. - - Provide data to the learning model. - - Specify the neural network structure. - - Train the model. - - Inference (make prediction on test examples). - - -1. Preprocess data into standardized format - - In the text classification example, you will start with a text file with one training example per line. Each line contains category id (in machine learning, often denoted the target y), followed by the input text (often denoted x); these two elements are separated by a Tab. For example: ```positive [tab] This monitor is fantastic```. You will preprocess this raw data into a format that Paddle can use. - -2. Provide data to the learning model. - - You can write data providers in Python. For any required data preprocessing step, you can add the preprocessing code to the PyDataProvider Python file. - - In our text classification example, every word or character will be converted into an integer id, specified in a dictionary file. It perform a dictionary lookup in PyDataProvider to get the id. -3. Specify neural network structure. (From easy to hard, we provide 4 kinds of network configurations) - - A logistic regression model. - - A word embedding model. - - A convolutional neural network model. - - A sequential recurrent neural network model. - - You will also learn different learning algorithms. -4. Training model. -5. Inference. - -## Preprocess data into standardized format -In this example, you are going to use [Amazon electronic product review dataset](http://jmcauley.ucsd.edu/data/amazon/) to build a bunch of deep neural network models for text classification. Each text in this dataset is a product review. This dataset has two categories: “positive” and “negative”. Positive means the reviewer likes the product, while negative means the reviewer does not like the product. - -`demo/quick_start` in the [source code](https://github.com/PaddlePaddle/Paddle) provides script for downloading the preprocessed data as shown below. (If you want to process the raw data, you can use the script `demo/quick_start/data/proc_from_raw_data/get_data.sh`). - -```bash -cd demo/quick_start -./data/get_data.sh -``` - -## Transfer Data to Model -### Write Data Provider with Python -The following `dataprovider_bow.py` gives a complete example of writing data provider with Python. It includes the following parts: - -* initalizer： define the additional meta-data of the data provider and the types of the input data. -* process： Each `yield` returns a data sample. In this case, it return the text representation and category id. The order of features in the returned result needs to be consistent with the definition of the input types in `initalizer`. - -```python -from paddle.trainer.PyDataProvider2 import * - -# id of the word not in dictionary -UNK_IDX = 0 - -# initializer is called by the framework during initialization. -# It allows the user to describe the data types and setup the -# necessary data structure for later use. -# `settings` is an object. initializer need to properly fill settings.input_types. -# initializer can also store other data structures needed to be used at process(). -# In this example, dictionary is stored in settings. -# `dictionay` and `kwargs` are arguments passed from trainer_config.lr.py -def initializer(settings, dictionary, **kwargs): - # Put the word dictionary into settings - settings.word_dict = dictionary - - # setting.input_types specifies what the data types the data provider - # generates. - settings.input_types = [ - # The first input is a sparse_binary_vector, - # which means each dimension of the vector is either 0 or 1. It is the - # bag-of-words (BOW) representation of the texts. - sparse_binary_vector(len(dictionary)), - # The second input is an integer. It represents the category id of the - # sample. 2 means there are two labels in the dataset. - # (1 for positive and 0 for negative) - integer_value(2)] - -# Delaring a data provider. It has an initializer 'data_initialzer'. -# It will cache the generated data of the first pass in memory, so that -# during later pass, no on-the-fly data generation will be needed. -# `setting` is the same object used by initializer() -# `file_name` is the name of a file listed train_list or test_list file given -# to define_py_data_sources2(). See trainer_config.lr.py. -@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM) -def process(settings, file_name): - # Open the input data file. - with open(file_name, 'r') as f: - # Read each line. - for line in f: - # Each line contains the label and text of the comment, separated by \t. - label, comment = line.strip().split('\t') - - # Split the words into a list. - words = comment.split() - - # convert the words into a list of ids by looking them up in word_dict. - word_vector = [settings.word_dict.get(w, UNK_IDX) for w in words] - - # Return the features for the current comment. The first is a list - # of ids representing a 0-1 binary sparse vector of the text, - # the second is the integer id of the label. - yield word_vector, int(label) -``` - -### Define Python Data Provider in Configuration files. -You need to add a data provider definition `define_py_data_sources2` in our network configuration. This definition specifies: - -- The path of the training and testing data (`data/train.list`, `data/test.list`). -- The location of the data provider file (`dataprovider_bow`). -- The function to call to get data. (`process`). -- Additional arguments or data. Here it passes the path of word dictionary. - -```python -from paddle.trainer_config_helpers import * - -file = "data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i -# define the data sources for the model. -# We need to use different process for training and prediction. -# For training, the input data includes both word IDs and labels. -# For prediction, the input data only includs word Ids. -define_py_data_sources2(train_list='data/train.list', - test_list='data/test.list', - module="dataprovider_bow", - obj="process", - args={"dictionary": word_dict}) -``` -You can refer to the following link for more detailed examples and data formats: PyDataProvider2. - -## Network Architecture -We will describe four kinds of network architectures in this section. -
![](./src/PipelineNetwork_en.jpg)
- -First, you will build a logistic regression model. Later, you will also get chance to build other more powerful network architectures. -For more detailed documentation, you could refer to: layer documentation. All configuration files are in `demo/quick_start` directory. - -### Logistic Regression -The architecture is illustrated in the following picture: -
![](./src/NetLR_en.png)
- -- You need define the data for text features. The size of the data layer is the number of words in the dictionary. - -```python -word = data_layer(name="word", size=voc_dim) -``` - -- You also need to define the category id for each example. The size of the data layer is the number of labels. - -```python -label = data_layer(name="label", size=label_dim) -``` - -- It uses logistic regression model to classify the vector, and it will output the classification error during training. - - Each layer has an *input* argument that specifies its input layer. Some layers can have multiple input layers. You can use a list of the input layers as input in that case. - - *size* for each layer means the number of neurons of the layer. - - *act_type* means activation function applied to the output of each neuron independently. - - Some layers can have additional special inputs. For example, `classification_cost` needs ground truth label as input to compute classification loss and error. -```python -# Define a fully connected layer with logistic activation (also called softmax activation). -output = fc_layer(input=word, - size=label_dim, - act_type=SoftmaxActivation()) -# Define cross-entropy classification loss and error. -classification_cost(input=output, label=label) -``` - -Performance summary: You can refer to the training and testing scripts later. In order to compare different network architectures, the model complexity and test classification error are listed in the following table: - - -
- - - - - - - - - - - - - - - - - -
Network nameNumber of parametersTest error
Logistic regression252 KB8.652%
- -
- -### Word Embedding Model -In order to use the word embedding model, you need to change the data provider a little bit to make the input words as a sequence of word IDs. The revised data provider `dataprovider_emb.py` is listed below. You only need to change initializer() for the type of the first input. It is changed from sparse_binary_vector to sequence of intergers. process() remains the same. This data provider can also be used for later sequence models. - -```python -def initializer(settings, dictionary, **kwargs): - # Put the word dictionary into settings - settings.word_dict = dictionary - settings.input_types = [ - # Define the type of the first input as a sequence of integers. - integer_value_sequence(len(dictionary)), - # Define the second input for label id - integer_value(2)] - -@provider(init_hook=initializer) -def process(settings, file_name): - ... - # omitted, it is same as the data provider for LR model -``` - -This model is very similar to the framework of logistic regression, but it uses word embedding vectors instead of a sparse vectors to represent words. -
![](./src/NetContinuous_en.png)
- -- It can look up the dense word embedding vector in the dictionary (its words embedding vector is `word_dim`). The input is a sequence of N words, the output is N word_dim dimensional vectors. - -```python -emb = embedding_layer(input=word, dim=word_dim) -``` - -- It averages all the word embedding in a sentence to get its sentence representation. - -```python -avg = pooling_layer(input=emb, pooling_type=AvgPooling()) -``` - -The other parts of the model are the same as logistic regression network. - -The performance is summarized in the following table: - - -
- - - - - - - - - - - - - - - - - -
Network nameNumber of parametersTest error
Word embedding model15 MB8.484%
-
-
- -### Convolutional Neural Network Model -Convolutional neural network converts a sequence of word embeddings into a sentence representation using temporal convolutions. You will transform the fully connected layer of the word embedding model to 3 new sub-steps. -
![](./src/NetConv_en.png)
- - -Text convolution has 3 steps: -1. Get K nearest neighbor context of each word in a sentence, stack them into a 2D vector representation. -2. Apply temporal convolution to this representation to produce a new hidden_dim dimensional vector. -3. Apply max-pooling to the new vectors at all the time steps in a sentence to get a sentence representation. - -```python -# context_len means convolution kernel size. -# context_start means the start of the convolution. It can be negative. In that case, zero padding is applied. -text_conv = sequence_conv_pool(input=emb, - context_start=k, - context_len=2 * k + 1) -``` - -The performance is summarized in the following table： - - -
- - - - - - - - - - - - - - - - - -
Network nameNumber of parametersTest error
Convolutional model16 MB5.628%
-
- -### Recurrent Model -
![](./src/NetRNN_en.png)
- -You can use Recurrent neural network as our time sequence model, including simple RNN model, GRU model, and LSTM model。 - -- GRU model can be specified via： - -```python -gru = simple_gru(input=emb, size=gru_size) -``` - -- LSTM model can be specified via： - -```python -lstm = simple_lstm(input=emb, size=lstm_size) -``` - -You can use single layer LSTM model with Dropout for our text classification problem. The performance is summarized in the following table: - - -
- - - - - - - - - - - - - - - - - -
Network nameNumber of parametersTest error
Recurrent model16 MB4.812%
- -
- -## Optimization Algorithm -Optimization algorithms include Momentum, RMSProp, AdaDelta, AdaGrad, Adam, and Adamax. You can use Adam optimization method here, with L2 regularization and gradient clipping, because Adam has been proved to work very well for training recurrent neural network. - -```python -settings(batch_size=128, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) -``` - -## Training Model -After completing data preparation and network architecture specification, you will run the training script. -
![](./src/PipelineTrain_en.png)
- -Training script: our training script is in `train.sh` file. The training arguments are listed below: - -```bash -paddle train \ ---config=trainer_config.py \ ---log_period=20 \ ---save_dir=./output \ ---num_passes=15 \ ---use_gpu=false -``` - -We do not provide examples on how to train on clusters here. If you want to train on clusters, please follow the distributed training documentation or other demos for more details. - -## Inference -You can use the trained model to perform prediction on the dataset with no labels. You can also evaluate the model on dataset with labels to obtain its test accuracy. -
![](./src/PipelineTest_en.png)
- -The test script is listed below. PaddlePaddle can evaluate a model on the data with labels specified in `test.list`. - -```bash -paddle train \ ---config=trainer_config.lstm.py \ ---use_gpu=false \ ---job=test \ ---init_model_path=./output/pass-0000x -``` - -We will give an example of performing prediction using Recurrent model on a dataset with no labels. You can refer to Python Prediction API tutorial，or other demo for the prediction process using Python. You can also use the following script for inference or evaluation. - -inference script (predict.sh)： - -```bash -model="output/pass-00003" -paddle train \ - --config=trainer_config.lstm.py \ - --use_gpu=false \ - --job=test \ - --init_model_path=$model \ - --config_args=is_predict=1 \ - --predict_output_dir=. \ - -mv rank-00000 result.txt -``` -User can choose the best model base on the training log instead of model `output/pass-00003`. There are several differences between training and inference network configurations. -- You do not need labels during inference. -- Outputs need to be specified to the classification probability layer (the output of softmax layer), or the id of maximum probability (`max_id` layer). An example to output the id and probability is given in the code snippet. -- batch_size = 1. -- You need to specify the location of `test_list` in the test data. - -The results in `result.txt` is as follows, each line is one sample. - -``` -predicted_label_id;probability_of_label_0 probability_of_label_1 # the first sample -predicted_label_id;probability_of_label_0 probability_of_label_1 # the second sample -``` - - -```python -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -obj = 'process' if not is_predict else 'process_pre' -batch_size = 128 if not is_predict else 1 -if is_predict: - maxid = maxid_layer(output) - outputs([maxid,output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) outputs(cls) -``` - -## Summary -The scripts of data downloading, network configurations, and training scrips are in `/demo/quick_start`. The following table summarizes the performance of our network architecture on Amazon-Elec dataset(25k): - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Network nameNumber of parametersError rateConfiguration file name
Logistic regression model(BOW) 252KB 8.652%trainer_config.lr.py
Word embedding 15MB 8.484%trainer_config.emb.py
Convolution model 16MB 5.628%trainer_config.cnn.py
Time sequence model 16MB 4.812%trainer_config.lstm.py
-
-
- -## Appendix -### Command Line Argument - -* \--config：network architecture path. -* \--save_dir：model save directory. -* \--log_period：the logging period per batch. -* \--num_passes：number of training passes. One pass means the training would go over the whole training dataset once. -* \--config_args：Other configuration arguments. -* \--init_model_path：The path of the initial model parameter. - -By default, the trainer will save model every pass. You can also specify `saving_period_by_batches` to set the frequency of batch saving. You can use `show_parameter_stats_period` to print the statistics of the parameters, which are very useful for tuning parameters. Other command line arguments can be found in command line argument documentation。 - -### Log - -``` -TrainerInternal.cpp:160] Batch=20 samples=2560 AvgCost=0.628761 CurrentCost=0.628761 Eval: classification_error_evaluator=0.304297 CurrentEval: classification_error_evaluator=0.304297 -``` -During model training, you will see the log like the examples above: -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
NameExplanation
Batch=20 You have trained 20 batches.
samples=2560 You have trained 2560 examples.
AvgCost The average cost from the first batch to the current batch.
CurrentCost the average cost of the last log_period batches
Eval: classification_error_evaluator The average classification error from the first batch to the current batch.
CurrentEval: classification_error_evaluator The average error rate of the last log_period batches
-
-
diff --git a/doc/v1_api_tutorials/quick_start/src/NetContinuous_cn.jpg b/doc/v1_api_tutorials/quick_start/src/NetContinuous_cn.jpg deleted file mode 100755 index b18e452a481232f47c453fc30521f8c78ad2da0b..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetContinuous_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetContinuous_en.png b/doc/v1_api_tutorials/quick_start/src/NetContinuous_en.png deleted file mode 100644 index 7bdef1aa366711806585d35c8653c987fd63d59e..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetContinuous_en.png and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetConv_cn.jpg b/doc/v1_api_tutorials/quick_start/src/NetConv_cn.jpg deleted file mode 100755 index 0f5ebfa52ff2abcfc63464d48f00e406d4cbfe86..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetConv_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetConv_en.png b/doc/v1_api_tutorials/quick_start/src/NetConv_en.png deleted file mode 100644 index ad618d1d6f8f4839f566f5f5cb5db37a4b7d9093..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetConv_en.png and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetLR_cn.jpg b/doc/v1_api_tutorials/quick_start/src/NetLR_cn.jpg deleted file mode 100755 index ee65d1f4127a347e991b4de5e5208347b2b0f51f..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetLR_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetLR_en.png b/doc/v1_api_tutorials/quick_start/src/NetLR_en.png deleted file mode 100644 index 9d514bf1b18a0c330f98c28785e5d008f409fc1d..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetLR_en.png and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetRNN_cn.jpg b/doc/v1_api_tutorials/quick_start/src/NetRNN_cn.jpg deleted file mode 100755 index f8bc081827f3a50af80495b4561813c5efe9c9dc..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetRNN_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/NetRNN_en.png b/doc/v1_api_tutorials/quick_start/src/NetRNN_en.png deleted file mode 100644 index 180f273d32ea59dc8ececa69c08e249f79f9d4f7..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/NetRNN_en.png and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/PipelineNetwork_cn.jpg b/doc/v1_api_tutorials/quick_start/src/PipelineNetwork_cn.jpg deleted file mode 100755 index 7e68891d7a536b06027ada6b40cb13dbc3c2d72e..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/PipelineNetwork_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/PipelineNetwork_en.jpg b/doc/v1_api_tutorials/quick_start/src/PipelineNetwork_en.jpg deleted file mode 100644 index e779aed06d5cdb2b442754e7915e79b72946418e..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/PipelineNetwork_en.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/PipelineTest_cn.jpg b/doc/v1_api_tutorials/quick_start/src/PipelineTest_cn.jpg deleted file mode 100755 index 01715db886d811a776c9e7d47f32ba8b1ffe6230..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/PipelineTest_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/PipelineTest_en.png b/doc/v1_api_tutorials/quick_start/src/PipelineTest_en.png deleted file mode 100644 index 7e7ef520b5effa2f43fd2964048f05c42f2ea890..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/PipelineTest_en.png and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/PipelineTrain_cn.jpg b/doc/v1_api_tutorials/quick_start/src/PipelineTrain_cn.jpg deleted file mode 100755 index 8049d3e53c483a4d4956a4840b9a0dce5c2990f1..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/PipelineTrain_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/PipelineTrain_en.png b/doc/v1_api_tutorials/quick_start/src/PipelineTrain_en.png deleted file mode 100644 index 132d29bfd5d678d2518161d0b5ed2e16a233a048..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/PipelineTrain_en.png and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/Pipeline_cn.jpg b/doc/v1_api_tutorials/quick_start/src/Pipeline_cn.jpg deleted file mode 100755 index d5d99253ea528ebd2d8af6cddc6a9f00e20de8cd..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/Pipeline_cn.jpg and /dev/null differ diff --git a/doc/v1_api_tutorials/quick_start/src/Pipeline_en.jpg b/doc/v1_api_tutorials/quick_start/src/Pipeline_en.jpg deleted file mode 100644 index 21a7a7bb6a1af746120e6f4f51f797b6aaafb9d8..0000000000000000000000000000000000000000 Binary files a/doc/v1_api_tutorials/quick_start/src/Pipeline_en.jpg and /dev/null differ diff --git a/go/pserver/client/c/test/test_mnist.py b/go/pserver/client/c/test/test_mnist.py index 7b50a10afc68b87f331c4d0afede9413b3aa2d35..821d9adfcb3164b0982873354badc382b13645ea 100644 --- a/go/pserver/client/c/test/test_mnist.py +++ b/go/pserver/client/c/test/test_mnist.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2 as paddle import gzip diff --git a/go/pserver/client/c/test/test_train.py b/go/pserver/client/c/test/test_train.py index 7ef0fca496e8a3836d1a38b0ff576652d72ce177..445a8d3aa4840b6e206dec813ed6dc04c2a9a21d 100644 --- a/go/pserver/client/c/test/test_train.py +++ b/go/pserver/client/c/test/test_train.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2 as paddle import paddle.v2.dataset.uci_housing as uci_housing import paddle.v2.master as master diff --git a/paddle/CMakeLists.txt b/paddle/CMakeLists.txt index 4a98ede278fad85ff2beef3c8e7dd158912f693a..3f9c132ef6ae03c7614e10484715676c8019821e 100644 --- a/paddle/CMakeLists.txt +++ b/paddle/CMakeLists.txt @@ -18,7 +18,7 @@ else() add_subdirectory(capi) endif() - if(Boost_FOUND) + if(NOT ANDROID AND NOT IOS) add_subdirectory(memory) add_subdirectory(platform) add_subdirectory(framework) diff --git a/paddle/api/test/testTrainConfig.py b/paddle/api/test/testTrainConfig.py index ab9a83e4a35305bcab42a5e05d8f4880f4edd9bb..1a1283e1168650066ad6eca356ae0c1e5bd967ee 100644 --- a/paddle/api/test/testTrainConfig.py +++ b/paddle/api/test/testTrainConfig.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=100, learning_method=AdamOptimizer()) diff --git a/paddle/capi/examples/model_inference/dense/main.c b/paddle/capi/examples/model_inference/dense/main.c index 376cd46fb09a156d426453986c87dcff6e2f71dd..f795bfe11d73dd6a4431b89d33bf143bec89ef76 100644 --- a/paddle/capi/examples/model_inference/dense/main.c +++ b/paddle/capi/examples/model_inference/dense/main.c @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #include #include diff --git a/paddle/capi/examples/model_inference/dense/merge_v2_model.py b/paddle/capi/examples/model_inference/dense/merge_v2_model.py index 760a485a53f5edd31622a0f25a7bf32e6230e9a4..7aeb482903aa9fca1a23f9b68988ee3bd9886cca 100644 --- a/paddle/capi/examples/model_inference/dense/merge_v2_model.py +++ b/paddle/capi/examples/model_inference/dense/merge_v2_model.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.utils.merge_model import merge_v2_model from mnist_v2 import network diff --git a/paddle/capi/examples/model_inference/dense/mnist_v2.py b/paddle/capi/examples/model_inference/dense/mnist_v2.py index 174436bd1d7c1efaf8b25b002ada38b6babeba1c..183eecfdf2cf146d9d919f94dd0fd5416c5ff97e 100644 --- a/paddle/capi/examples/model_inference/dense/mnist_v2.py +++ b/paddle/capi/examples/model_inference/dense/mnist_v2.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import os import sys import gzip diff --git a/paddle/capi/examples/model_inference/dense/trainer_config.py b/paddle/capi/examples/model_inference/dense/trainer_config.py index fbf08903578b309ee163b1f4c70b80133a4324b8..b94a21a7e406b833797f8f521c62a2351c2bc30a 100644 --- a/paddle/capi/examples/model_inference/dense/trainer_config.py +++ b/paddle/capi/examples/model_inference/dense/trainer_config.py @@ -1,31 +1,13 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from paddle.trainer_config_helpers import * - -img = data_layer(name='pixel', size=784) - -hidden = fc_layer( - input=img, - size=200, - param_attr=ParamAttr(name='hidden.w'), - bias_attr=ParamAttr(name='hidden.b')) - -prob = fc_layer( - input=hidden, - size=10, - act=SoftmaxActivation(), - param_attr=ParamAttr(name='prob.w'), - bias_attr=ParamAttr(name='prob.b')) - -outputs(prob) +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/paddle/capi/examples/model_inference/multi_thread/main.c b/paddle/capi/examples/model_inference/multi_thread/main.c index d7675cd80a52f752b1a8567dae34123978113831..eecb9138e76e6e0c0e1065b57fa26e4fa4c7f1d6 100644 --- a/paddle/capi/examples/model_inference/multi_thread/main.c +++ b/paddle/capi/examples/model_inference/multi_thread/main.c @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #include #include #include diff --git a/paddle/capi/examples/model_inference/multi_thread/main_gpu.c b/paddle/capi/examples/model_inference/multi_thread/main_gpu.c index 6fd376e0d1a2fee4f9a0f676b53c6f2891795cab..85bb45658438c7410d29943a401338bd314f383f 100644 --- a/paddle/capi/examples/model_inference/multi_thread/main_gpu.c +++ b/paddle/capi/examples/model_inference/multi_thread/main_gpu.c @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #include #include #include diff --git a/paddle/capi/examples/model_inference/sequence/main.c b/paddle/capi/examples/model_inference/sequence/main.c index 50bc0c9201f207eff7389bfbee3bc2e43261b19a..80937c830d67432ff8cfb125172901bdd4257ffb 100644 --- a/paddle/capi/examples/model_inference/sequence/main.c +++ b/paddle/capi/examples/model_inference/sequence/main.c @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #include #include #include "../common/common.h" diff --git a/paddle/capi/examples/model_inference/sequence/trainer_config.py b/paddle/capi/examples/model_inference/sequence/trainer_config.py index c1326bb95550dc3a7ac6385a81158b5bbd88d5af..889f8acdfd254548ead975c5b8c19b7372ccf3d7 100644 --- a/paddle/capi/examples/model_inference/sequence/trainer_config.py +++ b/paddle/capi/examples/model_inference/sequence/trainer_config.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * WORD_DIM = 3000 diff --git a/paddle/capi/examples/model_inference/sparse_binary/main.c b/paddle/capi/examples/model_inference/sparse_binary/main.c index 029b94ee63ba282aa48193ffd4f625657ddc3a60..efec010a91ab933c8e1ab1715293cdeaaf376327 100644 --- a/paddle/capi/examples/model_inference/sparse_binary/main.c +++ b/paddle/capi/examples/model_inference/sparse_binary/main.c @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #include #include diff --git a/paddle/capi/tests/test_predict_network.py b/paddle/capi/tests/test_predict_network.py index 46a985d47652faa573d1b9817952a2b81d0db8fc..6560417b2a111c3c1b6849468697abeea6e159b6 100644 --- a/paddle/capi/tests/test_predict_network.py +++ b/paddle/capi/tests/test_predict_network.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=100) diff --git a/paddle/framework/CMakeLists.txt b/paddle/framework/CMakeLists.txt index 597ea959f230d88350796cef05b7d6f2a42e594a..8b71f73c36c33d882b34c833031c50cd14817e76 100644 --- a/paddle/framework/CMakeLists.txt +++ b/paddle/framework/CMakeLists.txt @@ -1,7 +1,7 @@ # ddim lib proto_library(framework_proto SRCS framework.proto) -cc_library(ddim SRCS ddim.cc DEPS eigen3) +cc_library(ddim SRCS ddim.cc DEPS eigen3 boost) cc_test(ddim_test SRCS ddim_test.cc DEPS ddim) nv_test(dim_test SRCS dim_test.cu DEPS ddim) @@ -22,24 +22,30 @@ cc_test(eigen_test SRCS eigen_test.cc DEPS tensor) cc_library(lod_tensor SRCS lod_tensor.cc DEPS ddim place tensor framework_proto) cc_test(lod_tensor_test SRCS lod_tensor_test.cc DEPS lod_tensor paddle_memory) -nv_test(lod_tensor_gpu_test SRCS lod_tensor_test.cu DEPS lod_tensor) +nv_test(lod_tensor_gpu_test SRCS lod_tensor_test.cu DEPS lod_tensor init) cc_test(variable_test SRCS variable_test.cc) -cc_library(threadpool SRCS threadpool.cc) +cc_library(threadpool SRCS threadpool.cc DEPS enforce) cc_test(threadpool_test SRCS threadpool_test.cc DEPS threadpool) cc_library(scope SRCS scope.cc DEPS glog threadpool) cc_test(scope_test SRCS scope_test.cc DEPS scope) cc_library(data_device_transform SRCS data_device_transform.cc DEPS tensor) +nv_test(data_device_transform_test SRCS data_device_transform_test.cu + DEPS operator op_registry init math_function) + cc_library(data_type_transform SRCS data_type_transform.cc DEPS tensor) +cc_test(data_type_transform_test SRCS data_type_transform_test.cc DEPS data_type_transform) + cc_library(data_layout_transform SRCS data_layout_transform.cc DEPS tensor math_function) +cc_test(data_layout_transform_test SRCS data_layout_transform_test.cc DEPS data_layout_transform) cc_library(data_transform SRCS data_transform.cc DEPS math_function tensor framework_proto selected_rows data_device_transform data_type_transform data_layout_transform) -cc_library(attribute SRCS attribute.cc DEPS framework_proto) +cc_library(attribute SRCS attribute.cc DEPS framework_proto boost) cc_test(program_desc_test SRCS program_desc_test.cc DEPS proto_desc device_context) cc_library(op_proto_maker SRCS op_proto_maker.cc DEPS framework_proto attribute) @@ -68,7 +74,10 @@ cc_library(backward SRCS backward.cc DEPS net_op) cc_test(backward_test SRCS backward_test.cc DEPS backward recurrent_op device_context fill_constant_op) cc_library(lod_rank_table SRCS lod_rank_table.cc DEPS lod_tensor) -cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto backward glog lod_rank_table) +cc_library(feed_fetch_method SRCS feed_fetch_method.cc DEPS lod_tensor scope glog) + +cc_library(executor SRCS executor.cc DEPS op_registry device_context scope +framework_proto backward glog lod_rank_table profiler feed_fetch_method) cc_library(prune SRCS prune.cc DEPS framework_proto) cc_test(prune_test SRCS prune_test.cc DEPS op_info prune recurrent_op device_context) @@ -82,5 +91,12 @@ cc_test(init_test SRCS init_test.cc DEPS init) cc_test(op_kernel_type_test SRCS op_kernel_type_test.cc DEPS place device_context framework_proto) cc_test(cow_ptr_tests SRCS details/cow_ptr_test.cc) -nv_test(data_device_transform_test SRCS data_device_transform_test.cu - DEPS operator op_registry init math_function) + +if(NOT WITH_C_API AND WITH_FLUID) + file(GLOB FRAMEWORK_HEADERS *.h) + install(FILES ${FRAMEWORK_HEADERS} DESTINATION include/paddle/framework) + install(FILES ${CMAKE_CURRENT_BINARY_DIR}/framework.pb.h DESTINATION include/paddle/framework) + install(FILES details/cow_ptr.h details/op_registry.h DESTINATION include/paddle/framework/details) +endif() + +cc_test(channel_test SRCS channel_test.cc) diff --git a/paddle/framework/attribute.cc b/paddle/framework/attribute.cc index b0fd4d2750eb2529706d871947332d39494505cd..5074e8f5a05ed4e824b3db7e506b30eb1b70c3fd 100644 --- a/paddle/framework/attribute.cc +++ b/paddle/framework/attribute.cc @@ -61,6 +61,9 @@ Attribute GetAttrValue(const proto::OpDesc::Attr& attr_desc) { } return val; } + case proto::AttrType::LONG: { + return attr_desc.l(); + } default: PADDLE_THROW("Unsupport attr type %d", attr_desc.type()); } diff --git a/paddle/framework/attribute.h b/paddle/framework/attribute.h index c1c63d9cb13acb195b3bc3b30088f5fa7daf2a3d..bcff9bc4c48f8f233b7f811640c2789f9618a972 100644 --- a/paddle/framework/attribute.h +++ b/paddle/framework/attribute.h @@ -168,6 +168,32 @@ struct ExtractAttribute { const std::string& attr_name_; }; +template <> +struct ExtractAttribute { + explicit ExtractAttribute(const std::string& attr_name) + : attr_name_(attr_name) {} + + int64_t* operator()(Attribute& attr) const { + if (attr.type() == typeid(int)) { // NOLINT + int val = boost::get(attr); + attr = static_cast(val); + } else if (attr.type() == typeid(float)) { // NOLINT + int val = boost::get(attr); + attr = static_cast(val); + } + int64_t* attr_value = nullptr; + try { + attr_value = &boost::get(attr); + } catch (boost::bad_get& bad_get) { + PADDLE_THROW("Cannot get attribute %s by type int64_t, its type is %s", + attr_name_, attr.type().name()); + } + return attr_value; + } + + const std::string& attr_name_; +}; + // check whether a certain attribute fit its limits // an attribute can have more than one limits template diff --git a/paddle/framework/block_desc.cc b/paddle/framework/block_desc.cc index 54498e175dacfa0a220e3d839f4feb02502b2c03..dd2ed87252102aee6d384f37365d19305f19b281 100644 --- a/paddle/framework/block_desc.cc +++ b/paddle/framework/block_desc.cc @@ -75,7 +75,7 @@ std::vector BlockDesc::AllVars() const { OpDesc *BlockDesc::AppendOp() { need_update_ = true; - ops_.emplace_back(new OpDesc()); + ops_.emplace_back(new OpDesc(this)); return ops_.back().get(); } @@ -86,7 +86,7 @@ void BlockDesc::AppendAllocatedOp(std::unique_ptr &&op_desc) { OpDesc *BlockDesc::PrependOp() { need_update_ = true; - ops_.emplace_front(new OpDesc()); + ops_.emplace_front(new OpDesc(this)); return ops_.front().get(); } @@ -153,7 +153,7 @@ BlockDesc::BlockDesc(ProgramDesc *prog, proto::BlockDesc *desc) vars_[var_desc.name()].reset(new VarDesc(var_desc)); } for (const proto::OpDesc &op_desc : desc_->ops()) { - ops_.emplace_back(new OpDesc(op_desc, prog)); + ops_.emplace_back(new OpDesc(op_desc, prog, this)); } } @@ -162,7 +162,7 @@ BlockDesc::BlockDesc(const BlockDesc &other, proto::BlockDesc *desc, : prog_(prog), desc_(desc) { need_update_ = true; for (auto &op : other.ops_) { - ops_.emplace_back(new OpDesc(*op)); + ops_.emplace_back(new OpDesc(*op, this)); } for (auto &it : other.vars_) { diff --git a/paddle/framework/channel.h b/paddle/framework/channel.h new file mode 100644 index 0000000000000000000000000000000000000000..0570980c5a4d7fa45e672ae5baac65d2c65ddad9 --- /dev/null +++ b/paddle/framework/channel.h @@ -0,0 +1,58 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include // for size_t + +namespace paddle { +namespace framework { + +// Channel is the abstract class of buffered and un-buffered channels. +template +class Channel { + public: + virtual void Send(T*) = 0; + virtual void Receive(T*) = 0; + virtual size_t Cap() = 0; + virtual void Close() = 0; + virtual ~Channel() {} +}; + +// Forward declaration of channel implementations. +namespace details { +template +class Buffered; +template +class UnBuffered; +} // namespace details + +template +Channel* MakeChannel(size_t buffer_size) { + if (buffer_size > 0) { + return new details::Buffered(buffer_size); + } + return new details::UnBuffered(); +} + +template +void CloseChannel(Channel* ch) { + ch->Close(); +} + +} // namespace framework +} // namespace paddle + +#include "paddle/framework/details/buffered_channel.h" +#include "paddle/framework/details/unbuffered_channel.h" diff --git a/paddle/framework/channel_test.cc b/paddle/framework/channel_test.cc new file mode 100644 index 0000000000000000000000000000000000000000..1510fb8abf54f05804bd404d9bd00ecc42fbef63 --- /dev/null +++ b/paddle/framework/channel_test.cc @@ -0,0 +1,80 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/framework/channel.h" + +#include +#include + +#include "gtest/gtest.h" + +using paddle::framework::Channel; +using paddle::framework::MakeChannel; +using paddle::framework::CloseChannel; + +TEST(Channel, MakeAndClose) { + using paddle::framework::details::Buffered; + using paddle::framework::details::UnBuffered; + { + // MakeChannel should return a buffered channel is buffer_size > 0. + auto ch = MakeChannel(10); + EXPECT_NE(dynamic_cast*>(ch), nullptr); + EXPECT_EQ(dynamic_cast*>(ch), nullptr); + CloseChannel(ch); + delete ch; + } + { + // MakeChannel should return an un-buffered channel is buffer_size = 0. + auto ch = MakeChannel(0); + EXPECT_EQ(dynamic_cast*>(ch), nullptr); + EXPECT_NE(dynamic_cast*>(ch), nullptr); + CloseChannel(ch); + delete ch; + } +} + +TEST(Channel, SufficientBufferSizeDoesntBlock) { + const size_t buffer_size = 10; + auto ch = MakeChannel(buffer_size); + for (size_t i = 0; i < buffer_size; ++i) { + ch->Send(&i); // should not block + } + + size_t out; + for (size_t i = 0; i < buffer_size; ++i) { + ch->Receive(&out); // should not block + EXPECT_EQ(out, i); + } + CloseChannel(ch); + delete ch; +} + +TEST(Channel, ConcurrentSendNonConcurrentReceiveWithSufficientBufferSize) { + const size_t buffer_size = 10; + auto ch = MakeChannel(buffer_size); + size_t sum = 0; + std::thread t([&]() { + // Try to write more than buffer size. + for (size_t i = 0; i < 2 * buffer_size; ++i) { + ch->Send(&i); // should not block + sum += i; + } + }); + std::this_thread::sleep_for(std::chrono::milliseconds(100)); // wait 0.5 sec + EXPECT_EQ(sum, 45U); + + CloseChannel(ch); + t.join(); + delete ch; +} diff --git a/paddle/framework/data_device_transform.cc b/paddle/framework/data_device_transform.cc index d38d87927fc7ee0aa32eabff5cf83d7c4ca7b2b0..5daf5a4e0ab47cf90119ee2f48f1fe89559a1972 100644 --- a/paddle/framework/data_device_transform.cc +++ b/paddle/framework/data_device_transform.cc @@ -31,7 +31,7 @@ static const platform::DeviceContext* GetDeviceContext( } } -void DeviceTransform(const Tensor& in, const platform::Place& dst_place, +void TransDataDevice(const Tensor& in, const platform::Place& dst_place, Tensor* out) { VLOG(3) << "DeviceTransform in, src_place " << in.place() << " dst_place: " << dst_place; diff --git a/paddle/framework/data_device_transform.h b/paddle/framework/data_device_transform.h index b21ed0be34ad868ae4a404e913e9623db308b530..39750a85f2787c6d35723bb908011349a771fb7f 100644 --- a/paddle/framework/data_device_transform.h +++ b/paddle/framework/data_device_transform.h @@ -21,7 +21,7 @@ limitations under the License. */ namespace paddle { namespace framework { -void DeviceTransform(const Tensor& in, const platform::Place& dst_place, +void TransDataDevice(const Tensor& in, const platform::Place& dst_place, Tensor* out); } // namespace framework diff --git a/paddle/framework/data_device_transform_test.cu b/paddle/framework/data_device_transform_test.cu index 5d89f5546fa87241dec6364d86a100ca51bce687..efc05b3106b40bdaa6cd03ce707c677dd58b0730 100644 --- a/paddle/framework/data_device_transform_test.cu +++ b/paddle/framework/data_device_transform_test.cu @@ -150,6 +150,7 @@ TEST(Operator, CPUtoGPU) { // get output auto* output2 = scope.Var("OUT2"); gpu_op->Run(scope, cuda_place); + VLOG(3) << "after gpu_op run"; // auto* output2_ptr = output2->Get().data(); DeviceContextPool& pool = DeviceContextPool::Instance(); diff --git a/paddle/framework/data_layout_transform.cc b/paddle/framework/data_layout_transform.cc index 96794cae97d460e86fe83ac1395e1dfc7e371e3b..9d0a6d5ea3eb4127763acbd1f7a219aa19db4eca 100644 --- a/paddle/framework/data_layout_transform.cc +++ b/paddle/framework/data_layout_transform.cc @@ -1,25 +1,36 @@ -/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. #include "paddle/framework/data_layout_transform.h" -#include "paddle/framework/tensor.h" #include "paddle/operators/math/math_function.h" namespace paddle { namespace framework { +std::vector GetAxis(const DataLayout& from, const DataLayout& to) { + PADDLE_ENFORCE_NE(from, to, + "layout transform should transform different layout"); + if (from == DataLayout::kNCHW && to == DataLayout::kNHWC) { + return {0, 2, 3, 1}; + } else if (from == DataLayout::kNHWC && to == DataLayout::kNCHW) { + return {0, 3, 1, 2}; + } else { + PADDLE_THROW("unsupported transform"); + } +} + struct CastDataLayout { CastDataLayout(const platform::DeviceContext* ctx, const std::vector& axis, const framework::Tensor& in, @@ -44,38 +55,36 @@ struct CastDataLayout { } }; -void TransDataLayout(const std::vector& axis, - const platform::DeviceContext* ctx, - const KernelTypePair& kernel_pair, const Variable& in, - Variable* out) { - PADDLE_ENFORCE(in.IsType(), "Only support Tensor transform!."); +void TransDataLayout(const OpKernelType& kernel_type_for_var, + const OpKernelType& expected_kernel_type, const Tensor& in, + Tensor* out) { PADDLE_ENFORCE( - platform::places_are_same_class(kernel_pair.first.place_, - kernel_pair.second.place_), + platform::places_are_same_class(kernel_type_for_var.place_, + expected_kernel_type.place_), "TransDataLayout only support DataLayout transform on same place!"); - PADDLE_ENFORCE(kernel_pair.first.data_type_ == kernel_pair.second.data_type_, - "TransDataLayout only support Datatype are same!"); - auto src = in.Get(); - auto* dst = out->GetMutable(); - PADDLE_ENFORCE(arity(src.dims()) == 4, "Input Arity Only Suppport 4!"); + PADDLE_ENFORCE(arity(in.dims()) == 4, "Input Arity only support 4!"); + + auto& pool = platform::DeviceContextPool::Instance(); - auto src_dim = src.dims(); + auto src_dim = in.dims(); std::vector dst_dim; + auto axis = GetAxis(kernel_type_for_var.data_layout_, + expected_kernel_type.data_layout_); dst_dim.resize(axis.size()); for (size_t i = 0; i < axis.size(); i++) { dst_dim[i] = src_dim[axis[i]]; } - dst->Resize(make_ddim(dst_dim)); - auto place = kernel_pair.second.place_; - dst->mutable_data(place, src.type()); + out->Resize(make_ddim(dst_dim)); + out->mutable_data(expected_kernel_type.place_, in.type()); - auto src_type = kernel_pair.first.data_type_; - framework::VisitDataType(src_type, CastDataLayout(ctx, axis, src, dst)); + framework::VisitDataType( + framework::ToDataType(in.type()), + CastDataLayout(pool.Get(expected_kernel_type.place_), axis, in, out)); - dst->set_layout(kernel_pair.second.data_layout_); + out->set_layout(expected_kernel_type.data_layout_); } } // namespace framework diff --git a/paddle/framework/data_layout_transform.h b/paddle/framework/data_layout_transform.h index befae1f63616a4c21d998c6b784b8ef288d00617..368f7fc9898338af0f9502cbc1e94cc40ae12e3b 100644 --- a/paddle/framework/data_layout_transform.h +++ b/paddle/framework/data_layout_transform.h @@ -1,31 +1,31 @@ -/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. #pragma once #include "paddle/framework/op_kernel_type.h" +#include "paddle/framework/tensor.h" #include "paddle/framework/variable.h" namespace paddle { namespace framework { -using KernelTypePair = std::pair; +std::vector GetAxis(const DataLayout& from, const DataLayout& to); -void TransDataLayout(const std::vector& axis, - const platform::DeviceContext* ctx, - const KernelTypePair& kernel_pair, const Variable& in, - Variable* out); +void TransDataLayout(const OpKernelType& kernel_type_for_var, + const OpKernelType& expected_kernel_type, const Tensor& in, + Tensor* out); } // namespace framework } // namespace paddle diff --git a/paddle/framework/data_layout_transform_test.cc b/paddle/framework/data_layout_transform_test.cc new file mode 100644 index 0000000000000000000000000000000000000000..093e8d4d3458479763415dbc24957eaaa0f6f0fe --- /dev/null +++ b/paddle/framework/data_layout_transform_test.cc @@ -0,0 +1,44 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "paddle/framework/data_layout_transform.h" + +#include "gtest/gtest.h" +#include "paddle/platform/device_context.h" + +TEST(DataTransform, DataLayoutFunction) { + using namespace paddle::framework; + using namespace paddle::platform; + + auto place = CPUPlace(); + Tensor in = Tensor(); + Tensor out = Tensor(); + in.mutable_data(make_ddim({2, 3, 1, 2}), place); + in.set_layout(DataLayout::kNHWC); + + auto kernel_nhwc = OpKernelType(proto::DataType::FP32, place, + DataLayout::kNHWC, LibraryType::kPlain); + auto kernel_ncwh = OpKernelType(proto::DataType::FP32, place, + DataLayout::kNCHW, LibraryType::kPlain); + + TransDataLayout(kernel_nhwc, kernel_ncwh, in, &out); + + EXPECT_TRUE(out.layout() == DataLayout::kNCHW); + EXPECT_TRUE(out.dims() == make_ddim({2, 2, 3, 1})); + + TransDataLayout(kernel_ncwh, kernel_nhwc, in, &out); + + EXPECT_TRUE(in.layout() == DataLayout::kNHWC); + EXPECT_TRUE(in.dims() == make_ddim({2, 3, 1, 2})); +} \ No newline at end of file diff --git a/paddle/framework/data_transform.cc b/paddle/framework/data_transform.cc index d826f0edace6d5afee5cd83f6e65d6dbaefae874..b6fd46401ffdd50cfdb00cc6e4ecd821bb39aba5 100644 --- a/paddle/framework/data_transform.cc +++ b/paddle/framework/data_transform.cc @@ -15,18 +15,50 @@ limitations under the License. */ #include "paddle/framework/data_transform.h" #include "paddle/framework/data_device_transform.h" +#include "paddle/framework/data_layout_transform.h" +#include "paddle/framework/data_type_transform.h" namespace paddle { namespace framework { +static void PassTensorData(Tensor* from, Tensor* to) { + to->ShareDataWith(*from); + *from = Tensor(); +} + void DataTransform(const OpKernelType& expected_kernel_type, const OpKernelType& kernel_type_for_var, - const Tensor& input_tensor, Tensor* out) { + const Tensor& input_tensor, Tensor* output_tensor) { + bool transformed = false; + Tensor in; + in.ShareDataWith(input_tensor); + Tensor out; + + // do layout transform + if (NeedTransformLayout(expected_kernel_type.data_layout_, + kernel_type_for_var.data_layout_)) { + TransDataLayout(kernel_type_for_var, expected_kernel_type, in, &out); + transformed = true; + PassTensorData(&out, &in); + } + + if (expected_kernel_type.data_type_ != kernel_type_for_var.data_type_) { + TransDataType(kernel_type_for_var, expected_kernel_type, in, &out); + transformed = true; + PassTensorData(&out, &in); + } + + // do device transform if (!platform::is_same_place(kernel_type_for_var.place_, expected_kernel_type.place_)) { - DeviceTransform(input_tensor, expected_kernel_type.place_, out); + TransDataDevice(in, expected_kernel_type.place_, &out); + transformed = true; + PassTensorData(&out, &in); } - PADDLE_ENFORCE_NOT_NULL(out, "out should not be null"); + + PADDLE_ENFORCE(transformed, "No transform is applied, please check!"); + // get output data + output_tensor->ShareDataWith(in); } void CopyVariableWithTensor(const Variable& in_var, const Tensor& tensor, diff --git a/paddle/framework/data_type.h b/paddle/framework/data_type.h index 6a372ac32e48131eed28e2d42125feb5b92a11c7..98eb3e857d1943e71f1d41f24ecbedbe09e85b7b 100644 --- a/paddle/framework/data_type.h +++ b/paddle/framework/data_type.h @@ -79,5 +79,33 @@ inline void VisitDataType(proto::DataType type, Visitor visitor) { } } +inline std::string DataTypeToString(const proto::DataType type) { + using namespace paddle::framework::proto; + switch (type) { + case DataType::FP16: + return "float16"; + case DataType::FP32: + return "float32"; + case DataType::FP64: + return "float64"; + case DataType::INT16: + return "int16"; + case DataType::INT32: + return "int32"; + case DataType::INT64: + return "int64"; + case DataType::BOOL: + return "bool"; + default: + PADDLE_THROW("Not support type %d", type); + } +} + +inline std::ostream& operator<<(std::ostream& out, + const proto::DataType& type) { + out << DataTypeToString(type); + return out; +} + } // namespace framework } // namespace paddle diff --git a/paddle/framework/data_type_transform.cc b/paddle/framework/data_type_transform.cc index 63373232e910d44eb0996f9280f9c166ad092030..7df1cc6b75b4e87b57813aa01a1b29302d616158 100644 --- a/paddle/framework/data_type_transform.cc +++ b/paddle/framework/data_type_transform.cc @@ -38,14 +38,11 @@ struct CastDataType { template void operator()() { - auto place = ctx_->GetPlace(); - auto* in_begin = in_.data(); - auto numel = in_.numel(); - auto* in_end = in_begin + numel; - auto* out_begin = out_->mutable_data(place); + auto* in_end = in_begin + in_.numel(); + auto* out_begin = out_->mutable_data(in_.place()); - if (platform::is_cpu_place(place)) { + if (platform::is_cpu_place(in_.place())) { platform::Transform trans; auto* context = static_cast(ctx_); trans(*context, in_begin, in_end, out_begin, @@ -57,38 +54,31 @@ struct CastDataType { } }; -void TransDataType(const platform::DeviceContext* ctx, - const KernelTypePair& kernel_pair, const Variable& in, - Variable* out) { - PADDLE_ENFORCE(in.IsType(), "Only Support Tensor transform!."); - PADDLE_ENFORCE( - platform::places_are_same_class(kernel_pair.first.place_, - kernel_pair.second.place_), - "TransDataType Only Support DataType transform on same place!"); - - auto src = in.Get(); - auto* dst = out->GetMutable(); +void TransDataType(const OpKernelType& kernel_type_for_var, + const OpKernelType& expected_kernel_type, const Tensor& in, + Tensor* out) { + platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance(); - auto dims = src.dims(); - dst->Resize(dims); - auto dst_type = kernel_pair.second.data_type_; - auto src_type = kernel_pair.first.data_type_; + out->Resize(in.dims()); + auto src_type = kernel_type_for_var.data_type_; + auto dst_type = expected_kernel_type.data_type_; + auto ctx = pool.Get(in.place()); switch (src_type) { case proto::DataType::FP32: - framework::VisitDataType(dst_type, CastDataType(src, dst, ctx)); + framework::VisitDataType(dst_type, CastDataType(in, out, ctx)); break; case proto::DataType::FP64: - framework::VisitDataType(dst_type, CastDataType(src, dst, ctx)); + framework::VisitDataType(dst_type, CastDataType(in, out, ctx)); break; case proto::DataType::INT32: - framework::VisitDataType(dst_type, CastDataType(src, dst, ctx)); + framework::VisitDataType(dst_type, CastDataType(in, out, ctx)); break; case proto::DataType::INT64: - framework::VisitDataType(dst_type, CastDataType(src, dst, ctx)); + framework::VisitDataType(dst_type, CastDataType(in, out, ctx)); break; case proto::DataType::BOOL: - framework::VisitDataType(dst_type, CastDataType(src, dst, ctx)); + framework::VisitDataType(dst_type, CastDataType(in, out, ctx)); break; default: PADDLE_THROW("Not support type %d", src_type); diff --git a/paddle/framework/data_type_transform.h b/paddle/framework/data_type_transform.h index 8ec90742256c2308a242d993838e46e51a6fc167..067c0c2a5b14465a31cabf7b5d4442d6a3a1c773 100644 --- a/paddle/framework/data_type_transform.h +++ b/paddle/framework/data_type_transform.h @@ -15,6 +15,7 @@ limitations under the License. */ #pragma once #include "paddle/framework/op_kernel_type.h" +#include "paddle/framework/tensor.h" #include "paddle/framework/variable.h" #include "paddle/platform/device_context.h" @@ -23,9 +24,9 @@ namespace framework { using KernelTypePair = std::pair; -void TransDataType(const platform::DeviceContext* ctx, - const KernelTypePair& kernel_pair, const Variable& in, - Variable* out); +void TransDataType(const OpKernelType& kernel_type_for_var, + const OpKernelType& expected_kernel_type, const Tensor& in, + Tensor* out); } // namespace framework } // namespace paddle diff --git a/paddle/framework/data_type_transform_test.cc b/paddle/framework/data_type_transform_test.cc new file mode 100644 index 0000000000000000000000000000000000000000..89d32f528334a159e1b015802ff2670f9cfc1584 --- /dev/null +++ b/paddle/framework/data_type_transform_test.cc @@ -0,0 +1,53 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/framework/data_type_transform.h" + +#include "gtest/gtest.h" + +TEST(DataTypeTransform, CPUTransform) { + using namespace paddle::framework; + using namespace paddle::platform; + + auto place = CPUPlace(); + + Tensor in; + Tensor out; + + float* ptr = in.mutable_data(make_ddim({2, 3}), place); + int data_number = 2 * 3; + + for (int i = 0; i < data_number; ++i) { + ptr[i] = i / 3; + } + + auto kernel_fp32 = OpKernelType(proto::DataType::FP32, place, + DataLayout::kAnyLayout, LibraryType::kPlain); + auto kernel_fp64 = OpKernelType(proto::DataType::FP64, place, + DataLayout::kAnyLayout, LibraryType::kPlain); + auto kernel_int32 = OpKernelType(proto::DataType::INT32, place, + DataLayout::kAnyLayout, LibraryType::kPlain); + + TransDataType(kernel_fp32, kernel_fp64, in, &out); + double* out_data_double = out.data(); + for (int i = 0; i < data_number; ++i) { + ASSERT_EQ(out_data_double[i], static_cast(i / 3)); + } + + TransDataType(kernel_fp32, kernel_int32, in, &out); + int* out_data_int = out.data(); + for (int i = 0; i < data_number; ++i) { + ASSERT_EQ(out_data_int[i], static_cast(i / 3)); + } +} diff --git a/paddle/framework/details/buffered_channel.h b/paddle/framework/details/buffered_channel.h new file mode 100644 index 0000000000000000000000000000000000000000..b093e1589293b030ef2bedb82504a8e86b3dc857 --- /dev/null +++ b/paddle/framework/details/buffered_channel.h @@ -0,0 +1,102 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include +#include +#include + +#include "paddle/framework/channel.h" +#include "paddle/platform/enforce.h" + +namespace paddle { +namespace framework { +namespace details { + +template +class Buffered : public paddle::framework::Channel { + friend Channel* paddle::framework::MakeChannel(size_t); + friend void paddle::framework::CloseChannel(Channel*); + + public: + virtual void Send(T*); + virtual void Receive(T*); + virtual size_t Cap() { return cap_; } + virtual void Close(); + virtual ~Buffered(); + + private: + size_t cap_; + std::mutex mu_; + std::condition_variable empty_cond_var_; + std::condition_variable full_cond_var_; + std::deque channel_; + bool closed_; + + Buffered(size_t cap) : cap_(cap), closed_(false) { + PADDLE_ENFORCE_GT(cap, 0); + } + + void NotifyAllSenders(std::unique_lock*); +}; + +template +void Buffered::Send(T* item) { + std::unique_lock lock(mu_); + full_cond_var_.wait(lock, + [this]() { return channel_.size() < cap_ || closed_; }); + if (!closed_) { + channel_.push_back(std::move(*item)); + lock.unlock(); + empty_cond_var_.notify_one(); + } +} + +template +void Buffered::Receive(T* item) { + std::unique_lock lock(mu_); + empty_cond_var_.wait(lock, [this]() { return !channel_.empty() || closed_; }); + if (!closed_) { + *item = std::move(channel_.front()); + channel_.pop_front(); + NotifyAllSenders(&lock); + } else { + item = nullptr; + } +} + +template +void Buffered::Close() { + std::unique_lock lock(mu_); + closed_ = true; + NotifyAllSenders(&lock); +} + +template +Buffered::~Buffered() { + std::unique_lock lock(mu_); + closed_ = true; + channel_.clear(); + NotifyAllSenders(&lock); +} + +template +void Buffered::NotifyAllSenders(std::unique_lock* lock) { + lock->unlock(); + full_cond_var_.notify_all(); +} + +} // namespace details +} // namespace framework +} // namespace paddle diff --git a/paddle/framework/details/unbuffered_channel.h b/paddle/framework/details/unbuffered_channel.h new file mode 100644 index 0000000000000000000000000000000000000000..cc2d2e587eca981307d4e522bd569fbffa450207 --- /dev/null +++ b/paddle/framework/details/unbuffered_channel.h @@ -0,0 +1,56 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include +#include +#include + +#include "paddle/framework/channel.h" + +namespace paddle { +namespace framework { +namespace details { + +template +class UnBuffered : public paddle::framework::Channel { + friend Channel* paddle::framework::MakeChannel(size_t); + friend void paddle::framework::CloseChannel(Channel*); + + public: + virtual void Send(T*); + virtual void Receive(T*); + virtual size_t Cap() { return 0; } + virtual void Close(); + virtual ~UnBuffered(); + + private: + UnBuffered() {} +}; + +template +void UnBuffered::Send(T* channel_element) {} + +template +void UnBuffered::Receive(T*) {} + +template +void UnBuffered::Close() {} + +template +UnBuffered::~UnBuffered() {} + +} // namespace details +} // namespace framework +} // namespace paddle diff --git a/paddle/framework/eigen_test.cc b/paddle/framework/eigen_test.cc index c6f77ecfabdddbfdd1df646a1c9310c03930bd8e..9e368a522ce71d73213ba4c781e8a56fad917b0a 100644 --- a/paddle/framework/eigen_test.cc +++ b/paddle/framework/eigen_test.cc @@ -11,18 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "paddle/framework/eigen.h" #include diff --git a/paddle/framework/executor.cc b/paddle/framework/executor.cc index c0418c9266e257bd7567861543e557f354451b17..9a232b08434d299d10bb2acdb6e96295de875d56 100644 --- a/paddle/framework/executor.cc +++ b/paddle/framework/executor.cc @@ -17,12 +17,15 @@ limitations under the License. */ #include #include "gflags/gflags.h" +#include "paddle/framework/feed_fetch_method.h" #include "paddle/framework/feed_fetch_type.h" #include "paddle/framework/lod_rank_table.h" #include "paddle/framework/lod_tensor_array.h" #include "paddle/framework/op_registry.h" #include "paddle/platform/place.h" +#include "paddle/platform/profiler.h" +DECLARE_bool(benchmark); DEFINE_bool(check_nan_inf, false, "Checking whether operator produce NAN/INF or not. It will be " "extremely slow so please use this flag wisely."); @@ -30,9 +33,6 @@ DEFINE_bool(check_nan_inf, false, namespace paddle { namespace framework { -const std::string kFeedOpType = "feed"; -const std::string kFetchOpType = "fetch"; - Executor::Executor(const platform::Place& place) : place_(place) {} static void CreateTensor(Variable* var, proto::VarDesc::VarType var_type) { @@ -115,8 +115,17 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id, for (auto& op_desc : block.AllOps()) { auto op = paddle::framework::OpRegistry::CreateOp(*op_desc); - VLOG(3) << op->DebugStringEx(local_scope); + VLOG(4) << op->DebugStringEx(local_scope); + + platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance(); + platform::RecordEvent record_event(op->Type(), pool.Get(place_)); + op->Run(*local_scope, place_); + VLOG(3) << op->DebugStringEx(local_scope); + if (FLAGS_benchmark) { + VLOG(2) << "Memory used after operator " + op->Type() + " running: " + << memory::memory_usage(place_); + } if (FLAGS_check_nan_inf) { for (auto& vname : op->OutputVars(true)) { auto* var = local_scope->FindVar(vname); @@ -130,6 +139,171 @@ void Executor::Run(const ProgramDesc& pdesc, Scope* scope, int block_id, if (create_vars && create_local_scope) { scope->DeleteScope(local_scope); } + if (FLAGS_benchmark) { + VLOG(2) << "-------------------------------------------------------"; + VLOG(2) << "Memory used after deleting local scope: " + << memory::memory_usage(place_); + VLOG(2) << "-------------------------------------------------------"; + } +} + +// Check whether the block already has feed operators and feed_holder. +// Return false if the block does not have any feed operators. +// If some feed operators have been prepended to the block, check that +// the info contained in these feed operators matches the feed_targets +// and feed_holder_name. Raise exception when any mismatch is found. +// Return true if the block has feed operators and holder of matching info. +static bool has_feed_operators( + BlockDesc* block, std::map& feed_targets, + const std::string& feed_holder_name) { + size_t feed_count = 0; + for (auto* op : block->AllOps()) { + if (op->Type() == kFeedOpType) { + feed_count++; + PADDLE_ENFORCE_EQ(op->Input("X")[0], feed_holder_name, + "Input to feed op should be '%s'", feed_holder_name); + std::string feed_target_name = op->Output("Out")[0]; + PADDLE_ENFORCE( + feed_targets.find(feed_target_name) != feed_targets.end(), + "Feed operator output name '%s' cannot be found in 'feed_targets'", + feed_target_name); + } + } + + if (feed_count > 0) { + PADDLE_ENFORCE_EQ( + feed_count, feed_targets.size(), + "The number of feed operators should match 'feed_targets'"); + + // When feed operator are present, so should be feed_holder + auto var = block->FindVar(feed_holder_name); + PADDLE_ENFORCE_NOT_NULL(var, "Block should already have a '%s' variable", + feed_holder_name); + PADDLE_ENFORCE_EQ(var->GetType(), proto::VarDesc::FEED_MINIBATCH, + "'%s' variable should be 'FEED_MINIBATCH' type", + feed_holder_name); + } + + return feed_count > 0; +} + +// Check whether the block already has fetch operators and fetch_holder. +// Return false if the block does not have any fetch operators. +// If some fetch operators have been appended to the block, check that +// the info contained in these fetch operators matches the fetch_targets +// and fetch_holder_name. Raise exception when any mismatch is found. +// Return true if the block has fetch operators and holder of matching info. +static bool has_fetch_operators( + BlockDesc* block, std::map& fetch_targets, + const std::string& fetch_holder_name) { + size_t fetch_count = 0; + for (auto* op : block->AllOps()) { + if (op->Type() == kFetchOpType) { + fetch_count++; + PADDLE_ENFORCE_EQ(op->Output("Out")[0], fetch_holder_name, + "Output of fetch op should be '%s'", fetch_holder_name); + std::string fetch_target_name = op->Input("X")[0]; + PADDLE_ENFORCE( + fetch_targets.find(fetch_target_name) != fetch_targets.end(), + "Fetch operator input name '%s' cannot be found in 'fetch_targets'", + fetch_target_name); + } + } + + if (fetch_count > 0) { + PADDLE_ENFORCE_EQ( + fetch_count, fetch_targets.size(), + "The number of fetch operators should match 'fetch_targets'"); + + // When fetch operator are present, so should be fetch_holder + auto var = block->FindVar(fetch_holder_name); + PADDLE_ENFORCE_NOT_NULL(var, "Block should already have a '%s' variable", + fetch_holder_name); + PADDLE_ENFORCE_EQ(var->GetType(), proto::VarDesc::FETCH_LIST, + "'%s' variable should be 'FETCH_LIST' type", + fetch_holder_name); + } + + return fetch_count > 0; +} + +void Executor::Run(const ProgramDesc& program, Scope* scope, + std::map& feed_targets, + std::map& fetch_targets, + const std::string& feed_holder_name, + const std::string& fetch_holder_name) { + auto* copy_program = new ProgramDesc(program); + auto* global_block = copy_program->MutableBlock(0); + + if (!has_feed_operators(global_block, feed_targets, feed_holder_name)) { + // create feed_holder variable + auto* feed_holder = global_block->Var(feed_holder_name); + feed_holder->SetType(proto::VarDesc::FEED_MINIBATCH); + feed_holder->SetPersistable(true); + + int i = 0; + for (auto& feed_target : feed_targets) { + std::string var_name = feed_target.first; + VLOG(3) << "feed target's name: " << var_name; + + // prepend feed op + auto* op = global_block->PrependOp(); + op->SetType(kFeedOpType); + op->SetInput("X", {feed_holder_name}); + op->SetOutput("Out", {var_name}); + op->SetAttr("col", {static_cast(i)}); + op->CheckAttrs(); + + i++; + } + } + + // map the data of feed_targets to feed_holder + for (auto* op : global_block->AllOps()) { + if (op->Type() == kFeedOpType) { + std::string feed_target_name = op->Output("Out")[0]; + int idx = boost::get(op->GetAttr("col")); + SetFeedVariable(scope, *feed_targets[feed_target_name], feed_holder_name, + idx); + } + } + + if (!has_fetch_operators(global_block, fetch_targets, fetch_holder_name)) { + // create fetch_holder variable + auto* fetch_holder = global_block->Var(fetch_holder_name); + fetch_holder->SetType(proto::VarDesc::FETCH_LIST); + fetch_holder->SetPersistable(true); + + int i = 0; + for (auto& fetch_target : fetch_targets) { + std::string var_name = fetch_target.first; + VLOG(3) << "fetch target's name: " << var_name; + + // append fetch op + auto* op = global_block->AppendOp(); + op->SetType(kFetchOpType); + op->SetInput("X", {var_name}); + op->SetOutput("Out", {fetch_holder_name}); + op->SetAttr("col", {static_cast(i)}); + op->CheckAttrs(); + + i++; + } + } + + Run(*copy_program, scope, 0, true, true); + + // obtain the data of fetch_targets from fetch_holder + for (auto* op : global_block->AllOps()) { + if (op->Type() == kFetchOpType) { + std::string fetch_target_name = op->Input("X")[0]; + int idx = boost::get(op->GetAttr("col")); + *fetch_targets[fetch_target_name] = + GetFetchVariable(*scope, fetch_holder_name, idx); + } + } + + delete copy_program; } } // namespace framework diff --git a/paddle/framework/executor.h b/paddle/framework/executor.h index d869e18901b82959a40cc296aa0844c20ea63ac1..035ff48a52bd2fc4b1a46b48b1fbf1fbcb2ac70b 100644 --- a/paddle/framework/executor.h +++ b/paddle/framework/executor.h @@ -41,6 +41,12 @@ class Executor { void Run(const ProgramDesc&, Scope*, int, bool create_local_scope = true, bool create_vars = true); + void Run(const ProgramDesc& program, Scope* scope, + std::map& feed_targets, + std::map& fetch_targets, + const std::string& feed_holder_name = "feed", + const std::string& fetch_holder_name = "fetch"); + private: const platform::Place place_; }; diff --git a/paddle/framework/feed_fetch_method.cc b/paddle/framework/feed_fetch_method.cc new file mode 100644 index 0000000000000000000000000000000000000000..21201b675519e34b11e9f1f3a6f2a135c06d63a7 --- /dev/null +++ b/paddle/framework/feed_fetch_method.cc @@ -0,0 +1,56 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/framework/feed_fetch_method.h" +#include "glog/logging.h" +#include "paddle/framework/variable.h" + +namespace paddle { +namespace framework { + +void SetFeedVariable(Scope* scope, const LoDTensor& input, + const std::string& var_name, size_t index) { + // If var_name Variable is not found in GlobalScope, a new variable will + // be created. + VLOG(3) << "SetFeedVariable name=" << var_name << " index=" << index; + Variable* g_feed_value = scope->Var(var_name); + auto& feed_inputs = + *(g_feed_value->GetMutable>()); + if (index >= feed_inputs.size()) { + feed_inputs.resize(index + 1); + } + // shared data with input tensor + feed_inputs[index].ShareDataWith(input); + // set lod + feed_inputs[index].set_lod(input.lod()); +} + +LoDTensor& GetFetchVariable(const Scope& scope, const std::string& var_name, + size_t index) { + // Since we want to fetch LodTensor from a variable, the variable must + // be created alreadly. + Variable* g_fetch_value = scope.FindVar(var_name); + PADDLE_ENFORCE(g_fetch_value->IsType(), + "Only %s can be invoked by GetFetchVariable", + typeid(FeedFetchList).name()); + auto& fetch_outputs = *g_fetch_value->GetMutable(); + auto& tensor = fetch_outputs[index]; + VLOG(3) << "Fetch " << var_name << " with index " << index + << " shape= " << tensor.dims(); + PADDLE_ENFORCE_LT(index, fetch_outputs.size()); + return tensor; +} + +} // namespace framework +} // namespace paddle diff --git a/paddle/framework/feed_fetch_method.h b/paddle/framework/feed_fetch_method.h index 7feacb1e24708411e7fbb610f9909447cba9e291..b71945fcc8834d2e5fe21151e1e88788b4acd5c1 100644 --- a/paddle/framework/feed_fetch_method.h +++ b/paddle/framework/feed_fetch_method.h @@ -13,46 +13,18 @@ See the License for the specific language governing permissions and limitations under the License. */ #pragma once -#include "glog/logging.h" + #include "paddle/framework/feed_fetch_type.h" #include "paddle/framework/scope.h" -#include "paddle/framework/variable.h" namespace paddle { namespace framework { void SetFeedVariable(Scope* scope, const LoDTensor& input, - const std::string& var_name, size_t index) { - // If var_name Variable is not found in GlobalScope, a new variable will - // be created. - VLOG(3) << "SetFeedVariable name=" << var_name << " index=" << index; - Variable* g_feed_value = scope->Var(var_name); - auto& feed_inputs = - *(g_feed_value->GetMutable>()); - if (index >= feed_inputs.size()) { - feed_inputs.resize(index + 1); - } - // shared data with input tensor - feed_inputs[index].ShareDataWith(input); - // set lod - feed_inputs[index].set_lod(input.lod()); -} + const std::string& var_name, size_t index); LoDTensor& GetFetchVariable(const Scope& scope, const std::string& var_name, - size_t index) { - // Since we want to fetch LodTensor from a variable, the variable must - // be created alreadly. - Variable* g_fetch_value = scope.FindVar(var_name); - PADDLE_ENFORCE(g_fetch_value->IsType(), - "Only %s can be invoked by GetFetchVariable", - typeid(FeedFetchList).name()); - auto& fetch_outputs = *g_fetch_value->GetMutable(); - auto& tensor = fetch_outputs[index]; - VLOG(3) << "Fetch " << var_name << " with index " << index - << " shape= " << tensor.dims(); - PADDLE_ENFORCE_LT(index, fetch_outputs.size()); - return tensor; -} + size_t index); } // namespace framework } // namespace paddle diff --git a/paddle/framework/feed_fetch_type.h b/paddle/framework/feed_fetch_type.h index 9bc4a90c44828ecb7458d524f59609f01848cc5c..168f456675af508df86dd0520cdeb5d16d94ad31 100644 --- a/paddle/framework/feed_fetch_type.h +++ b/paddle/framework/feed_fetch_type.h @@ -13,6 +13,7 @@ See the License for the specific language governing permissions and limitations under the License. */ #pragma once +#include #include #include "paddle/framework/lod_tensor.h" @@ -20,5 +21,8 @@ namespace paddle { namespace framework { using FeedFetchType = LoDTensor; using FeedFetchList = std::vector; + +static const std::string kFeedOpType = "feed"; +static const std::string kFetchOpType = "fetch"; } // namespace framework } // namespace paddle diff --git a/paddle/framework/framework.proto b/paddle/framework/framework.proto index ea69b87e2ac7dc587333b623c310182bb39eb452..5b6ef03f610926578d2c02dcf06f399f106a30a1 100644 --- a/paddle/framework/framework.proto +++ b/paddle/framework/framework.proto @@ -26,6 +26,7 @@ enum AttrType { BOOLEAN = 6; BOOLEANS = 7; BLOCK = 8; + LONG = 9; } // OpDesc describes an instance of a C++ framework::OperatorBase @@ -44,6 +45,7 @@ message OpDesc { optional bool b = 10; repeated bool bools = 11; optional int32 block_idx = 12; + optional int64 l = 13; }; message Var { diff --git a/paddle/framework/init.cc b/paddle/framework/init.cc index 4ef82a541efaa35bcf831d5122570154f2fa2423..3f6ea121b3994979d89a7d5a8c20c59240a0c111 100644 --- a/paddle/framework/init.cc +++ b/paddle/framework/init.cc @@ -13,6 +13,7 @@ See the License for the specific language governing permissions and limitations under the License. */ #include // for strdup #include +#include #include #include "paddle/framework/init.h" @@ -46,17 +47,23 @@ void InitDevices() { std::vector places; places.emplace_back(platform::CPUPlace()); + int count = 0; #ifdef PADDLE_WITH_CUDA - int count = platform::GetCUDADeviceCount(); - for (int i = 0; i < count; ++i) { - places.emplace_back(platform::CUDAPlace(i)); + try { + count = platform::GetCUDADeviceCount(); + } catch (const std::exception &exp) { + LOG(WARNING) << "Compiled with WITH_GPU, but no GPU found in runtime."; } #else LOG(WARNING) - << "'GPU' is not supported, Please re-compile with WITH_GPU option"; + << "'CUDA' is not supported, Please re-compile with WITH_GPU option"; #endif + for (int i = 0; i < count; ++i) { + places.emplace_back(platform::CUDAPlace(i)); + } + platform::DeviceContextPool::Init(places); } diff --git a/paddle/framework/init_test.cc b/paddle/framework/init_test.cc index f837a965d3be7d40c20803ae4462b3bfd91bffd0..01e076dd8ea24831e3ed7c8a7f8fae6818a89335 100644 --- a/paddle/framework/init_test.cc +++ b/paddle/framework/init_test.cc @@ -20,7 +20,21 @@ TEST(InitDevices, CPU) { using paddle::framework::InitDevices; using paddle::platform::DeviceContextPool; +#ifndef PADDLE_WITH_CUDA InitDevices(); DeviceContextPool& pool = DeviceContextPool::Instance(); - ASSERT_GE(pool.size(), 1U); + ASSERT_EQ(pool.size(), 1U); +#endif +} + +TEST(InitDevices, CUDA) { + using paddle::framework::InitDevices; + using paddle::platform::DeviceContextPool; + +#ifdef PADDLE_WITH_CUDA + int count = paddle::platform::GetCUDADeviceCount(); + InitDevices(); + DeviceContextPool& pool = DeviceContextPool::Instance(); + ASSERT_EQ(pool.size(), 1U + static_cast(count)); +#endif } diff --git a/paddle/framework/lod_tensor.cc b/paddle/framework/lod_tensor.cc index 87a57d095141cc456af2cbabbc227715a02375e9..cb27de6991674247e6215ce64a2da5000fa78ed4 100644 --- a/paddle/framework/lod_tensor.cc +++ b/paddle/framework/lod_tensor.cc @@ -24,8 +24,6 @@ limitations under the License. */ #include #include -#include - namespace paddle { namespace framework { @@ -107,9 +105,10 @@ LoD ToAbsOffset(const LoD &in) { // the lowest level stores relative offsets if (in.empty() || in.size() == 1) return in; LoD result = in; - for (int level = result.size() - 2; level >= 0; level--) { - for (auto &ele : result[level]) { - ele = result[level + 1][ele]; + for (auto level = static_cast(in.size() - 2); level >= 0; level--) { + for (size_t i = 0; i < in[level].size(); ++i) { + size_t index = in[level][i]; + result[level][i] = result[level + 1][index]; } } return result; @@ -135,6 +134,65 @@ bool operator==(const LoD &a, const LoD &b) { return true; } +bool CheckLoD(const LoD &in, int tensor_height) { + if (in.empty()) return true; + for (const auto &level : in) { + // check: there should be more than 2 offsets existing in each level. + if (level.size() < 2) return false; + // check: the first offset(the begin offset) of each level should be 0. + if (level.front() != 0) return false; + // check: all the offsets in a level should be ascending(no same items + // allows). + if (!std::is_sorted(level.begin(), level.begin(), [](size_t a, size_t b) { + if (a < b) return true; + return false; + })) { + LOG(INFO) << "ascending error"; + return false; + } + } + // check: the lowest level's last offset should equals `tensor_height` if + // tensor_height>0. + if (tensor_height > 0 && (size_t)tensor_height != in.back().back()) + return false; + + // check: the higher level's last offset should equals the lower level's + // size-1. + // NOTE LoD store the levels from top to bottom, so the higher level goes + // first. + for (size_t level = 0; level < in.size() - 1; level++) { + if (in[level].back() != in[level + 1].size() - 1) return false; + } + return true; +} + +bool CheckAbsLoD(const LoD &in, int tensor_height) { + if (in.empty()) return true; + for (const auto &level : in) { + // check: all the offsets in a level should be ascending(no same items + // allows). + if (!std::is_sorted(level.begin(), level.begin(), [](size_t a, size_t b) { + if (a < b) return true; + return false; + })) { + return false; + } + + // check: there should be more than 2 offsets existing in each level. + if (level.size() < 2) return false; + + // check: the first offset of each level should be 0, and the last should be + // the same(the height of underlying tensor). + if (level.front() != 0) return false; + if (tensor_height < 0) { + tensor_height = level.back(); + } else if ((size_t)tensor_height != level.back()) { + return false; + } + } + return true; +} + using LoDAndOffset = std::pair>; LoDAndOffset GetSubLoDAndAbsoluteOffset(const LoD &lod, size_t start_idx, size_t end_idx, size_t start_level) { @@ -227,48 +285,86 @@ void DeserializeFromStream(std::istream &is, LoDTensor *tensor, DeserializeFromStream(is, static_cast(tensor), dev_ctx); } -// TODO(tonyyang-svail): make this function support LoD std::vector LoDTensor::SplitLoDTensor( const std::vector places) const { check_memory_size(); - PADDLE_ENFORCE(lod().empty(), "Disable parallel lod for now"); - PADDLE_ENFORCE(dims()[0] % places.size() == 0, - "Batch size should be divided by places size"); - - std::vector lods; - for (size_t place_idx = 0; place_idx < places.size(); ++place_idx) { - int begin = place_idx * dims()[0] / places.size(); - int end = (place_idx + 1) * dims()[0] / places.size(); + int batch_size = + lod().empty() ? dims()[0] : static_cast(lod()[0].size()) - 1; + size_t result_size = std::min(static_cast(batch_size), places.size()); + size_t remainder = batch_size % places.size(); + + std::vector results; + results.reserve(result_size); + + int step_width = static_cast(batch_size / result_size); + for (size_t i = 0; i < result_size; ++i) { + int begin = static_cast(i * step_width); + int end = static_cast((i + 1) * step_width); + if (i + 1 == places.size()) { // last + end += remainder; + } - auto src = Slice(begin, end); - auto &dst_place = places[place_idx]; LoDTensor dst; - framework::Copy(src, dst_place, &dst); - - lods.emplace_back(dst); + if (lod().empty()) { + auto src = Slice(begin, end); + auto &dst_place = places[i]; + framework::Copy(src, dst_place, &dst); + } else { + auto lod_and_offset = GetSubLoDAndAbsoluteOffset(lod(), begin, end, 0); + + auto &offset = lod_and_offset.second; + auto src = Slice(offset.first, offset.second); + auto &dst_place = places[i]; + framework::Copy(src, dst_place, &dst); + + LoD my_lod; + for (auto &l : lod_and_offset.first) { + std::vector v{0}; + for (auto &ll : l) { + v.push_back(ll + v.back()); + } + my_lod.emplace_back(v); + } + dst.set_lod(my_lod); + } + results.emplace_back(dst); } - return lods; + return results; } -// TODO(tonyyang-svail): make this function support LoD void LoDTensor::MergeLoDTensor( const std::vector &lod_tensors, platform::Place dst_place) { PADDLE_ENFORCE(!lod_tensors.empty()); + framework::DDim new_dim = lod_tensors[0]->dims(); std::type_index new_type = lod_tensors[0]->type(); - auto new_layout = lod_tensors[0]->layout(); - for (auto *lod : lod_tensors) { - PADDLE_ENFORCE(new_dim == lod->dims()); - PADDLE_ENFORCE(new_type == lod->type()); - PADDLE_ENFORCE(new_layout == lod->layout()); + framework::DataLayout new_layout = lod_tensors[0]->layout(); + LoD new_lod = lod_tensors[0]->lod(); + for (size_t i = 1; i < lod_tensors.size(); ++i) { + auto *t = lod_tensors[i]; + PADDLE_ENFORCE_EQ(new_type.hash_code(), t->type().hash_code()); + PADDLE_ENFORCE_EQ(new_layout, t->layout()); + + PADDLE_ENFORCE_EQ(framework::product(new_dim) / new_dim[0], + framework::product(t->dims()) / t->dims()[0]); + new_dim[0] += t->dims()[0]; + + auto &lod = t->lod(); + for (size_t j = 0; j < lod.size(); ++j) { + auto &sub_lod = new_lod[j]; + auto &offset = sub_lod.back(); + for (size_t k = 1; k < lod[j].size(); ++k) { + sub_lod.push_back(lod[j][k] + offset); + } + } } - new_dim[0] *= lod_tensors.size(); Resize(new_dim); set_layout(new_layout); - + set_lod(new_lod); mutable_data(dst_place, new_type); + int begin = 0; for (auto *src : lod_tensors) { int end = begin + src->dims()[0]; diff --git a/paddle/framework/lod_tensor.h b/paddle/framework/lod_tensor.h index 88ea78f2682b2ffc962c9663f6b3c636dedb931d..d0ab640485baf6d76ee629ea420b603f42b031b4 100644 --- a/paddle/framework/lod_tensor.h +++ b/paddle/framework/lod_tensor.h @@ -18,11 +18,11 @@ limitations under the License. */ #ifdef PADDLE_WITH_CUDA #include #include -#include #endif #include #include "paddle/framework/ddim.h" +#include "paddle/framework/mixed_vector.h" #include "paddle/framework/tensor.h" #include "paddle/framework/tensor_util.h" #include "paddle/platform/enforce.h" @@ -31,15 +31,6 @@ limitations under the License. */ namespace paddle { namespace framework { -#ifndef PADDLE_WITH_CUDA -template -using Vector = std::vector; -#else -template -using Vector = thrust::host_vector< - T, thrust::system::cuda::experimental::pinned_allocator>; -#endif - /* * LoD is short for Level of Details. * @@ -55,7 +46,15 @@ using Vector = thrust::host_vector< * 0 2 4 7 * 0 2 5 7 10 12 15 20 */ -using LoD = std::vector>; +struct LoD : public std::vector> { + using std::vector>::vector; + + void CopyFromCUDA() { + for (auto it = this->begin(); it != this->end(); ++it) { + it->CopyFromCUDA(); + } + } +}; std::ostream& operator<<(std::ostream& os, const LoD& lod); std::ostream& operator<<(std::ostream& os, const LoDTensor& t); @@ -71,13 +70,48 @@ LoD ToAbsOffset(const LoD& in); bool operator==(const LoD& a, const LoD& b); +/* + * Check whether this lod's format is valid. + * + * ATTENTION: + * - Empty lod is treated as valid. + * + * It will check two things: + * + * 1. all the offsets in a level should be ascending(no same items allows). + * 2. there should be more than 2 offsets existing in each level. + * 3. the higher level's last offset should equals the lower level's size-1. + * 4. the first offset(the begin offset) of each level should be 0. + * 5. the lowest level's last offset should equals `tensor_height` if + * tensor_height>0. + */ + +bool CheckLoD(const LoD& in, int tensor_height = -1); +/* + * Check whether this absolute lod's format is valid. + * + * ATTENTION: + * - Empty lod is treated as valid. + * + * It will check two things: + * 1. all the offsets in a level should be ascending(no same items allows) + * 2. there should be more than 2 offsets existing in each level. + * 3. the first offset of each level should be 0, and the last should be the + * same(the height of underlying tensor) or `tensor_height` if + * tensor_height>0. + */ +bool CheckAbsLoD(const LoD& in, int tensor_height = -1); + /* * LoDTensor (Level of details Tensor) * see https://en.wikipedia.org/wiki/Level_of_details for reference. */ class LoDTensor : public Tensor { public: - LoDTensor() {} + LoDTensor() : Tensor() {} + + /* Constructor with place should only be used in pybind */ + explicit LoDTensor(const platform::Place& place) : Tensor(place) {} explicit LoDTensor(const LoD& lod) : lod_(lod) {} diff --git a/paddle/framework/lod_tensor_test.cc b/paddle/framework/lod_tensor_test.cc index 19ae7815cc1008f9f468821d61b26f221e266d5e..3b63020e685436396071fa05cd7697630ae56c95 100644 --- a/paddle/framework/lod_tensor_test.cc +++ b/paddle/framework/lod_tensor_test.cc @@ -1,28 +1,16 @@ -// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. // You may obtain a copy of the License at // -// http://www.apache.org/licenses/LICENSE-2.0 +// http://www.apache.org/licenses/LICENSE-2.0 // // Unless required by applicable law or agreed to in writing, software // distributed under the License is distributed on an "AS IS" BASIS, // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "paddle/framework/lod_tensor.h" @@ -35,37 +23,16 @@ namespace paddle { namespace framework { -const int kLodTensorSize = 20 * 128; - -class LoDTensorTester : public ::testing::Test { - public: - virtual void SetUp() override { - // tensor's batch_size: 30 - // 3 levels - // 0 10 20 - // 0 5 10 15 20 - // 0 2 5 7 10 12 15 20 - LoD lod; - lod.push_back(std::vector{0, 2, 3}); - lod.push_back(std::vector{0, 2, 5, 8}); - lod.push_back(std::vector{0, 2, 5, 7, 10, 12, 15, 17, 20}); - - ASSERT_EQ(lod.size(), 3UL); - - lod_tensor_.Resize({20 /*batch size*/, 128 /*dim*/}); - // malloc memory - float* dst_ptr = lod_tensor_.mutable_data(place); - for (int i = 0; i < kLodTensorSize; ++i) { - dst_ptr[i] = i; - } - - lod_tensor_.set_lod(lod); - } +TEST(LoD, data) { + LoD lod{{0, 1, 2}}; + lod.push_back({0, 2, 4, 5}); + lod.push_back(std::vector({0, 1, 6, 8, 10, 11})); - protected: - platform::CPUPlace place; - LoDTensor lod_tensor_; -}; + auto& v = lod[0]; + for (size_t i = 0; i < v.size(); ++i) { + EXPECT_EQ(v[i], i); + } +} TEST(LodExpand, test) { LoD lod{{0, 2}}; @@ -144,5 +111,118 @@ TEST(LoD, ToAbsOffset) { EXPECT_EQ(abs_lod, expected); } +TEST(LoD, SplitLoDTensor) { + LoD lod; + lod.push_back(std::vector({0, 2, 4, 5, 6})); + lod.push_back(std::vector({0, 1, 6, 8, 13, 15, 20})); + + platform::CPUPlace place; + LoDTensor lod_tensor; + lod_tensor.Resize({20, 1}); + float* dst_ptr = lod_tensor.mutable_data(place); + for (int i = 0; i < lod_tensor.numel(); ++i) { + dst_ptr[i] = i; + } + lod_tensor.set_lod(lod); + + std::vector places{platform::CPUPlace(), + platform::CPUPlace()}; + LoD lod0; + lod0.push_back(std::vector({0, 2, 4})); + lod0.push_back(std::vector({0, 1, 6, 8, 13})); + LoD lod1; + lod1.push_back(std::vector({0, 1, 2})); + lod1.push_back(std::vector({0, 2, 7})); + + auto lods = lod_tensor.SplitLoDTensor(places); + EXPECT_EQ(lods[0].lod(), lod0); + EXPECT_EQ(lods[1].lod(), lod1); +} + +TEST(LoD, MergeLoDTensor) { + LoD lod; + lod.push_back(std::vector({0, 2, 4, 5, 6})); + lod.push_back(std::vector({0, 1, 6, 8, 13, 15, 20})); + + platform::CPUPlace place; + + LoDTensor lod_tensor0; + LoD lod0; + lod0.push_back(std::vector({0, 2, 4})); + lod0.push_back(std::vector({0, 1, 6, 8, 13})); + lod_tensor0.set_lod(lod0); + + lod_tensor0.Resize({13, 1}); + float* dst_ptr = lod_tensor0.mutable_data(place); + for (int i = 0; i < lod_tensor0.numel(); ++i) { + dst_ptr[i] = i; + } + + LoDTensor lod_tensor1; + LoD lod1; + lod1.push_back(std::vector({0, 1, 2})); + lod1.push_back(std::vector({0, 2, 7})); + lod_tensor1.set_lod(lod1); + lod_tensor1.Resize({7, 1}); + dst_ptr = lod_tensor1.mutable_data(place); + for (int i = 0; i < lod_tensor1.numel(); ++i) { + dst_ptr[i] = i; + } + + std::vector lods{&lod_tensor0, &lod_tensor1}; + + LoDTensor lod_tensor; + lod_tensor.MergeLoDTensor(lods, place); + EXPECT_EQ(lod_tensor.lod(), lod); +} + +TEST(LoD, CheckLoD) { + LoD relative_lod; + relative_lod.push_back(std::vector({0, 2})); + relative_lod.push_back(std::vector({0, 1, 3})); + relative_lod.push_back(std::vector({0, 2, 4, 5})); + + // check compatible + ASSERT_TRUE(CheckLoD(relative_lod)); + relative_lod[1].back()++; + ASSERT_FALSE(CheckLoD(relative_lod)); + relative_lod[1].back()--; // recover it + + // check empty + LoD empty_lod; + ASSERT_TRUE(CheckLoD(empty_lod)); + + // check less than 2 offsets in a level + LoD some_lod0; + some_lod0.push_back(std::vector({0})); + ASSERT_FALSE(CheckLoD(some_lod0)); + + // check with underlying tensor storage. + ASSERT_TRUE(CheckLoD(relative_lod, 5)); + ASSERT_FALSE(CheckLoD(relative_lod, 9)); +} + +TEST(LoD, CheckAbsLoD) { + LoD relative_lod; + relative_lod.push_back(std::vector({0, 2})); + relative_lod.push_back(std::vector({0, 1, 3})); + relative_lod.push_back(std::vector({0, 2, 4, 5})); + + auto abs_lod = ToAbsOffset(relative_lod); + + ASSERT_TRUE(CheckAbsLoD(abs_lod)); + + // check less than 2 offsets in a level. + + // check the last item should be compatible with tensor height. + abs_lod.back().back()++; + ASSERT_FALSE(CheckAbsLoD(abs_lod)); + abs_lod.back().back()--; // restore + + // check less than 2 offsets in a lod. + LoD abs_lod0; + abs_lod0.push_back(std::vector({0})); + ASSERT_FALSE(CheckAbsLoD(abs_lod0)); +} } // namespace framework } // namespace paddle diff --git a/paddle/framework/lod_tensor_test.cu b/paddle/framework/lod_tensor_test.cu index 0f46e9b1e3966a49ff0673231c65f608797c80fe..d4c9f00bd9c00f3cae68858ca46c5320fc117405 100644 --- a/paddle/framework/lod_tensor_test.cu +++ b/paddle/framework/lod_tensor_test.cu @@ -1,31 +1,21 @@ -// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. // // Licensed under the Apache License, Version 2.0 (the "License"); // you may not use this file except in compliance with the License. // You may obtain a copy of the License at // -// http://www.apache.org/licenses/LICENSE-2.0 +// http://www.apache.org/licenses/LICENSE-2.0 // // Unless required by applicable law or agreed to in writing, software // distributed under the License is distributed on an "AS IS" BASIS, // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include #include +#include +#include "paddle/framework/init.h" #include "paddle/framework/lod_tensor.h" #include "paddle/platform/assert.h" @@ -38,7 +28,48 @@ __global__ void test(size_t* a, int size) { } } +TEST(Vector, Normal) { + using namespace paddle::framework; + using namespace paddle::platform; + using namespace paddle::memory; + + paddle::framework::InitDevices(); + + paddle::framework::Vector vec({1, 2, 3}); + size_t* ptr = vec.data(); + for (size_t i = 0; i < vec.size(); ++i) { + EXPECT_EQ(vec[i], *(ptr + i)); + } + + vec.clear(); + vec.CopyFromCUDA(); + + std::vector v = {1, 2, 3}; + for (size_t i = 0; i < v.size(); ++i) { + EXPECT_EQ(v[i], vec[i]); + } +} + +TEST(LoD, data) { + paddle::framework::InitDevices(); + + paddle::framework::LoD lod{{0, 1, 2}}; + lod.push_back({0, 2, 4, 5}); + lod.push_back(std::vector({0, 1, 6, 8, 10, 11})); + + auto& v = lod[0]; + test<<<1, 1>>>(v.cuda_data(), v.size()); + cudaDeviceSynchronize(); + + v.CopyFromCUDA(); + for (size_t i = 0; i < v.size(); ++i) { + EXPECT_EQ(v[i], i * 2); + } +} + TEST(LoDTensor, LoDInGPU) { + paddle::framework::InitDevices(); + paddle::framework::LoDTensor lod_tensor; paddle::platform::CUDAPlace place(0); @@ -54,8 +85,9 @@ TEST(LoDTensor, LoDInGPU) { auto lod = lod_tensor.lod(); - test<<<1, 8>>>(lod[0].data(), lod[0].size()); + test<<<1, 8>>>(lod[0].cuda_data(), lod[0].size()); cudaDeviceSynchronize(); + lod.CopyFromCUDA(); for (size_t i = 0; i < src_lod[0].size(); ++i) { EXPECT_EQ(lod[0].data()[i], src_lod[0].data()[i] * 2); diff --git a/paddle/framework/mixed_vector.h b/paddle/framework/mixed_vector.h new file mode 100644 index 0000000000000000000000000000000000000000..0e0e23958602343f8e0106e3a88eaac9c6d71066 --- /dev/null +++ b/paddle/framework/mixed_vector.h @@ -0,0 +1,154 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. */ + +#pragma once + +#include +#include + +#include "paddle/memory/memcpy.h" +#include "paddle/memory/memory.h" +#include "paddle/platform/device_context.h" +#include "paddle/platform/enforce.h" +#include "paddle/platform/place.h" + +namespace paddle { +namespace framework { + +/** + * @brief Vector support both cpu and gpu. + * host vector lifetime is same with Vector + * device vector is lazily malloc and modified. + */ + +template +class Vector : public std::vector { + public: + /* NOTE(dzhwinter): + * Data always store and modified on Host. + * If the data is modified when use cuda_data interface, + * You need to call the CopyFromCUDA explicitly to synchronize data. + * + */ + enum class kDataPosition { + kDataOnHost = 0, + kDataOnDevice = 1, + }; + + public: + using std::vector::vector; + + Vector() {} + Vector(const std::vector &v) : std::vector(v) {} // NOLINT + + virtual ~Vector() { +#ifdef PADDLE_WITH_CUDA + if (cuda_ptr_ != nullptr) { + memory::Free(place_, static_cast(cuda_ptr_)); + } +#endif + } + + T *cuda_data() { + CopyToCUDA(); + PADDLE_ENFORCE_NOT_NULL( + cuda_ptr_, "No data or Insufficient CUDA memory to allocation"); + return static_cast(cuda_ptr_); + } + + T *data() { return std::vector::data(); } + + const T *data() const { return std::vector::data(); } + + void CopyToCUDA(); + + void CopyFromCUDA(); + + void CopyToPeer(platform::Place); + + private: + void *cuda_ptr_ = nullptr; + size_t cuda_size_ = 0; + /*The DataPosition is unused now, + if we want support random access from cpu and cuda, + we need to overload all the vector method */ + + kDataPosition position_ = kDataPosition::kDataOnHost; + platform::CUDAPlace place_; +}; + +template +void Vector::CopyToCUDA() { +#ifdef PADDLE_WITH_CUDA + if (cuda_ptr_ == nullptr) { + cuda_ptr_ = + memory::Alloc(place_, this->size() * sizeof(T)); + } + platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); + auto *cuda_ctx = pool.GetByPlace(place_); + + memory::Copy(place_, static_cast(cuda_ptr_), platform::CPUPlace(), + static_cast(this->data()), + this->size() * sizeof(T), cuda_ctx->stream()); + cuda_ctx->Wait(); + + cuda_size_ = this->size(); +#endif +} + +template +void Vector::CopyFromCUDA() { +#ifdef PADDLE_WITH_CUDA + platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); + auto *cuda_ctx = pool.GetByPlace(place_); + if (cuda_ptr_ == nullptr) { + LOG(WARNING) << "No uncommited cuda data."; + return; + } + this->resize(cuda_size_); + memory::Copy(platform::CPUPlace(), static_cast(this->data()), place_, + static_cast(cuda_ptr_), this->size() * sizeof(T), + cuda_ctx->stream()); + cuda_ctx->Wait(); + +#endif +} + +template +void Vector::CopyToPeer(platform::Place peer_place) { + if (platform::is_cpu_place(peer_place)) { + return; + } +#ifdef PADDLE_WITH_CUDA + auto *cuda_ctx = platform::DeviceContextPool::Instance().GetByPlace(place_); + void *peer_cuda_ptr_ = memory::Alloc( + boost::get(peer_place), this->size() * sizeof(T)); + memory::Copy(boost::get(peer_place), + static_cast(peer_cuda_ptr_), place_, + static_cast(cuda_ptr_), this->size() * sizeof(T), + cuda_ctx->stream()); + cuda_ctx->Wait(); + memory::Free(place_, static_cast(cuda_ptr_)); + place_ = boost::get(peer_place); + cuda_ptr_ = peer_cuda_ptr_; +#endif +} + +template class Vector; +template class Vector; +template class Vector; +template class Vector; + +} // namespace framework +} // namespace paddle diff --git a/paddle/framework/op_desc.cc b/paddle/framework/op_desc.cc index 1c0372bb16c04e155a68a0411939e4887322107a..f8df2cf97ad532f06cb1393b1a24cd789f8bde29 100644 --- a/paddle/framework/op_desc.cc +++ b/paddle/framework/op_desc.cc @@ -97,7 +97,7 @@ void OpDesc::CopyFrom(const OpDesc &op_desc) { need_update_ = true; } -OpDesc::OpDesc(const proto::OpDesc &desc, ProgramDesc *prog) +OpDesc::OpDesc(const proto::OpDesc &desc, ProgramDesc *prog, BlockDesc *block) : desc_(desc), need_update_(false) { // restore inputs_ int input_size = desc_.inputs_size(); @@ -131,6 +131,7 @@ OpDesc::OpDesc(const proto::OpDesc &desc, ProgramDesc *prog) attrs_[attr_name] = prog->MutableBlock(bid); } } + this->block_ = block; } proto::OpDesc *OpDesc::Proto() { @@ -282,6 +283,7 @@ struct SetAttrDescVisitor : public boost::static_visitor { VectorToRepeated(v, attr_->mutable_bools()); } void operator()(BlockDesc *desc) const { attr_->set_block_idx(desc->ID()); } + void operator()(int64_t v) const { attr_->set_l(v); } void operator()(boost::blank) const { PADDLE_THROW("Unexpected branch"); } }; diff --git a/paddle/framework/op_desc.h b/paddle/framework/op_desc.h index a5ffb162928bfd355d35d3f9b63aab59a88dd061..13695cff59f0bfd79c48eb28670ecc67a0309332 100644 --- a/paddle/framework/op_desc.h +++ b/paddle/framework/op_desc.h @@ -25,7 +25,6 @@ namespace framework { class BlockDesc; class ProgramDesc; - class OpDesc { public: OpDesc() {} @@ -33,7 +32,14 @@ class OpDesc { OpDesc(const std::string &type, const VariableNameMap &inputs, const VariableNameMap &outputs, const AttributeMap &attrs); - OpDesc(const proto::OpDesc &desc, ProgramDesc *prog); + OpDesc(const proto::OpDesc &desc, ProgramDesc *prog, BlockDesc *block); + + explicit OpDesc(BlockDesc *block) : block_(block) {} + + OpDesc(const OpDesc &other, BlockDesc *block) { + *this = other; + block_ = block; + } void CopyFrom(const OpDesc &op_desc); @@ -117,6 +123,10 @@ class OpDesc { void Flush(); + BlockDesc *Block() { return this->block_; } + + void SetBlock(BlockDesc *block) { this->block_ = block; } + private: template static std::vector MapKeys(const MapType &map) { @@ -129,6 +139,7 @@ class OpDesc { } proto::OpDesc desc_; + BlockDesc *block_; // not_own // input arg name => input variable names VariableNameMap inputs_; // output arg name => output variable names diff --git a/paddle/framework/op_kernel_type.h b/paddle/framework/op_kernel_type.h index 312bd5f892ac23c847c87388c9cadf2161028d3e..44adb94d2a8feb79a5ff93c6e32cdff52333166e 100644 --- a/paddle/framework/op_kernel_type.h +++ b/paddle/framework/op_kernel_type.h @@ -85,9 +85,14 @@ inline std::string KernelTypeToString(const OpKernelType& kernel_key) { return stream.str(); } +inline bool NeedTransformLayout(const DataLayout& l, const DataLayout& r) { + return l != DataLayout::kAnyLayout && r != DataLayout::kAnyLayout && l != r; +} + inline bool TransFromNeeded(const OpKernelType& l, const OpKernelType& r) { return (!platform::places_are_same_class(l.place_, r.place_)) || - (l.data_type_ != r.data_type_) || (l.data_layout_ != r.data_layout_); + (l.data_type_ != r.data_type_) || + NeedTransformLayout(l.data_layout_, r.data_layout_); } } // namespace framework diff --git a/paddle/framework/op_kernel_type_test.cc b/paddle/framework/op_kernel_type_test.cc index 649afeee8a846b0579545f2edff77e9dbe3b4dd8..cb23bbde01493d1a3b5845e77d6160a75f409c7a 100644 --- a/paddle/framework/op_kernel_type_test.cc +++ b/paddle/framework/op_kernel_type_test.cc @@ -26,9 +26,9 @@ TEST(OpKernelType, ToString) { OpKernelType op_kernel_type(DataType::FP32, CPUPlace(), DataLayout::kNCHW, LibraryType::kCUDNN); - ASSERT_EQ( - paddle::framework::KernelTypeToString(op_kernel_type), - "data_type[5]:data_layout[NCHW]:place[CPUPlace]:library_type[CUDNN]"); + ASSERT_EQ(paddle::framework::KernelTypeToString(op_kernel_type), + "data_type[float32]:data_layout[NCHW]:place[CPUPlace]:library_type[" + "CUDNN]"); } TEST(OpKernelType, Hash) { diff --git a/paddle/framework/op_registry.h b/paddle/framework/op_registry.h index d75c0233e8e0134ddf4edc50c07490a234b65cd0..5de9ae559c435439f30931c7840e54e0d2bb744c 100644 --- a/paddle/framework/op_registry.h +++ b/paddle/framework/op_registry.h @@ -177,16 +177,16 @@ class OpKernelRegistrar : public Registrar { /** * Macro to register OperatorKernel. */ -#define REGISTER_OP_KERNEL(op_type, DEVICE_TYPE, place_class, ...) \ - STATIC_ASSERT_GLOBAL_NAMESPACE( \ - __reg_op_kernel_##op_type##_##DEVICE_TYPE##__, \ - "REGISTER_OP_KERNEL must be called in global namespace"); \ - static ::paddle::framework::OpKernelRegistrar \ - __op_kernel_registrar_##op_type##_##DEVICE_TYPE##__(#op_type, \ - #DEVICE_TYPE); \ - int TouchOpKernelRegistrar_##op_type##_##DEVICE_TYPE() { \ - __op_kernel_registrar_##op_type##_##DEVICE_TYPE##__.Touch(); \ - return 0; \ +#define REGISTER_OP_KERNEL(op_type, LIBRARY_TYPE, place_class, ...) \ + STATIC_ASSERT_GLOBAL_NAMESPACE( \ + __reg_op_kernel_##op_type##_##LIBRARY_TYPE##__, \ + "REGISTER_OP_KERNEL must be called in global namespace"); \ + static ::paddle::framework::OpKernelRegistrar \ + __op_kernel_registrar_##op_type##_##LIBRARY_TYPE##__(#op_type, \ + #LIBRARY_TYPE); \ + int TouchOpKernelRegistrar_##op_type##_##LIBRARY_TYPE() { \ + __op_kernel_registrar_##op_type##_##LIBRARY_TYPE##__.Touch(); \ + return 0; \ } #define REGISTER_OP_CUDA_KERNEL(op_type, ...) \ @@ -208,14 +208,14 @@ class OpKernelRegistrar : public Registrar { static int use_op_itself_##op_type##_ __attribute__((unused)) = \ TouchOpRegistrar_##op_type() -#define USE_OP_DEVICE_KERNEL(op_type, DEVICE_TYPE) \ - STATIC_ASSERT_GLOBAL_NAMESPACE( \ - __use_op_kernel_##op_type##_##DEVICE_TYPE##__, \ - "USE_OP_DEVICE_KERNEL must be in global namespace"); \ - extern int TouchOpKernelRegistrar_##op_type##_##DEVICE_TYPE(); \ - static int use_op_kernel_##op_type##_##DEVICE_TYPE##_ \ - __attribute__((unused)) = \ - TouchOpKernelRegistrar_##op_type##_##DEVICE_TYPE() +#define USE_OP_DEVICE_KERNEL(op_type, LIBRARY_TYPE) \ + STATIC_ASSERT_GLOBAL_NAMESPACE( \ + __use_op_kernel_##op_type##_##LIBRARY_TYPE##__, \ + "USE_OP_DEVICE_KERNEL must be in global namespace"); \ + extern int TouchOpKernelRegistrar_##op_type##_##LIBRARY_TYPE(); \ + static int use_op_kernel_##op_type##_##LIBRARY_TYPE##_ \ + __attribute__((unused)) = \ + TouchOpKernelRegistrar_##op_type##_##LIBRARY_TYPE() // TODO(fengjiayi): The following macros // seems ugly, do we have better method? diff --git a/paddle/framework/operator.cc b/paddle/framework/operator.cc index 84c010df7c396fc21904ae3c980f5fad70b2ceac..4e854f54dd43d760bab44fb5f7cafeb13314b27c 100644 --- a/paddle/framework/operator.cc +++ b/paddle/framework/operator.cc @@ -22,9 +22,7 @@ limitations under the License. */ #include "paddle/framework/shape_inference.h" #include "paddle/framework/var_type.h" -DEFINE_bool(op_sync, false, - "Default cuda is asynchronous device, set to True will" - "force op run in synchronous mode."); +DECLARE_bool(benchmark); namespace paddle { namespace framework { @@ -485,9 +483,15 @@ void OperatorWithKernel::Run(const Scope& scope, // } auto expected_kernel_key = this->GetExpectedKernelType(ctx); - VLOG(3) << "expected_kernel_key:" << expected_kernel_key; + auto kernel_iter = kernels.find(expected_kernel_key); + if (kernel_iter == kernels.end()) { + PADDLE_THROW("op %s does not have kernel for %s", type_, + KernelTypeToString(expected_kernel_key)); + } + + // do data transform Scope& new_scope = scope.NewScope(); for (auto& var_name_item : this->Inputs()) { @@ -520,14 +524,12 @@ void OperatorWithKernel::Run(const Scope& scope, } } - auto kernel_iter = kernels.find(expected_kernel_key); - auto* new_dev_ctx = pool.Get(expected_kernel_key.place_); kernel_iter->second->Compute( ExecutionContext(*this, new_scope, *new_dev_ctx)); /*For profiling/benchmark only*/ - if (FLAGS_op_sync) { + if (FLAGS_benchmark) { new_dev_ctx->Wait(); } } diff --git a/paddle/framework/program_desc.cc b/paddle/framework/program_desc.cc index b5d9e5e385c1ba57169ef885824fc23b0f130692..15ea4035c6e6193105b621210a900e74d1466941 100644 --- a/paddle/framework/program_desc.cc +++ b/paddle/framework/program_desc.cc @@ -14,6 +14,7 @@ limitations under the License. */ #include "paddle/framework/program_desc.h" #include "paddle/framework/block_desc.h" +#include "paddle/framework/feed_fetch_type.h" namespace paddle { namespace framework { @@ -64,5 +65,27 @@ ProgramDesc::ProgramDesc(const std::string &binary_str) { } } +const std::vector ProgramDesc::GetFeedTargetNames() { + BlockDesc *global_block = blocks_[0].get(); + std::vector feed_target_names; + for (auto *op : global_block->AllOps()) { + if (op->Type() == kFeedOpType) { + feed_target_names.insert(feed_target_names.begin(), op->Output("Out")[0]); + } + } + return feed_target_names; +} + +const std::vector ProgramDesc::GetFetchTargetNames() { + BlockDesc *global_block = blocks_[0].get(); + std::vector fetch_target_names; + for (auto *op : global_block->AllOps()) { + if (op->Type() == kFetchOpType) { + fetch_target_names.push_back(op->Input("X")[0]); + } + } + return fetch_target_names; +} + } // namespace framework } // namespace paddle diff --git a/paddle/framework/program_desc.h b/paddle/framework/program_desc.h index 15a962bb696d6172acd1a83cf9bb1ffd0846d449..8e958eab6ee08436ca73b13bac010e66c7df2b8b 100644 --- a/paddle/framework/program_desc.h +++ b/paddle/framework/program_desc.h @@ -16,6 +16,7 @@ limitations under the License. */ #include #include +#include "paddle/framework/block_desc.h" #include "paddle/framework/framework.pb.h" #include "paddle/framework/proto_desc.h" #include "paddle/platform/macros.h" @@ -45,6 +46,9 @@ class ProgramDesc { proto::ProgramDesc *Proto(); + const std::vector GetFeedTargetNames(); + const std::vector GetFetchTargetNames(); + private: proto::ProgramDesc desc_; diff --git a/paddle/framework/prune.cc b/paddle/framework/prune.cc index 25eb813ffb96e9b1e13299421ead9f85c02da59f..bff8e0bceaca9749101b2c45edddba526d565624 100644 --- a/paddle/framework/prune.cc +++ b/paddle/framework/prune.cc @@ -17,6 +17,7 @@ limitations under the License. */ #include #include #include +#include #include #include @@ -102,6 +103,32 @@ void prune_impl(const proto::ProgramDesc& input, proto::ProgramDesc* output, *op_field->Add() = input.blocks(block_id).ops(i); } } + + // remove the VarDescs in BlockDesc that are not referenced in + // the pruned OpDescs + std::unordered_map var_map; + auto* var_field = output->mutable_blocks(block_id)->mutable_vars(); + for (const auto& var : *var_field) { + var_map[var.name()] = var; + } + + var_field->Clear(); + for (const auto& op : *op_field) { + // add VarDescs of all input arguments for each OpDesc + auto& input_field = op.inputs(); + for (auto& input_var : input_field) { + for (auto& arg : input_var.arguments()) { + *var_field->Add() = var_map[arg]; + } + } + // add VarDescs of all output arguments for each OpDesc + auto& output_field = op.outputs(); + for (auto& output_var : output_field) { + for (auto& arg : output_var.arguments()) { + *var_field->Add() = var_map[arg]; + } + } + } } // TODO(fengjiayi): Prune() could be inplaced to avoid unnecessary copies diff --git a/paddle/framework/scope.cc b/paddle/framework/scope.cc index 2bd0ac8f5a9eb6439a4196dd9c61e13797c1a8e3..af08b2ab816f63c05d4c65df9601c787e57994f5 100644 --- a/paddle/framework/scope.cc +++ b/paddle/framework/scope.cc @@ -20,6 +20,12 @@ limitations under the License. */ #include "paddle/framework/threadpool.h" #include "paddle/string/printf.h" +DEFINE_bool(benchmark, false, + "Doing memory benchmark. It will make deleting scope synchronized, " + "and add some memory usage logs." + "Default cuda is asynchronous device, set to True will" + "force op run in synchronous mode."); + namespace paddle { namespace framework { @@ -88,8 +94,12 @@ void Scope::DeleteScope(Scope* scope) { auto it = std::find(this->kids_.begin(), this->kids_.end(), scope); PADDLE_ENFORCE(it != this->kids_.end(), "Cannot find %p as kid scope", scope); this->kids_.erase(it); - // Make delete async. - Async([scope] { delete scope; }); + // When making memory benchmark on Fluid, we have to delete scope sync. + if (FLAGS_benchmark) { + delete scope; + } else { + Async([scope] { delete scope; }); + } } void Scope::Rename(const std::string& origin_name, diff --git a/paddle/framework/tensor.h b/paddle/framework/tensor.h index 4aaa29d794c95592832a1fe990e2dce274eba9d5..f0ea709a5c37e769e3ffa1b2e9d1e39721979251 100644 --- a/paddle/framework/tensor.h +++ b/paddle/framework/tensor.h @@ -47,6 +47,11 @@ class Tensor { public: Tensor() : offset_(0) {} + /*! Constructor with place should only be used in pybind. */ + explicit Tensor(const platform::Place& place) : offset_(0) { + holder_->set_place(place); + } + /*! Return a pointer to mutable memory block. */ template inline T* data(); @@ -137,6 +142,7 @@ class Tensor { virtual std::type_index type() const = 0; virtual platform::Place place() const = 0; virtual void set_type(std::type_index type) = 0; + virtual void set_place(platform::Place place) = 0; }; template @@ -156,6 +162,7 @@ class Tensor { virtual void* ptr() const { return static_cast(ptr_.get()); } virtual std::type_index type() const { return type_; } virtual void set_type(std::type_index type) { type_ = type; } + virtual void set_place(platform::Place place) { place_ = place; } /*! the pointer of memory block. */ std::unique_ptr> ptr_; diff --git a/paddle/framework/tensor_test.cc b/paddle/framework/tensor_test.cc index c04cd38697f85b1d51aaa42b34a189f22bb4d5cc..9a387526ac2b0982aa4c931a9d92d98b08fb4f98 100644 --- a/paddle/framework/tensor_test.cc +++ b/paddle/framework/tensor_test.cc @@ -11,18 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "paddle/framework/tensor.h" #include @@ -60,9 +48,6 @@ TEST(Tensor, DataAssert) { ASSERT_TRUE(caught); } -/* following tests are not available at present - because Memory::Alloc() and Memory::Free() have not been ready. -*/ TEST(Tensor, MutableData) { { framework::Tensor src_tensor; diff --git a/paddle/framework/tensor_util_test.cc b/paddle/framework/tensor_util_test.cc index f541927c0e7e044adeabee01e8ec91e7d8ef7baf..906b0b5656301eebbb9f61fad2a3cb4e464a83e8 100644 --- a/paddle/framework/tensor_util_test.cc +++ b/paddle/framework/tensor_util_test.cc @@ -11,18 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "paddle/framework/tensor_util.h" #include diff --git a/paddle/framework/threadpool.cc b/paddle/framework/threadpool.cc index 109a7e7dc440d91e8223f2c0924f489f54a06f64..b7d7c00bcf9d9770f58284023ca2defcda299d64 100644 --- a/paddle/framework/threadpool.cc +++ b/paddle/framework/threadpool.cc @@ -1,24 +1,95 @@ /* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 + http://www.apache.org/licenses/LICENSE-2.0 -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. */ #include "paddle/framework/threadpool.h" +#include "paddle/platform/enforce.h" + namespace paddle { namespace framework { -std::unique_ptr ThreadPool::threadpool(nullptr); -std::once_flag ThreadPool::init_flag; +std::unique_ptr ThreadPool::threadpool_(nullptr); +std::once_flag ThreadPool::init_flag_; + +ThreadPool* ThreadPool::GetInstance() { + std::call_once(init_flag_, &ThreadPool::Init); + return threadpool_.get(); +} + +void ThreadPool::Init() { + if (threadpool_.get() == nullptr) { + // TODO(Yancey1989): specify the max threads number + int num_threads = std::thread::hardware_concurrency(); + PADDLE_ENFORCE_GT(num_threads, 0); + threadpool_.reset(new ThreadPool(num_threads)); + } +} + +ThreadPool::ThreadPool(int num_threads) + : total_threads_(num_threads), idle_threads_(num_threads), running_(true) { + threads_.resize(num_threads); + for (auto& thread : threads_) { + // TODO(Yancey1989): binding the thread on the specify CPU number + thread.reset(new std::thread(std::bind(&ThreadPool::TaskLoop, this))); + } +} + +ThreadPool::~ThreadPool() { + { + // notify all threads to stop running + running_ = false; + scheduled_.notify_all(); + } + + for (auto& t : threads_) { + t->join(); + t.reset(nullptr); + } +} + +void ThreadPool::Wait() { + std::unique_lock lock(mutex_); + completed_.wait(lock, [=] { return Done() == true; }); +} + +void ThreadPool::TaskLoop() { + while (running_) { + std::unique_lock lock(mutex_); + scheduled_.wait(lock, [=] { return !tasks_.empty() || !running_; }); + + if (!running_) { + break; + } + // pop a task from the task queue + auto task = std::move(tasks_.front()); + tasks_.pop(); + + --idle_threads_; + lock.unlock(); + + // run the task + task(); + + { + std::unique_lock lock(mutex_); + ++idle_threads_; + if (Done()) { + completed_.notify_all(); + } + } + } +} } // namespace framework } // namespace paddle diff --git a/paddle/framework/threadpool.h b/paddle/framework/threadpool.h index 3ac345851c38557f82698786dd3bc8e1202a4256..4e9b58679d9e7c84adf76b6245b397c7a8872483 100644 --- a/paddle/framework/threadpool.h +++ b/paddle/framework/threadpool.h @@ -20,52 +20,36 @@ limitations under the License. */ #include #include #include +#include -#include "paddle/platform/enforce.h" +#include "paddle/platform/macros.h" // for DISABLE_COPY_AND_ASSIGN namespace paddle { namespace framework { +// ThreadPool maintains a queue of tasks, and runs them using a fixed +// number of threads. class ThreadPool { public: typedef std::packaged_task Task; - /** - * @brief Get a instance of threadpool, the thread number will - * be specified as the number of hardware thread contexts - */ - static ThreadPool* GetInstance() { - std::call_once(init_flag, &ThreadPool::Init); - return threadpool.get(); - } + // Returns the singleton of ThreadPool. + static ThreadPool* GetInstance(); - ~ThreadPool() { - { - // notify all threads to stop running - running_ = false; - scheduled_.notify_all(); - } - - for (auto& t : threads_) { - t->join(); - t.reset(nullptr); - } - } + ~ThreadPool(); - int GetNumThreads() const { return num_threads_; } + // Returns the number of threads created by the constructor. + size_t Threads() const { return total_threads_; } - int GetAvailable() { + // Returns the number of currently idle threads. + size_t IdleThreads() { std::unique_lock lock(mutex_); - return available_; + return idle_threads_; } - /** - * @brief Push a function to the queue, and will be scheduled and - * executed if a thread is available. - * @param[in] Task, will be pushed to the task queue. - * @return std::future, we could wait for the task finished by - * f.wait(). - */ + // Run pushes a function to the task queue and returns a std::future + // object. To wait for the completion of the task, call + // std::future::wait(). template std::future Run(Callback fn) { std::unique_lock lock(mutex_); @@ -77,84 +61,40 @@ class ThreadPool { return f; } - /** - * @brief Wait until all the tasks are completed. - */ - void Wait() { - std::unique_lock lock(mutex_); - completed_.wait(lock, [=] { return Done() == true; }); - } + // Wait until all the tasks are completed. + void Wait(); private: DISABLE_COPY_AND_ASSIGN(ThreadPool); - explicit ThreadPool(int num_threads) - : num_threads_(num_threads), available_(num_threads), running_(true) { - threads_.resize(num_threads); - for (auto& thread : threads_) { - // TODO(Yancey1989): binding the thread on the specify CPU number - thread.reset(new std::thread(std::bind(&ThreadPool::TaskLoop, this))); - } - } + explicit ThreadPool(int num_threads); - /** - * @brief If the task queue is empty and avaialbe - * is equal to the number of threads, means that - * all tasks are completed. - * - * Note: this function is not thread-safe. - * - * @return true if all tasks are completed. - */ - bool Done() { return tasks_.empty() && available_ == num_threads_; } - - void TaskLoop() { - while (running_) { - std::unique_lock lock(mutex_); - scheduled_.wait(lock, [=] { return !tasks_.empty() || !running_; }); - - if (!running_) { - break; - } - // pop a task from the task queue - auto task = std::move(tasks_.front()); - tasks_.pop(); - - --available_; - lock.unlock(); - - // run the task - task(); - - { - std::unique_lock lock(mutex_); - ++available_; - if (Done()) { - completed_.notify_all(); - } - } - } - } + // If the task queue is empty and avaialbe is equal to the number of + // threads, means that all tasks are completed. Note: this function + // is not thread-safe. Returns true if all tasks are completed. + // Note: don't delete the data member total_threads_ and use + // threads_.size() instead; because you'd need to lock the mutex + // before accessing threads_. + bool Done() { return tasks_.empty() && idle_threads_ == total_threads_; } - static void Init() { - if (threadpool.get() == nullptr) { - // TODO(Yancey1989): specify the max threads number - int num_threads = std::thread::hardware_concurrency(); - PADDLE_ENFORCE_GT(num_threads, 0); - threadpool.reset(new ThreadPool(num_threads)); - } - } + // The constructor starts threads to run TaskLoop, which retrieves + // and runs tasks from the queue. + void TaskLoop(); + + // Init is called by GetInstance. + static void Init(); private: - static std::unique_ptr threadpool; - static std::once_flag init_flag; + static std::unique_ptr threadpool_; + static std::once_flag init_flag_; - int num_threads_; - int available_; - bool running_; - std::queue tasks_; std::vector> threads_; + const size_t total_threads_; + size_t idle_threads_; + + std::queue tasks_; std::mutex mutex_; + bool running_; std::condition_variable scheduled_; std::condition_variable completed_; }; diff --git a/paddle/framework/threadpool_test.cc b/paddle/framework/threadpool_test.cc index 50b6238cd8786be9d8cf2d5f821daadea12bd208..3fbfe7efc867144dbd0dd2613c824c6a3c41b7d8 100644 --- a/paddle/framework/threadpool_test.cc +++ b/paddle/framework/threadpool_test.cc @@ -22,11 +22,7 @@ namespace framework = paddle::framework; void do_sum(framework::ThreadPool* pool, std::atomic& sum, int cnt) { std::vector> fs; for (int i = 0; i < cnt; ++i) { - auto f = pool->Run([&sum]() { sum.fetch_add(1); }); - fs.push_back(std::move(f)); - } - for (auto& f : fs) { - f.wait(); + fs.push_back(framework::Async([&sum]() { sum.fetch_add(1); })); } } diff --git a/paddle/framework/type_defs.h b/paddle/framework/type_defs.h index d834d343759fa279a1444c6337956ffce1b9061a..1eedbbc419ab660f5ce00aa891ef80ca245bc0a8 100644 --- a/paddle/framework/type_defs.h +++ b/paddle/framework/type_defs.h @@ -35,7 +35,7 @@ using VariableNameMap = std::map>; using Attribute = boost::variant, std::vector, std::vector, bool, - std::vector, BlockDesc*>; + std::vector, BlockDesc*, int64_t>; using AttributeMap = std::unordered_map; diff --git a/paddle/framework/var_desc.h b/paddle/framework/var_desc.h index fc482c467404a6b9dfed64c43871d91d3d10c766..9316b14bb695c185efd6db4296d422ef0c476d57 100644 --- a/paddle/framework/var_desc.h +++ b/paddle/framework/var_desc.h @@ -66,6 +66,8 @@ class VarDesc { std::string Name() const { return desc_.name(); } + void SetName(std::string name) { desc_.set_name(name); } + void SetShape(const std::vector &dims); void SetDataType(proto::DataType data_type); diff --git a/paddle/framework/variable.h b/paddle/framework/variable.h index 03992c8608693259224901b2f9c89d458f126d09..3b7ec0a2a90d8f88bfb7f1629f484b3a8a8078df 100644 --- a/paddle/framework/variable.h +++ b/paddle/framework/variable.h @@ -11,18 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #pragma once #include diff --git a/paddle/framework/variable_test.cc b/paddle/framework/variable_test.cc index aea03bcf5719dacc01d2d78b52b33e8a0b29b5e5..e5585c8724d712e273d086001b6cbc3d59c46ebe 100644 --- a/paddle/framework/variable_test.cc +++ b/paddle/framework/variable_test.cc @@ -1,15 +1,16 @@ -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - http://www.apache.org/licenses/LICENSE-2.0 - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. #include #include diff --git a/paddle/gserver/layers/MKLDNNConcatLayer.cpp b/paddle/gserver/layers/MKLDNNConcatLayer.cpp index 44bb0883b89c712d70e2d4fdfe16bdfde86f81b7..520ccc1a995e966de73080b61a8c20cbee722267 100644 --- a/paddle/gserver/layers/MKLDNNConcatLayer.cpp +++ b/paddle/gserver/layers/MKLDNNConcatLayer.cpp @@ -43,7 +43,7 @@ void MKLDNNConcatLayer::reshape( channels_[0] = ic; oc = ic; for (size_t i = 1; i < inputLayers_.size(); i++) { - int batchsize, height, witdh; + int batchsize = 0, height = 0, witdh = 0; reshapeInput(batchsize, height, witdh, i); CHECK_EQ(bs, batchsize); CHECK_EQ(ih, height); @@ -84,6 +84,7 @@ void MKLDNNConcatLayer::resetFwdBuffers(std::vector& inputs, bool has8c = false, has16c = false, hasnc = false; for (size_t i = 0; i < inputs.size(); i++) { resetInValue(inputs[i], nullptr, i, channels_[i]); + inputs[i]->downSpatial(); CHECK(inputs[i]); auto dm = inputs[i]->getDims(); // inputs format can be different, but ndims must equal diff --git a/paddle/gserver/layers/PriorBox.cpp b/paddle/gserver/layers/PriorBox.cpp index 331bc7672ec0d39a7317c39f1d14e8dcadea471a..8faf032f550836579522016b4fff3db7e94746e3 100644 --- a/paddle/gserver/layers/PriorBox.cpp +++ b/paddle/gserver/layers/PriorBox.cpp @@ -65,14 +65,19 @@ bool PriorBoxLayer::init(const LayerMap& layerMap, std::copy(pbConf.aspect_ratio().begin(), pbConf.aspect_ratio().end(), std::back_inserter(tmp)); - // flip - int inputRatioLength = tmp.size(); - for (int index = 0; index < inputRatioLength; index++) { - aspectRatio_.push_back(tmp[index]); - aspectRatio_.push_back(1 / tmp[index]); + + if (maxSize_.size() > 0) CHECK_EQ(minSize_.size(), maxSize_.size()); + + // flip aspect ratios + for (unsigned index = 0; index < tmp.size(); index++) { + real ar = tmp[index]; + if (fabs(ar - 1.) < 1e-6) continue; + aspectRatio_.push_back(ar); + aspectRatio_.push_back(1. / ar); } - numPriors_ = aspectRatio_.size(); - if (maxSize_.size() > 0) numPriors_++; + + numPriors_ = aspectRatio_.size() * minSize_.size() + maxSize_.size(); + return true; } @@ -99,50 +104,39 @@ void PriorBoxLayer::forward(PassType passType) { for (int w = 0; w < layerWidth; ++w) { real centerX = (w + 0.5) * stepW; real centerY = (h + 0.5) * stepH; - real minSize = 0; for (size_t s = 0; s < minSize_.size(); s++) { - // first prior. - minSize = minSize_[s]; + real minSize = minSize_[s]; real boxWidth = minSize; real boxHeight = minSize; - // xmin, ymin, xmax, ymax. - tmpPtr[idx++] = (centerX - boxWidth / 2.) / imageWidth; - tmpPtr[idx++] = (centerY - boxHeight / 2.) / imageHeight; - tmpPtr[idx++] = (centerX + boxWidth / 2.) / imageWidth; - tmpPtr[idx++] = (centerY + boxHeight / 2.) / imageHeight; - // set the variance. - for (int t = 0; t < 4; t++) tmpPtr[idx++] = variance_[t]; + + // priors with different aspect ratios + for (size_t r = 0; r < aspectRatio_.size(); r++) { + real ar = aspectRatio_[r]; + boxWidth = minSize * sqrt(ar); + boxHeight = minSize / sqrt(ar); + tmpPtr[idx++] = (centerX - boxWidth / 2.) / imageWidth; + tmpPtr[idx++] = (centerY - boxHeight / 2.) / imageHeight; + tmpPtr[idx++] = (centerX + boxWidth / 2.) / imageWidth; + tmpPtr[idx++] = (centerY + boxHeight / 2.) / imageHeight; + // set the variance. + for (int t = 0; t < 4; t++) tmpPtr[idx++] = variance_[t]; + } if (maxSize_.size() > 0) { - CHECK_EQ(minSize_.size(), maxSize_.size()); - // second prior. - for (size_t s = 0; s < maxSize_.size(); s++) { - real maxSize = maxSize_[s]; - boxWidth = boxHeight = sqrt(minSize * maxSize); - tmpPtr[idx++] = (centerX - boxWidth / 2.) / imageWidth; - tmpPtr[idx++] = (centerY - boxHeight / 2.) / imageHeight; - tmpPtr[idx++] = (centerX + boxWidth / 2.) / imageWidth; - tmpPtr[idx++] = (centerY + boxHeight / 2.) / imageHeight; - // set the variance. - for (int t = 0; t < 4; t++) tmpPtr[idx++] = variance_[t]; - } + // square prior with size sqrt(minSize * maxSize) + real maxSize = maxSize_[s]; + boxWidth = boxHeight = sqrt(minSize * maxSize); + tmpPtr[idx++] = (centerX - boxWidth / 2.) / imageWidth; + tmpPtr[idx++] = (centerY - boxHeight / 2.) / imageHeight; + tmpPtr[idx++] = (centerX + boxWidth / 2.) / imageWidth; + tmpPtr[idx++] = (centerY + boxHeight / 2.) / imageHeight; + // set the variance. + for (int t = 0; t < 4; t++) tmpPtr[idx++] = variance_[t]; } } - // rest of priors. - for (size_t r = 0; r < aspectRatio_.size(); r++) { - real ar = aspectRatio_[r]; - if (fabs(ar - 1.) < 1e-6) continue; - real boxWidth = minSize * sqrt(ar); - real boxHeight = minSize / sqrt(ar); - tmpPtr[idx++] = (centerX - boxWidth / 2.) / imageWidth; - tmpPtr[idx++] = (centerY - boxHeight / 2.) / imageHeight; - tmpPtr[idx++] = (centerX + boxWidth / 2.) / imageWidth; - tmpPtr[idx++] = (centerY + boxHeight / 2.) / imageHeight; - // set the variance. - for (int t = 0; t < 4; t++) tmpPtr[idx++] = variance_[t]; - } } } + // clip the prior's coordidate such that it is within [0, 1] for (int d = 0; d < dim * 2; ++d) if ((d % 8) < 4) diff --git a/paddle/gserver/tests/img_conv_cudnn.py b/paddle/gserver/tests/img_conv_cudnn.py index e424261bda2c0fea4b1796ce443810e8e4e4599b..0ea6d6bae66b0a307748bd0d0fa9a53ed5f7927d 100644 --- a/paddle/gserver/tests/img_conv_cudnn.py +++ b/paddle/gserver/tests/img_conv_cudnn.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # #Licensed under the Apache License, Version 2.0 (the "License"); #you may not use this file except in compliance with the License. @@ -11,20 +11,6 @@ #WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #See the License for the specific language governing permissions and #limitations under the License. -#edit-mode: -*- python -*- -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/paddle/gserver/tests/img_conv_exconv.py b/paddle/gserver/tests/img_conv_exconv.py index 3b59cd80b8b8349f022b774b36d4e4db2d517b73..c618cdab27c52d70042b0a118f7f6fe935a6b9d7 100644 --- a/paddle/gserver/tests/img_conv_exconv.py +++ b/paddle/gserver/tests/img_conv_exconv.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # #Licensed under the Apache License, Version 2.0 (the "License"); #you may not use this file except in compliance with the License. @@ -11,20 +11,6 @@ #WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #See the License for the specific language governing permissions and #limitations under the License. -#edit-mode: -*- python -*- -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/paddle/gserver/tests/pyDataProvider.py b/paddle/gserver/tests/pyDataProvider.py index 7235a239439b7544805d1bd06dfb1a72c2e0e937..d2ad5888b5a4c79d8b663ce8c2f313184151beb6 100644 --- a/paddle/gserver/tests/pyDataProvider.py +++ b/paddle/gserver/tests/pyDataProvider.py @@ -1,17 +1,16 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. import numpy import struct import traceback diff --git a/paddle/gserver/tests/rnn_data_provider.py b/paddle/gserver/tests/rnn_data_provider.py index 913365a5a4037d14fcba1e1546508ba89668e0d6..063a4127e542d23012359a2eac0045bf69a51356 100644 --- a/paddle/gserver/tests/rnn_data_provider.py +++ b/paddle/gserver/tests/rnn_data_provider.py @@ -1,17 +1,16 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. from paddle.trainer.PyDataProvider2 import * # Note that each config should has an independent provider diff --git a/paddle/gserver/tests/sequenceGen.py b/paddle/gserver/tests/sequenceGen.py index fd725727c04677b5ea8918f6721f0c007e80915d..04a1732d61c8618984d16550acf7c94da1bd3578 100644 --- a/paddle/gserver/tests/sequenceGen.py +++ b/paddle/gserver/tests/sequenceGen.py @@ -1,17 +1,16 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. import os import sys diff --git a/paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py b/paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py index 7303d088043d5096a3491d3b3b32b231bde09a0a..aeaaa221f9fab981af88cfd63c30349e1b02a0ee 100644 --- a/paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py +++ b/paddle/gserver/tests/sequence_nest_rnn_multi_unequalength_inputs.py @@ -1,18 +1,16 @@ -# edit-mode: -*- python -*- -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. from paddle.trainer_config_helpers import * ######################## data source ################################ diff --git a/paddle/gserver/tests/sequence_recurrent.py b/paddle/gserver/tests/sequence_recurrent.py index d20d3331ae6887aa3bca9df7e7817e08ebb91a48..8786a5465db82d786d3772357b02ab837073a576 100644 --- a/paddle/gserver/tests/sequence_recurrent.py +++ b/paddle/gserver/tests/sequence_recurrent.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # #Licensed under the Apache License, Version 2.0 (the "License"); #you may not use this file except in compliance with the License. @@ -11,20 +11,6 @@ #WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #See the License for the specific language governing permissions and #limitations under the License. -#!/usr/bin/env python -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/paddle/gserver/tests/sequence_recurrent_group.py b/paddle/gserver/tests/sequence_recurrent_group.py index a1d54542e3bc4e89f70d31d5e89c0f44953c9f90..8b5a3d49838c9bb49321a9d7514fc0241e6d67cd 100644 --- a/paddle/gserver/tests/sequence_recurrent_group.py +++ b/paddle/gserver/tests/sequence_recurrent_group.py @@ -1,18 +1,16 @@ -#!/usr/bin/env python -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. - from paddle.trainer_config_helpers import * ######################## data source ################################ diff --git a/paddle/gserver/tests/sequence_rnn_matched_inputs.py b/paddle/gserver/tests/sequence_rnn_matched_inputs.py index b156fa9baaf386a8625d63020d72828047ddca2f..0c55f2cf9d07b194aa06f88892f831f1a9ce6436 100644 --- a/paddle/gserver/tests/sequence_rnn_matched_inputs.py +++ b/paddle/gserver/tests/sequence_rnn_matched_inputs.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # #Licensed under the Apache License, Version 2.0 (the "License"); #you may not use this file except in compliance with the License. @@ -11,20 +11,6 @@ #WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #See the License for the specific language governing permissions and #limitations under the License. -# edit-mode: -*- python -*- -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/paddle/gserver/tests/sequence_rnn_mixed_inputs.py b/paddle/gserver/tests/sequence_rnn_mixed_inputs.py index c2c98aae5f5c5f73c61b309382f2e3d7d649ef47..22b376b91aa4736d16fead698105466d679dd248 100644 --- a/paddle/gserver/tests/sequence_rnn_mixed_inputs.py +++ b/paddle/gserver/tests/sequence_rnn_mixed_inputs.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # #Licensed under the Apache License, Version 2.0 (the "License"); #you may not use this file except in compliance with the License. @@ -11,20 +11,6 @@ #WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #See the License for the specific language governing permissions and #limitations under the License. -# edit-mode: -*- python -*- -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py b/paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py index c812c6ced2458b6c40e141606655aa45e2bbc55a..3ce87490bbd0f30a3c42b947b073adb2a6c5b51c 100644 --- a/paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py +++ b/paddle/gserver/tests/sequence_rnn_multi_unequalength_inputs.py @@ -1,4 +1,4 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # #Licensed under the Apache License, Version 2.0 (the "License"); #you may not use this file except in compliance with the License. @@ -11,20 +11,6 @@ #WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. #See the License for the specific language governing permissions and #limitations under the License. -#edit-mode: -*- python -*- -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. from paddle.trainer_config_helpers import * diff --git a/paddle/gserver/tests/test_LayerGrad.cpp b/paddle/gserver/tests/test_LayerGrad.cpp index ba83667ebc9a89c37f77a7f71e6df90b54723cc0..aab02f16849582db4b41087046b810463a855e1a 100644 --- a/paddle/gserver/tests/test_LayerGrad.cpp +++ b/paddle/gserver/tests/test_LayerGrad.cpp @@ -991,8 +991,10 @@ TEST(Layer, SequenceLastInstanceLayer) { "seqlastins", "non-seq", -1); // hasSubseq seqlastins to non-seq - testDegradeLayer( - true, "seqlastins", "seq", -1); // hasSubseq seqlastins to seq + testDegradeLayer(true, + "seqlastins", + "seq", + -1); // hasSubseq seqlastins to seq } TEST(Layer, AverageLayer) { @@ -1001,8 +1003,10 @@ TEST(Layer, AverageLayer) { "average", "non-seq", 5); // seq average to a shorten seq, stride window = 5 - testDegradeLayer( - true, "average", "non-seq", -1); // hasSubseq average to non-seq + testDegradeLayer(true, + "average", + "non-seq", + -1); // hasSubseq average to non-seq testDegradeLayer(true, "average", "seq", -1); // hasSubseq average to seq } @@ -1287,8 +1291,9 @@ TEST(Layer, PoolLayer) { testPoolLayer("cudnn-avg-pool", /* trans= */ false, /* useGpu= */ true); testPoolLayer2("cudnn-max-pool", /* trans= */ false, /* useGpu= */ true); testPoolLayer2("cudnn-avg-pool", /* trans= */ false, /* useGpu= */ true); - testPoolLayer2( - "cudnn-avg-incl-pad-pool", /* trans= */ false, /* useGpu= */ true); + testPoolLayer2("cudnn-avg-incl-pad-pool", + /* trans= */ false, + /* useGpu= */ true); testPoolLayer("max-pool-with-mask", /* trans= */ false, /* useGpu= */ true); #endif } @@ -2431,18 +2436,21 @@ TEST(Layer, test3DDeConvLayer) { } TEST(Layer, ScaleShiftLayer) { - const size_t batchSize = 16; - const size_t size = 32; - TestConfig config; - config.layerConfig.set_type("scale_shift"); - config.layerConfig.set_size(size); - config.biasSize = 1; - config.inputDefs.push_back( - {INPUT_DATA, "input", /* dim= */ size, /* paraSize= */ 1}); - config.layerConfig.add_inputs(); - for (auto useGpu : {false, true}) { - testLayerGrad(config, "scale_shift", batchSize, false, useGpu, false); - } + // FIXME: Disable ScaleShiftLayer because it is not stable. + // https://github.com/PaddlePaddle/Paddle/issues/7781 + return; + // const size_t batchSize = 16; + // const size_t size = 32; + // TestConfig config; + // config.layerConfig.set_type("scale_shift"); + // config.layerConfig.set_size(size); + // config.biasSize = 1; + // config.inputDefs.push_back( + // {INPUT_DATA, "input", /* dim= */ size, /* paraSize= */ 1}); + // config.layerConfig.add_inputs(); + // for (auto useGpu : {false, true}) { + // testLayerGrad(config, "scale_shift", batchSize, false, useGpu, false); + // } } TEST(Layer, ScaleSubRegionLayer) { diff --git a/paddle/gserver/tests/test_PyDataProvider2.py b/paddle/gserver/tests/test_PyDataProvider2.py index 0d0fe476ff5eac8bf8ad1c9fe09b32c1a8f73ebc..044aede98e684a432c48b3ea5bb82a4a677682d4 100644 --- a/paddle/gserver/tests/test_PyDataProvider2.py +++ b/paddle/gserver/tests/test_PyDataProvider2.py @@ -1,17 +1,16 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. # -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. import random from paddle.trainer.PyDataProvider2 import * diff --git a/paddle/inference/CMakeLists.txt b/paddle/inference/CMakeLists.txt index 8437b2b21942ead544dab8636db1b355b7cf7bd5..2289ddc139cbddfbaa5238e683b2f8e784a7291e 100644 --- a/paddle/inference/CMakeLists.txt +++ b/paddle/inference/CMakeLists.txt @@ -1,47 +1,29 @@ -set(FLUID_CORE_MODULES - backward proto_desc paddle_memory executor prune init ${GLOB_OP_LIB}) +set(FLUID_CORE_MODULES proto_desc paddle_memory lod_tensor executor prune init) cc_library(paddle_fluid_api - SRCS inference.cc - DEPS ${FLUID_CORE_MODULES}) + SRCS io.cc + DEPS ${FLUID_CORE_MODULES} ${GLOB_OP_LIB}) -# Merge all modules into a simgle static library -cc_library(paddle_fluid DEPS paddle_fluid_api ${FLUID_CORE_MODULES}) +# Merge all modules into a single static library +cc_library(paddle_fluid DEPS paddle_fluid_api ${FLUID_CORE_MODULES} ${GLOB_OP_LIB}) -# ptools -# just for testing, we may need to change the storing format for inference_model -# and move the dependent of pickle. -# download from http://www.picklingtools.com/ -# build in the C++ sub-directory, using command -# make -f Makefile.Linux libptools.so -set(PTOOLS_LIB) -set(PTOOLS_ROOT $ENV{PTOOLS_ROOT} CACHE PATH "Folder contains PicklingTools") -find_path(PTOOLS_INC_DIR chooseser.h PATHS ${PTOOLS_ROOT}/C++) -find_library(PTOOLS_SHARED_LIB NAMES ptools PATHS ${PTOOLS_ROOT}/C++) -if(PTOOLS_INC_DIR AND PTOOLS_SHARED_LIB) - add_definitions(-DPADDLE_USE_PTOOLS) - set(PTOOLS_LIB ptools) - message(STATUS "Found PicklingTools: ${PTOOLS_SHARED_LIB}") - add_library(${PTOOLS_LIB} SHARED IMPORTED GLOBAL) - set_property(TARGET ${PTOOLS_LIB} PROPERTY IMPORTED_LOCATION ${PTOOLS_SHARED_LIB}) - include_directories(${PTOOLS_ROOT}/C++) - include_directories(${PTOOLS_ROOT}/C++/opencontainers_1_8_5/include) - add_definitions(-DOC_NEW_STYLE_INCLUDES) # used in ptools +# Create shared library +add_library(paddle_fluid_shared SHARED io.cc) + +target_circle_link_libraries(paddle_fluid_shared + ARCHIVE_START + ${GLOB_OP_LIB} + ARCHIVE_END + ${FLUID_CORE_MODULES}) + +SET_TARGET_PROPERTIES(paddle_fluid_shared PROPERTIES OUTPUT_NAME paddle_fluid) + +# install library & headers +if(NOT WITH_C_API AND WITH_FLUID) + install(FILES io.h DESTINATION include/paddle/inference) + install(TARGETS paddle_fluid_shared DESTINATION lib) endif() -add_executable(example example.cc) -if(APPLE) - set(OPTIONAL_LINK_FLAGS) - if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang" OR "${CMAKE_CXX_COMPILER_ID}" STREQUAL "AppleClang") - set(OPTIONAL_LINK_FLAGS "-undefined dynamic_lookup") - endif() - target_link_libraries(example - -Wl,-force_load paddle_fluid - ${OPTIONAL_LINK_FLAGS} - ${PTOOLS_LIB}) -else() - target_link_libraries(example - -Wl,--start-group -Wl,--whole-archive paddle_fluid - -Wl,--no-whole-archive -Wl,--end-group - ${PTOOLS_LIB}) +if(WITH_TESTING) + add_subdirectory(tests/book) endif() diff --git a/paddle/inference/example.cc b/paddle/inference/example.cc deleted file mode 100644 index 9711b20e6fb4099a2cc497029468ebd1fd0b3456..0000000000000000000000000000000000000000 --- a/paddle/inference/example.cc +++ /dev/null @@ -1,79 +0,0 @@ -/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ - -#include -#include -#include "gflags/gflags.h" -#include "paddle/inference/inference.h" - -DEFINE_string(dirname, "", "Directory of the inference model."); -DEFINE_string(feed_var_names, "", "Names of feeding variables"); -DEFINE_string(fetch_var_names, "", "Names of fetching variables"); - -int main(int argc, char** argv) { - google::ParseCommandLineFlags(&argc, &argv, true); - if (FLAGS_dirname.empty() || FLAGS_feed_var_names.empty() || - FLAGS_fetch_var_names.empty()) { - // Example: - // ./example --dirname=recognize_digits_mlp.inference.model - // --feed_var_names="x" - // --fetch_var_names="fc_2.tmp_2" - std::cout << "Usage: ./example --dirname=path/to/your/model " - "--feed_var_names=x --fetch_var_names=y" - << std::endl; - exit(1); - } - - std::cout << "FLAGS_dirname: " << FLAGS_dirname << std::endl; - std::cout << "FLAGS_feed_var_names: " << FLAGS_feed_var_names << std::endl; - std::cout << "FLAGS_fetch_var_names: " << FLAGS_fetch_var_names << std::endl; - - std::string dirname = FLAGS_dirname; - std::vector feed_var_names = {FLAGS_feed_var_names}; - std::vector fetch_var_names = {FLAGS_fetch_var_names}; - - paddle::InferenceEngine* engine = new paddle::InferenceEngine(); - engine->LoadInferenceModel(dirname, feed_var_names, fetch_var_names); - - paddle::framework::LoDTensor input; - srand(time(0)); - float* input_ptr = - input.mutable_data({1, 784}, paddle::platform::CPUPlace()); - for (int i = 0; i < 784; ++i) { - input_ptr[i] = rand() / (static_cast(RAND_MAX)); - } - - std::vector feeds; - feeds.push_back(input); - std::vector fetchs; - engine->Execute(feeds, fetchs); - - for (size_t i = 0; i < fetchs.size(); ++i) { - auto dims_i = fetchs[i].dims(); - std::cout << "dims_i:"; - for (int j = 0; j < dims_i.size(); ++j) { - std::cout << " " << dims_i[j]; - } - std::cout << std::endl; - std::cout << "result:"; - float* output_ptr = fetchs[i].data(); - for (int j = 0; j < paddle::framework::product(dims_i); ++j) { - std::cout << " " << output_ptr[j]; - } - std::cout << std::endl; - } - - delete engine; - return 0; -} diff --git a/paddle/inference/inference.cc b/paddle/inference/inference.cc deleted file mode 100644 index 37b8b20ddfcf2566b8410f950308309e5b2b2a7c..0000000000000000000000000000000000000000 --- a/paddle/inference/inference.cc +++ /dev/null @@ -1,195 +0,0 @@ -/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ - -#include "inference.h" -#include -#include "paddle/framework/executor.h" -#include "paddle/framework/feed_fetch_method.h" -#include "paddle/framework/init.h" -#include "paddle/framework/scope.h" - -#ifdef PADDLE_USE_PTOOLS -#include "chooseser.h" -#endif - -namespace paddle { - -void InferenceEngine::LoadInferenceModel( - const std::string& dirname, - const std::vector& feed_var_names, - const std::vector& fetch_var_names) { -#ifdef PADDLE_USE_PTOOLS - std::string model_filename = dirname + "/__model__"; - LOG(INFO) << "Using PicklingTools, loading model from " << model_filename; - Val v; - LoadValFromFile(model_filename.c_str(), v, SERIALIZE_P0); - std::string program_desc_str = v["program_desc_str"]; - LOG(INFO) << "program_desc_str's size: " << program_desc_str.size(); -// PicklingTools cannot parse the vector of strings correctly. -#else - std::string model_filename = dirname + "/__model__.dat"; - LOG(INFO) << "loading model from " << model_filename; - std::ifstream inputfs(model_filename, std::ios::in | std::ios::binary); - std::string program_desc_str; - inputfs.seekg(0, std::ios::end); - program_desc_str.resize(inputfs.tellg()); - inputfs.seekg(0, std::ios::beg); - LOG(INFO) << "program_desc_str's size: " << program_desc_str.size(); - inputfs.read(&program_desc_str[0], program_desc_str.size()); - inputfs.close(); -#endif - program_ = new framework::ProgramDesc(program_desc_str); - GenerateLoadProgram(dirname); - - if (feed_var_names.empty() || fetch_var_names.empty()) { - LOG(FATAL) << "Please specify the feed_var_names and fetch_var_names."; - } - feed_var_names_ = feed_var_names; - fetch_var_names_ = fetch_var_names; - PrependFeedOp(); - AppendFetchOp(); -} - -bool InferenceEngine::IsParameter(const framework::VarDesc* var) { - if (var->Persistable()) { - // There are many unreachable variables in the program - for (size_t i = 0; i < program_->Size(); ++i) { - const framework::BlockDesc& block = program_->Block(i); - for (auto* op : block.AllOps()) { - for (auto input_argument_name : op->InputArgumentNames()) { - if (input_argument_name == var->Name()) { - return true; - } - } - } - } - } - return false; -} - -void InferenceEngine::GenerateLoadProgram(const std::string& dirname) { - framework::BlockDesc* global_block = program_->MutableBlock(0); - - load_program_ = new framework::ProgramDesc(); - framework::BlockDesc* load_block = load_program_->MutableBlock(0); - for (auto* var : global_block->AllVars()) { - if (IsParameter(var)) { - LOG(INFO) << "parameter's name: " << var->Name(); - - framework::VarDesc* new_var = load_block->Var(var->Name()); - new_var->SetShape(var->Shape()); - new_var->SetDataType(var->GetDataType()); - new_var->SetType(var->GetType()); - new_var->SetLoDLevel(var->GetLoDLevel()); - new_var->SetPersistable(true); - - // append_op - framework::OpDesc* op = load_block->AppendOp(); - op->SetType("load"); - op->SetOutput("Out", {new_var->Name()}); - op->SetAttr("file_path", {dirname + "/" + new_var->Name()}); - op->CheckAttrs(); - } - } -} - -void InferenceEngine::PrependFeedOp() { - if (!program_) { - LOG(FATAL) << "Please initialize the program_ first."; - } - - framework::BlockDesc* global_block = program_->MutableBlock(0); - - // create_var - framework::VarDesc* feed_var = global_block->Var("feed"); - feed_var->SetType(framework::proto::VarDesc::FEED_MINIBATCH); - feed_var->SetPersistable(true); - - // prepend feed_op - for (size_t i = 0; i < feed_var_names_.size(); ++i) { - std::string var_name = feed_var_names_[i]; - LOG(INFO) << "feed var's name: " << var_name; - - // prepend_op - framework::OpDesc* op = global_block->PrependOp(); - op->SetType("feed"); - op->SetInput("X", {"feed"}); - op->SetOutput("Out", {var_name}); - op->SetAttr("col", {static_cast(i)}); - op->CheckAttrs(); - } -} - -void InferenceEngine::AppendFetchOp() { - if (!program_) { - LOG(FATAL) << "Please initialize the program_ first."; - } - - framework::BlockDesc* global_block = program_->MutableBlock(0); - - // create_var - framework::VarDesc* fetch_var = global_block->Var("fetch"); - fetch_var->SetType(framework::proto::VarDesc::FETCH_LIST); - fetch_var->SetPersistable(true); - - // append fetch_op - for (size_t i = 0; i < fetch_var_names_.size(); ++i) { - std::string var_name = fetch_var_names_[i]; - LOG(INFO) << "fetch var's name: " << var_name; - - // append_op - framework::OpDesc* op = global_block->AppendOp(); - op->SetType("fetch"); - op->SetInput("X", {var_name}); - op->SetOutput("Out", {"fetch"}); - op->SetAttr("col", {static_cast(i)}); - op->CheckAttrs(); - } -} - -void InferenceEngine::Execute(const std::vector& feeds, - std::vector& fetchs) { - if (!program_ || !load_program_) { - LOG(FATAL) << "Please initialize the program_ and load_program_ first."; - } - - if (feeds.size() < feed_var_names_.size()) { - LOG(FATAL) << "Please feed " << feed_var_names_.size() << " input Tensors."; - } - - auto* place = new platform::CPUPlace(); - framework::InitDevices(); - framework::Executor* executor = new framework::Executor(*place); - framework::Scope* scope = new framework::Scope(); - - executor->Run(*load_program_, scope, 0, true, true); - - // set_feed_variable - for (size_t i = 0; i < feed_var_names_.size(); ++i) { - framework::SetFeedVariable(scope, feeds[i], "feed", i); - } - - executor->Run(*program_, scope, 0, true, true); - - // get_fetch_variable - fetchs.resize(fetch_var_names_.size()); - for (size_t i = 0; i < fetch_var_names_.size(); ++i) { - fetchs[i] = framework::GetFetchVariable(*scope, "fetch", i); - } - - delete place; - delete scope; - delete executor; -} -} // namespace paddle diff --git a/paddle/inference/inference.h b/paddle/inference/inference.h deleted file mode 100644 index a3f3ef4b440036a0b27353cc092eed1bbf96eeb3..0000000000000000000000000000000000000000 --- a/paddle/inference/inference.h +++ /dev/null @@ -1,50 +0,0 @@ -/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ - -#pragma once - -#include "paddle/framework/block_desc.h" -#include "paddle/framework/lod_tensor.h" -#include "paddle/framework/program_desc.h" - -namespace paddle { - -class InferenceEngine { -public: - InferenceEngine() : program_(nullptr), load_program_(nullptr) {} - ~InferenceEngine() { - delete program_; - delete load_program_; - } - - void LoadInferenceModel(const std::string& dirname, - const std::vector& feed_var_names, - const std::vector& fetch_var_names); - void Execute(const std::vector& feeds, - std::vector& fetchs); - -private: - bool IsParameter(const framework::VarDesc* var); - void GenerateLoadProgram(const std::string& dirname); - void PrependFeedOp(); - void AppendFetchOp(); - -private: - framework::ProgramDesc* program_; - framework::ProgramDesc* load_program_; - std::vector feed_var_names_; - std::vector fetch_var_names_; -}; - -} // namespace paddle diff --git a/paddle/inference/io.cc b/paddle/inference/io.cc new file mode 100644 index 0000000000000000000000000000000000000000..60ad7af1c0a469beb6a07bf057a8647fcb98cca8 --- /dev/null +++ b/paddle/inference/io.cc @@ -0,0 +1,98 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/inference/io.h" + +#include +#include "paddle/framework/block_desc.h" +#include "paddle/framework/feed_fetch_type.h" + +namespace paddle { +namespace inference { + +bool IsParameter(const framework::VarDesc* var, + const framework::ProgramDesc& main_program) { + if (var->Persistable()) { + // There are many unreachable variables in the program + for (size_t i = 0; i < main_program.Size(); ++i) { + const framework::BlockDesc& block = main_program.Block(i); + for (auto* op : block.AllOps()) { + if (op->Type() == framework::kFeedOpType) { + continue; + } + for (auto input_argument_name : op->InputArgumentNames()) { + if (input_argument_name == var->Name()) { + return true; + } + } + } + } + } + return false; +} + +void LoadPersistables(framework::Executor& executor, + framework::Scope& scope, + const std::string& dirname, + const framework::ProgramDesc& main_program) { + const framework::BlockDesc& global_block = main_program.Block(0); + + framework::ProgramDesc* load_program = new framework::ProgramDesc(); + framework::BlockDesc* load_block = load_program->MutableBlock(0); + for (auto* var : global_block.AllVars()) { + if (IsParameter(var, main_program)) { + VLOG(3) << "parameter's name: " << var->Name(); + + framework::VarDesc* new_var = load_block->Var(var->Name()); + new_var->SetShape(var->Shape()); + new_var->SetDataType(var->GetDataType()); + new_var->SetType(var->GetType()); + new_var->SetLoDLevel(var->GetLoDLevel()); + new_var->SetPersistable(true); + + // append_op + framework::OpDesc* op = load_block->AppendOp(); + op->SetType("load"); + op->SetOutput("Out", {new_var->Name()}); + op->SetAttr("file_path", {dirname + "/" + new_var->Name()}); + op->CheckAttrs(); + } + } + executor.Run(*load_program, &scope, 0, true, true); + delete load_program; +} + +std::unique_ptr Load(framework::Executor& executor, + framework::Scope& scope, + const std::string& dirname) { + std::string model_filename = dirname + "/__model__"; + LOG(INFO) << "loading model from " << model_filename; + std::ifstream inputfs(model_filename, std::ios::in | std::ios::binary); + std::string program_desc_str; + inputfs.seekg(0, std::ios::end); + program_desc_str.resize(inputfs.tellg()); + inputfs.seekg(0, std::ios::beg); + LOG(INFO) << "program_desc_str's size: " << program_desc_str.size(); + inputfs.read(&program_desc_str[0], program_desc_str.size()); + inputfs.close(); + + std::unique_ptr main_program( + new framework::ProgramDesc(program_desc_str)); + + LoadPersistables(executor, scope, dirname, *main_program); + return main_program; +} + +} // namespace inference +} // namespace paddle diff --git a/paddle/inference/io.h b/paddle/inference/io.h new file mode 100644 index 0000000000000000000000000000000000000000..962b6c4e20d30de3cc28eae1c8c5c33b3ab5f6ac --- /dev/null +++ b/paddle/inference/io.h @@ -0,0 +1,37 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include +#include +#include +#include "paddle/framework/executor.h" +#include "paddle/framework/program_desc.h" +#include "paddle/framework/scope.h" + +namespace paddle { +namespace inference { + +void LoadPersistables(framework::Executor& executor, + framework::Scope& scope, + const std::string& dirname, + const framework::ProgramDesc& main_program); + +std::unique_ptr Load(framework::Executor& executor, + framework::Scope& scope, + const std::string& dirname); + +} // namespace inference +} // namespace paddle diff --git a/paddle/inference/tests/book/CMakeLists.txt b/paddle/inference/tests/book/CMakeLists.txt new file mode 100644 index 0000000000000000000000000000000000000000..d3798fb8fd8769aef5940d4ce724cb0cc8686422 --- /dev/null +++ b/paddle/inference/tests/book/CMakeLists.txt @@ -0,0 +1,7 @@ +set(PYTHON_TESTS_DIR ${PADDLE_SOURCE_DIR}/python/paddle/v2/fluid/tests) +cc_test(test_inference_recognize_digits_mlp + SRCS test_inference_recognize_digits.cc + DEPS ARCHIVE_START paddle_fluid ARCHIVE_END + ARGS --dirname=${PYTHON_TESTS_DIR}/book/recognize_digits_mlp.inference.model) +set_tests_properties(test_inference_recognize_digits_mlp + PROPERTIES DEPENDS test_recognize_digits_mlp_cpu) diff --git a/paddle/inference/tests/book/test_inference_recognize_digits.cc b/paddle/inference/tests/book/test_inference_recognize_digits.cc new file mode 100644 index 0000000000000000000000000000000000000000..26dc2aee04261d9a1fd29b4d75bfacc7870c09d8 --- /dev/null +++ b/paddle/inference/tests/book/test_inference_recognize_digits.cc @@ -0,0 +1,113 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include +#include +#include +#include "gflags/gflags.h" +#include "paddle/framework/lod_tensor.h" +#include "paddle/inference/io.h" + +DEFINE_string(dirname, "", "Directory of the inference model."); + +template +void TestInference(const std::string& dirname, + const std::vector& cpu_feeds, + std::vector& cpu_fetchs) { + // 1. Define place, executor and scope + auto place = Place(); + auto executor = paddle::framework::Executor(place); + auto* scope = new paddle::framework::Scope(); + + // 2. Initialize the inference_program and load all parameters from file + auto inference_program = paddle::inference::Load(executor, *scope, dirname); + + // 3. Get the feed_target_names and fetch_target_names + const std::vector& feed_target_names = + inference_program->GetFeedTargetNames(); + const std::vector& fetch_target_names = + inference_program->GetFetchTargetNames(); + + // 4. Prepare inputs: set up maps for feed targets + std::map feed_targets; + for (size_t i = 0; i < feed_target_names.size(); ++i) { + // Please make sure that cpu_feeds[i] is right for feed_target_names[i] + feed_targets[feed_target_names[i]] = cpu_feeds[i]; + } + + // 5. Define Tensor to get the outputs: set up maps for fetch targets + std::map fetch_targets; + for (size_t i = 0; i < fetch_target_names.size(); ++i) { + fetch_targets[fetch_target_names[i]] = cpu_fetchs[i]; + } + + // 6. Run the inference program + executor.Run(*inference_program, scope, feed_targets, fetch_targets); + + delete scope; +} + +TEST(inference, recognize_digits) { + if (FLAGS_dirname.empty()) { + LOG(FATAL) << "Usage: ./example --dirname=path/to/your/model"; + } + + LOG(INFO) << "FLAGS_dirname: " << FLAGS_dirname << std::endl; + std::string dirname = FLAGS_dirname; + + // 0. Call `paddle::framework::InitDevices()` initialize all the devices + // In unittests, this is done in paddle/testing/paddle_gtest_main.cc + + paddle::framework::LoDTensor input; + srand(time(0)); + float* input_ptr = + input.mutable_data({1, 28, 28}, paddle::platform::CPUPlace()); + for (int i = 0; i < 784; ++i) { + input_ptr[i] = rand() / (static_cast(RAND_MAX)); + } + std::vector cpu_feeds; + cpu_feeds.push_back(&input); + + paddle::framework::LoDTensor output1; + std::vector cpu_fetchs1; + cpu_fetchs1.push_back(&output1); + + // Run inference on CPU + TestInference( + dirname, cpu_feeds, cpu_fetchs1); + LOG(INFO) << output1.dims(); + +#ifdef PADDLE_WITH_CUDA + paddle::framework::LoDTensor output2; + std::vector cpu_fetchs2; + cpu_fetchs2.push_back(&output2); + + // Run inference on CUDA GPU + TestInference( + dirname, cpu_feeds, cpu_fetchs2); + LOG(INFO) << output2.dims(); + + EXPECT_EQ(output1.dims(), output2.dims()); + EXPECT_EQ(output1.numel(), output2.numel()); + + float err = 1E-3; + int count = 0; + for (int64_t i = 0; i < output1.numel(); ++i) { + if (fabs(output1.data()[i] - output2.data()[i]) > err) { + count++; + } + } + EXPECT_EQ(count, 0) << "There are " << count << " different elements."; +#endif +} diff --git a/paddle/memory/CMakeLists.txt b/paddle/memory/CMakeLists.txt index 8841c14ee083fccfd2271efd0c331805919a09d9..496098f80423854be62dc99b8601209ff6a6b182 100644 --- a/paddle/memory/CMakeLists.txt +++ b/paddle/memory/CMakeLists.txt @@ -1,7 +1,7 @@ add_subdirectory(detail) cc_library(memory SRCS memory.cc DEPS place enforce) -cc_library(memcpy SRCS memcpy.cc) +cc_library(memcpy SRCS memcpy.cc DEPS place) cc_library(paddle_memory DEPS @@ -14,3 +14,10 @@ cc_library(paddle_memory system_allocator) cc_test(memory_test SRCS memory_test.cc DEPS place paddle_memory) + +if(NOT WITH_C_API AND WITH_FLUID) + file(GLOB MEMORY_HEADERS *.h) + file(GLOB MEMORY_DETAIL_HEADERS detail/*.h) + install(FILES ${MEMORY_HEADERS} DESTINATION include/paddle/memory) + install(FILES ${MEMORY_DETAIL_HEADERS} DESTINATION include/paddle/memory/detail) +endif() diff --git a/paddle/operators/CMakeLists.txt b/paddle/operators/CMakeLists.txt index 2412ebd82a02c872e73fd310c56221309441f630..b2e73b6f23bd36e29be4e97237a269d12b92bd90 100644 --- a/paddle/operators/CMakeLists.txt +++ b/paddle/operators/CMakeLists.txt @@ -147,6 +147,7 @@ op_library(max_sequence_len_op DEPS lod_rank_table) op_library(sequence_conv_op DEPS context_project) op_library(sequence_pool_op DEPS sequence_pooling) op_library(lstm_op DEPS sequence2batch lstm_compute) +op_library(lstmp_op DEPS sequence2batch lstm_compute) op_library(gru_op DEPS sequence2batch gru_compute) op_library(recurrent_op DEPS executor) op_library(warpctc_op DEPS dynload_warpctc sequence_padding sequence_scale math_function) @@ -156,6 +157,7 @@ op_library(parallel_do_op DEPS executor) # Regist multiple Kernel to pybind if (WITH_GPU) op_library(conv_op SRCS conv_op.cc conv_op.cu.cc conv_cudnn_op.cu.cc DEPS vol2col) +op_library(edit_distance_op SRCS edit_distance_op.cc edit_distance_op.cu DEPS math_function) op_library(pool_op SRCS pool_op.cc pool_op.cu.cc pool_cudnn_op.cu.cc DEPS pooling) op_library(conv_transpose_op SRCS conv_transpose_op.cc conv_transpose_op.cu.cc conv_transpose_cudnn_op.cu.cc DEPS vol2col) @@ -171,6 +173,8 @@ endif() # FIXME(typhoonzero): save/load depends lodtensor serialization functions op_library(save_op DEPS lod_tensor) op_library(load_op DEPS lod_tensor) +op_library(save_combine_op DEPS lod_tensor) +op_library(load_combine_op DEPS lod_tensor) list(REMOVE_ITEM GENERAL_OPS ${DEPS_OPS}) foreach(src ${GENERAL_OPS}) @@ -178,16 +182,16 @@ foreach(src ${GENERAL_OPS}) endforeach() file(APPEND ${pybind_file} "USE_OP(less_than);\nUSE_OP(logical_and);\nUSE_NO_KERNEL_OP(read_from_array);\n") - set(GLOB_OP_LIB ${OP_LIBRARY} CACHE INTERNAL "Global OP library") - cc_test(gather_test SRCS gather_test.cc DEPS tensor) cc_test(net_op_test SRCS net_op_test.cc DEPS net_op) cc_test(scatter_test SRCS scatter_test.cc DEPS tensor) cc_test(beam_search_decode_op_test SRCS beam_search_decode_op_test.cc DEPS lod_tensor) +cc_test(beam_search_op_test SRCS beam_search_op_test.cc DEPS lod_tensor beam_search_op) cc_test(strided_memcpy_test SRCS strided_memcpy_test.cc DEPS tensor paddle_memory) if(WITH_GPU) cc_test(nccl_op_test SRCS nccl_op_test.cu.cc DEPS nccl_op gpu_info device_context) endif() cc_test(save_load_op_test SRCS save_load_op_test.cc DEPS save_op load_op) +cc_test(save_load_combine_op_test SRCS save_load_combine_op_test.cc DEPS save_combine_op load_combine_op) diff --git a/paddle/operators/activation_op.h b/paddle/operators/activation_op.h index 88c3d1c597a853abdee7753a5110be4a1726e905..c0809abc05104c1e8c1f42331c0530724dd1472f 100644 --- a/paddle/operators/activation_op.h +++ b/paddle/operators/activation_op.h @@ -323,7 +323,7 @@ template struct FloorFunctor : public BaseActivationFunctor { template void operator()(Device d, X x, Out out) const { - out.device(d) = x.ceil(); + out.device(d) = x.floor(); } }; diff --git a/paddle/operators/adagrad_op.cu b/paddle/operators/adagrad_op.cu index 4e579387924a5b0499f29609bc6b1322030a3c0d..00cb6e9cafb4e79ed3d59cd4a6e40ea132e5efda 100644 --- a/paddle/operators/adagrad_op.cu +++ b/paddle/operators/adagrad_op.cu @@ -82,7 +82,7 @@ struct SparseAdagradFunctor { math::scatter::MergeAdd merge_func; auto grad_merge = merge_func(context, grad); auto* grad_merge_data = grad_merge.mutable_value()->template data(); - auto& merge_rows = grad_merge.rows(); + framework::Vector merge_rows(grad_merge.rows()); // 2. m += g_m * g_m math::scatter::Mul sqare_func; auto grad_square = sqare_func(context, grad_merge, grad_merge); @@ -101,8 +101,8 @@ struct SparseAdagradFunctor { SparseAdagradFunctorKernel< T, 256><<(context) - .stream()>>>(grad_merge_data, grad_merge.rows().data(), - lr, param_data, moment_data, grad_width, + .stream()>>>(grad_merge_data, merge_rows.cuda_data(), lr, + param_data, moment_data, grad_width, epsilon); } }; diff --git a/paddle/operators/adam_op.h b/paddle/operators/adam_op.h index 9cc34bdded780e61e8700eb4fa4a295c84fb48bc..bf536687d398b8342e6ae76a07c11e5fe47483e0 100644 --- a/paddle/operators/adam_op.h +++ b/paddle/operators/adam_op.h @@ -199,7 +199,12 @@ class AdamOpKernel : public framework::OpKernel { merge_func(ctx.template device_context(), grad); auto& grad_tensor = grad_merge.value(); const T* grad_data = grad_tensor.template data(); - auto* rows = grad_merge.rows().data(); + int64_t* rows = nullptr; + if (platform::is_gpu_place(ctx.GetPlace())) { + rows = grad_merge.mutable_rows()->cuda_data(); + } else { + rows = grad_merge.mutable_rows()->data(); + } auto row_numel = grad_tensor.numel() / grad_merge.rows().size(); SparseAdamFunctor functor( diff --git a/paddle/operators/assign_value_op.cc b/paddle/operators/assign_value_op.cc index d5671c1183a0f58d2aedb0723bd462684ac5636e..8e3a53048920d9875f1b1a178b367cf02b2c9cf8 100644 --- a/paddle/operators/assign_value_op.cc +++ b/paddle/operators/assign_value_op.cc @@ -1,16 +1,16 @@ -/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. #include "paddle/operators/assign_value_op.h" diff --git a/paddle/operators/assign_value_op.h b/paddle/operators/assign_value_op.h index db2e43077999fa0f9aaada74026dd701ab2bf464..ec98c535132e454958e38c385f7da4df404fab50 100644 --- a/paddle/operators/assign_value_op.h +++ b/paddle/operators/assign_value_op.h @@ -1,16 +1,16 @@ -/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. - -Licensed under the Apache License, Version 2.0 (the "License"); -you may not use this file except in compliance with the License. -You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - -Unless required by applicable law or agreed to in writing, software -distributed under the License is distributed on an "AS IS" BASIS, -WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -See the License for the specific language governing permissions and -limitations under the License. */ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. #pragma once diff --git a/paddle/operators/beam_search_op.cc b/paddle/operators/beam_search_op.cc index ed2e7b738acd81140467ff22ad077155bba2fde1..844ade40eb2a7ae239b079daa609f03b9e7a06df 100644 --- a/paddle/operators/beam_search_op.cc +++ b/paddle/operators/beam_search_op.cc @@ -24,12 +24,22 @@ namespace operators { void BeamSearch::operator()(const framework::LoDTensor &pre_ids, framework::LoDTensor *selected_ids, framework::LoDTensor *selected_scores) { + auto abs_lod = framework::ToAbsOffset(ids_->lod()); + auto &high_level = abs_lod[lod_level_]; + auto items = SelectTopBeamSizeItems(); - auto selected_items = ToMap(items); + auto selected_items = ToMap(items, high_level.back()); + VLOG(3) << "selected_items:"; + for (size_t i = 0; i < selected_items.size(); ++i) { + VLOG(3) << "offset:" << i; + for (auto &item : selected_items[i]) { + VLOG(3) << ItemToString(item); + } + } PruneEndidCandidates(pre_ids, &selected_items); // calculate the output tensor's height size_t num_instances = std::accumulate( - std::begin(items), std::end(items), 0, + std::begin(selected_items), std::end(selected_items), 0, [](size_t a, std::vector &b) { return a + b.size(); }); // the output tensor shape should be [num_instances, 1] auto dims = framework::make_ddim( @@ -48,42 +58,54 @@ void BeamSearch::operator()(const framework::LoDTensor &pre_ids, size_t low_offset = 0; for (auto &items : selected_items) { low_level.push_back(low_offset); + sort(items.begin(), items.end(), [](const Item &a, const Item &b) { + if (a.offset < b.offset) { + return true; + } + return a.id < b.id; + }); for (auto &item : items) { ids_data[low_offset] = item.id; scores_data[low_offset] = item.score; low_offset++; } } + low_level.push_back(low_offset); + // fill lod - auto abs_lod = framework::ToAbsOffset(ids_->lod()); - auto &high_level = abs_lod[lod_level_]; framework::LoD lod(2); lod[0].assign(high_level.begin(), high_level.end()); lod[1].assign(low_level.begin(), low_level.end()); + if (!framework::CheckLoD(lod)) { + PADDLE_THROW("lod %s is not right", framework::LoDToString(lod)); + } selected_ids->set_lod(lod); selected_scores->set_lod(lod); } -void BeamSearch::PruneEndidCandidates(const framework::LoDTensor &pre_ids, - std::vector> *items) { +int BeamSearch::PruneEndidCandidates(const framework::LoDTensor &pre_ids, + std::vector> *items) { auto *pre_ids_data = pre_ids.data(); + int res = 0; for (size_t offset = 0; offset < items->size(); offset++) { auto prefix_id = pre_ids_data[offset]; if (prefix_id == end_id_) { items->at(offset).clear(); + } else { + res++; } } + + return res; } std::vector> BeamSearch::ToMap( - const std::vector> &items) { + const std::vector> &items, size_t element_num) { std::vector> result; + result.resize(element_num); for (auto &entries : items) { for (const auto &item : entries) { - if (item.offset >= result.size()) { - result.resize(item.offset + 1); - } result[item.offset].push_back(item); } } @@ -109,6 +131,14 @@ BeamSearch::SelectTopBeamSizeItems() { } result.emplace_back(items); } + VLOG(3) << "SelectTopBeamSizeItems result size " << result.size(); + for (auto &items : result) { + VLOG(3) << "item set:"; + for (auto &item : items) { + VLOG(3) << ItemToString(item); + } + } + return result; } @@ -121,11 +151,7 @@ bool BeamSearch::NextItemSet(std::vector *items) { auto ids = *ids_; auto scores = *scores_; - auto source_abs_two_level_lod = framework::SliceInLevel( - ids.lod(), lod_level_, sent_offset_, sent_offset_ + 1); - source_abs_two_level_lod = framework::ToAbsOffset(source_abs_two_level_lod); auto abs_lod = framework::ToAbsOffset(ids.lod()); - PADDLE_ENFORCE_GE(source_abs_two_level_lod.size(), 2UL); auto *ids_data = ids.data(); auto *scores_data = scores.data(); @@ -150,6 +176,22 @@ bool BeamSearch::NextItemSet(std::vector *items) { return true; } +std::ostream &operator<<(std::ostream &os, const BeamSearch::Item &item) { + os << "{"; + os << "offset: " << item.offset << ", "; + os << "id: " << item.id << ", "; + os << "score: " << item.score << ""; + os << "}"; + + return os; +} + +std::string ItemToString(const BeamSearch::Item &item) { + std::ostringstream stream; + stream << item; + return stream.str(); +} + class BeamSearchProtoAndCheckerMaker : public framework::OpProtoAndCheckerMaker { public: @@ -177,8 +219,40 @@ class BeamSearchProtoAndCheckerMaker } }; +class BeamSearchInferShape : public framework::InferShapeBase { + public: + void operator()(framework::InferShapeContext *context) const override { + for (const std::string &arg : + std::vector({"pre_ids", "ids", "scores"})) { + PADDLE_ENFORCE(context->HasInput(arg), + "BeamSearch need input argument '%s'", arg); + } + for (const std::string &arg : + std::vector({"selected_ids", "selected_scores"})) { + PADDLE_ENFORCE(context->HasOutput(arg), + "BeamSearch need output argument '%s'", arg); + } + } +}; + +class BeamSearchInferVarType : public framework::VarTypeInference { + public: + void operator()(const framework::OpDesc &op_desc, + framework::BlockDesc *block) const override { + for (auto &o : op_desc.Output("selected_ids")) { + block->Var(o)->SetType(framework::proto::VarDesc::LOD_TENSOR); + } + for (auto &o : op_desc.Output("selected_scores")) { + block->Var(o)->SetType(framework::proto::VarDesc::LOD_TENSOR); + } + } +}; + } // namespace operators } // namespace paddle -REGISTER_OP_WITHOUT_GRADIENT(beam_search, paddle::operators::BeamSearchOp, - paddle::operators::BeamSearchProtoAndCheckerMaker); +REGISTER_OPERATOR(beam_search, paddle::operators::BeamSearchOp, + paddle::operators::BeamSearchProtoAndCheckerMaker, + paddle::operators::BeamSearchInferShape, + paddle::operators::BeamSearchInferVarType, + paddle::framework::EmptyGradOpMaker); diff --git a/paddle/operators/beam_search_op.h b/paddle/operators/beam_search_op.h index 08b551ef9bd63106ed222d3a956a912294f827ec..7ad85874fcbd6ea48d688b32f2cc982d6b76d3c4 100644 --- a/paddle/operators/beam_search_op.h +++ b/paddle/operators/beam_search_op.h @@ -73,7 +73,15 @@ namespace operators { * second level: * [0, 2, 4] * - * tensor's data + * id tensor's data + * [[ + * 4, + * 1, + * 3, + * 8, + * ]] + * + * score tensor's data * [[ * 0.5, * 0.3, @@ -128,8 +136,6 @@ class BeamSearch { void operator()(const framework::LoDTensor& pre_ids, framework::LoDTensor* selected_ids, framework::LoDTensor* selected_scores); - - protected: /* * The basic items help to sort. */ @@ -137,23 +143,29 @@ class BeamSearch { Item() {} Item(size_t offset, size_t id, float score) : offset(offset), id(id), score(score) {} - // offset in the lod_level_+1 + // offset in the higher lod level. size_t offset; + // // prefix id in the lower lod level. + // size_t prefix; // the candidate id id_t id; // the corresponding score score_t score; }; - void PruneEndidCandidates(const framework::LoDTensor& pre_ids, - std::vector>* items); + protected: + /* + * Delete all the records that follows the end token. + */ + int PruneEndidCandidates(const framework::LoDTensor& pre_ids, + std::vector>* items); /* * Transform the items into a map whose key is offset, value is the items. * NOTE low performance */ std::vector> ToMap( - const std::vector>& inputs); + const std::vector>& inputs, size_t element_num); /* * For each source, select top beam_size records. @@ -174,6 +186,10 @@ class BeamSearch { int end_id_{0}; }; +std::ostream& operator<<(std::ostream& os, const BeamSearch::Item& item); + +std::string ItemToString(const BeamSearch::Item& item); + class BeamSearchOp : public framework::OperatorBase { public: BeamSearchOp(const std::string& type, @@ -190,7 +206,6 @@ class BeamSearchOp : public framework::OperatorBase { void Run(const framework::Scope& scope, const platform::Place& dev_place) const override { - LOG(INFO) << "run beam search op"; auto ids_var = scope.FindVar(Input("ids")); auto scores_var = scope.FindVar(Input("scores")); auto pre_ids_var = scope.FindVar(Input("pre_ids")); @@ -204,10 +219,8 @@ class BeamSearchOp : public framework::OperatorBase { size_t level = Attr("level"); size_t beam_size = Attr("beam_size"); int end_id = Attr("end_id"); - LOG(INFO) << "init beam search"; BeamSearch alg(ids, scores, level, beam_size, end_id); - LOG(INFO) << "after beam search"; auto selected_ids_var = scope.FindVar(Output("selected_ids")); auto selected_scores_var = scope.FindVar(Output("selected_scores")); PADDLE_ENFORCE_NOT_NULL(selected_ids_var); @@ -216,9 +229,7 @@ class BeamSearchOp : public framework::OperatorBase { *selected_ids_var->GetMutable(); auto& selected_scores_tensor = *selected_scores_var->GetMutable(); - LOG(INFO) << "run beam search"; alg(pre_ids, &selected_ids_tensor, &selected_scores_tensor); - LOG(INFO) << "finish beam search"; } }; diff --git a/paddle/operators/beam_search_op_test.cc b/paddle/operators/beam_search_op_test.cc new file mode 100644 index 0000000000000000000000000000000000000000..d4beb64a85a1645b3fed22c3325bd8c0b7cd12b1 --- /dev/null +++ b/paddle/operators/beam_search_op_test.cc @@ -0,0 +1,86 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. */ + +#include "paddle/operators/beam_search_op.h" + +#include +#include + +namespace paddle { +namespace test { + +using std::vector; +using framework::LoDTensor; +using framework::LoD; +using operators::BeamSearch; +using paddle::platform::CPUPlace; +using std::cout; +using std::endl; + +void CreateInput(LoDTensor* ids, LoDTensor* scores) { + LoD lod; + vector level0({0, 1, 4}); + vector level1({0, 1, 2, 3, 4}); + lod.push_back(level0); + lod.push_back(level1); + ids->set_lod(lod); + scores->set_lod(lod); + + auto dims = framework::make_ddim(vector({4, 3})); + ids->Resize(dims); + scores->Resize(dims); + CPUPlace place; + + auto* ids_data = ids->mutable_data(place); + auto* scores_data = scores->mutable_data(place); + vector _ids({4, 2, 5, 2, 1, 3, 3, 5, 2, 8, 2, 1}); + vector _scores( + {0.5, 0.3, 0.2, 0.6, 0.3, 0.1, 0.9, 0.5, 0.1, 0.7, 0.5, 0.1}); + + for (int i = 0; i < 12; i++) { + ids_data[i] = _ids[i]; + scores_data[i] = _scores[i]; + } +} + +TEST(beam_search_op, run) { + CPUPlace place; + LoDTensor ids, scores; + CreateInput(&ids, &scores); + + LoDTensor pre_ids; + pre_ids.Resize(framework::make_ddim(vector(4, 1))); + for (int i = 0; i < 4; i++) { + pre_ids.mutable_data(place)[i] = i + 1; + } + + BeamSearch beamsearch(ids, scores, (int64_t)0, (int64_t)2, 0); + LoDTensor sids, sscores; + beamsearch(pre_ids, &sids, &sscores); + + LOG(INFO) << "score: " << sscores << endl; + + ASSERT_EQ(sids.lod(), sscores.lod()); + + vector tids({2, 4, 3, 8}); + vector tscores({0.3, 0.5, 0.9, 0.7}); + + for (int i = 0; i < 4; i++) { + ASSERT_EQ(tids[i], sids.data()[i]); + ASSERT_EQ(tscores[i], sscores.data()[i]); + } +} + +} // namespace test +} // namespace paddle diff --git a/paddle/operators/bipartite_match_op.cc b/paddle/operators/bipartite_match_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..83c8778fe4cec4d9d80de691e117a39fdd92f494 --- /dev/null +++ b/paddle/operators/bipartite_match_op.cc @@ -0,0 +1,189 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + +http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/framework/op_registry.h" +#include "paddle/operators/math/math_function.h" + +namespace paddle { +namespace operators { + +using Tensor = framework::Tensor; +using LoDTensor = framework::LoDTensor; + +class BipartiteMatchOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("DistMat"), + "Input(DistMat) of BipartiteMatch should not be null."); + + auto dims = ctx->GetInputDim("DistMat"); + PADDLE_ENFORCE_EQ(dims.size(), 2, "The rank of Input(DistMat) must be 2."); + + ctx->SetOutputDim("ColToRowMatchIndices", dims); + ctx->SetOutputDim("ColToRowMatchDis", dims); + } +}; + +template +class BipartiteMatchKernel : public framework::OpKernel { + public: + // The match_indices must be initialized to -1 at first. + // The match_dist must be initialized to 0 at first. + void BipartiteMatch(const Tensor& dist, int* match_indices, + T* match_dist) const { + constexpr T kEPS = static_cast(1e-6); + PADDLE_ENFORCE_EQ(dist.dims().size(), 2, "The rank of dist must be 2."); + int64_t row = dist.dims()[0]; + int64_t col = dist.dims()[1]; + auto* dist_data = dist.data(); + std::vector row_pool; + for (int i = 0; i < row; ++i) { + row_pool.push_back(i); + } + while (row_pool.size() > 0) { + int max_idx = -1; + int max_row_idx = -1; + T max_dist = -1; + for (int64_t j = 0; j < col; ++j) { + if (match_indices[j] != -1) { + continue; + } + for (size_t k = 0; k < row_pool.size(); ++k) { + int m = row_pool[k]; + // distance is 0 between m-th row and j-th column + if (dist_data[m * col + j] < kEPS) { + continue; + } + if (dist_data[m * col + j] > max_dist) { + max_idx = j; + max_row_idx = m; + max_dist = dist_data[m * col + j]; + } + } + } + if (max_idx == -1) { + // Cannot find good match. + break; + } else { + PADDLE_ENFORCE_EQ(match_indices[max_idx], -1); + match_indices[max_idx] = max_row_idx; + match_dist[max_idx] = max_dist; + // Erase the row index. + row_pool.erase( + std::find(row_pool.begin(), row_pool.end(), max_row_idx)); + } + } + } + + void Compute(const framework::ExecutionContext& context) const override { + auto* dist_mat = context.Input("DistMat"); + auto* match_indices = context.Output("ColToRowMatchIndices"); + auto* match_dist = context.Output("ColToRowMatchDis"); + + auto& dev_ctx = context.device_context(); + + auto col = dist_mat->dims()[1]; + + int64_t n = dist_mat->lod().size() == 0UL + ? 1 + : static_cast(dist_mat->lod().back().size() - 1); + if (dist_mat->lod().size()) { + PADDLE_ENFORCE_EQ(dist_mat->lod().size(), 1UL, + "Only support 1 level of LoD."); + } + match_indices->mutable_data({n, col}, context.GetPlace()); + match_dist->mutable_data({n, col}, context.GetPlace()); + + math::SetConstant iset; + iset(dev_ctx, match_indices, static_cast(-1)); + math::SetConstant tset; + tset(dev_ctx, match_dist, static_cast(0)); + + int* indices = match_indices->data(); + T* dist = match_dist->data(); + if (n == 1) { + BipartiteMatch(*dist_mat, indices, dist); + } else { + auto lod = dist_mat->lod().back(); + for (size_t i = 0; i < lod.size() - 1; ++i) { + Tensor one_ins = dist_mat->Slice(lod[i], lod[i + 1]); + BipartiteMatch(one_ins, indices + i * col, dist + i * col); + } + } + } +}; + +class BipartiteMatchOpMaker : public framework::OpProtoAndCheckerMaker { + public: + BipartiteMatchOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput( + "DistMat", + "(LoDTensor or Tensor) this input is a 2-D LoDTensor with shape " + "[K, M]. It is pair-wise distance matrix between the entities " + "represented by each row and each column. For example, assumed one " + "entity is A with shape [K], another entity is B with shape [M]. The " + "DistMat[i][j] is the distance between A[i] and B[j]. The bigger " + "the distance is, the better macthing the pairs are. Please note, " + "This tensor can contain LoD information to represent a batch of " + "inputs. One instance of this batch can contain different numbers of " + "entities."); + AddOutput("ColToRowMatchIndices", + "(Tensor) A 2-D Tensor with shape [N, M] in int type. " + "N is the batch size. If ColToRowMatchIndices[i][j] is -1, it " + "means B[j] does not match any entity in i-th instance. " + "Otherwise, it means B[j] is matched to row " + "ColToRowMatchIndices[i][j] in i-th instance. The row number of " + "i-th instance is saved in ColToRowMatchIndices[i][j]."); + AddOutput("ColToRowMatchDis", + "(Tensor) A 2-D Tensor with shape [N, M] in float type. " + "N is batch size. If ColToRowMatchIndices[i][j] is -1, " + "ColToRowMatchDis[i][j] is also -1.0. Otherwise, assumed " + "ColToRowMatchIndices[i][j] = d, and the row offsets of each " + "instance are called LoD. Then " + "ColToRowMatchDis[i][j] = DistMat[d+LoD[i]][j]"); + AddComment(R"DOC( +This operator is a greedy bipartite matching algorithm, which is used to +obtain the matching with the maximum distance based on the input +distance matrix. For input 2D matrix, the bipartite matching algorithm can +find the matched column for each row, also can find the matched row for +each column. And this operator only calculate matched indices from column +to row. For each instance, the number of matched indices is the number of +of columns of the input ditance matrix. + +There are two outputs to save matched indices and distance. +A simple description, this algothrim matched the best (maximum distance) +row entity to the column entity and the matched indices are not duplicated +in each row of ColToRowMatchIndices. If the column entity is not matched +any row entity, set -1 in ColToRowMatchIndices. + +Please note that the input DistMat can be LoDTensor (with LoD) or Tensor. +If LoDTensor with LoD, the height of ColToRowMatchIndices is batch size. +If Tensor, the height of ColToRowMatchIndices is 1. + +)DOC"); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OPERATOR(bipartite_match, ops::BipartiteMatchOp, + ops::BipartiteMatchOpMaker, + paddle::framework::EmptyGradOpMaker); +REGISTER_OP_CPU_KERNEL(bipartite_match, ops::BipartiteMatchKernel, + ops::BipartiteMatchKernel); diff --git a/paddle/operators/clip_op.cc b/paddle/operators/clip_op.cc index 573bb9c7dfdac2366c2458dd9f27a035a9f9b813..7adb74eab78dcdd0251b8db60781f6e24e348634 100644 --- a/paddle/operators/clip_op.cc +++ b/paddle/operators/clip_op.cc @@ -51,8 +51,8 @@ class ClipOpMaker : public framework::OpProtoAndCheckerMaker { AddComment(R"DOC( Clip Operator. -The clip operator limits the value of given input within an interval. The interval is -specified with arguments 'min' and 'max': +The clip operator limits the value of given input within an interval. The +interval is specified with arguments 'min' and 'max': $$ Out = \min(\max(X, min), max) diff --git a/paddle/operators/compare_op.cc b/paddle/operators/compare_op.cc index daa2c193b48fe216ff284169a3dce1b4cd40a791..930c295a9cb31238954efeb87ff5ac2d3ca7bdc6 100644 --- a/paddle/operators/compare_op.cc +++ b/paddle/operators/compare_op.cc @@ -39,6 +39,11 @@ N-dim tensor. X and Y could be any type. The each element of the Out tensor is calculated by %s )DOC", comment.type, comment.equation)); + AddAttr("axis", + "(int, default -1). The start dimension index " + "for broadcasting Y onto X.") + .SetDefault(-1) + .EqualGreaterThan(-1); } }; @@ -95,11 +100,5 @@ REGISTER_LOGICAL_OP(less_than, "Out = X < Y"); REGISTER_LOGICAL_KERNEL(less_than, CPU, paddle::operators::LessThanFunctor); REGISTER_LOGICAL_OP(less_equal, "Out = X <= Y"); REGISTER_LOGICAL_KERNEL(less_equal, CPU, paddle::operators::LessEqualFunctor); -REGISTER_LOGICAL_OP(greater_than, "Out = X > Y"); -REGISTER_LOGICAL_KERNEL(greater_than, CPU, - paddle::operators::GreaterThanFunctor); -REGISTER_LOGICAL_OP(greater_equal, "Out = X >= Y"); -REGISTER_LOGICAL_KERNEL(greater_equal, CPU, - paddle::operators::GreaterEqualFunctor); REGISTER_LOGICAL_OP(equal, "Out = X == Y"); REGISTER_LOGICAL_KERNEL(equal, CPU, paddle::operators::EqualFunctor); diff --git a/paddle/operators/compare_op.cu b/paddle/operators/compare_op.cu index 26049271befd1fe57001659d1a406e73de0004a7..f625824dbc99d603f1e92700b4ad3d7fa25b471d 100644 --- a/paddle/operators/compare_op.cu +++ b/paddle/operators/compare_op.cu @@ -16,8 +16,4 @@ limitations under the License. */ REGISTER_LOGICAL_KERNEL(less_than, CUDA, paddle::operators::LessThanFunctor); REGISTER_LOGICAL_KERNEL(less_equal, CUDA, paddle::operators::LessEqualFunctor); -REGISTER_LOGICAL_KERNEL(greater_than, CUDA, - paddle::operators::GreaterThanFunctor); -REGISTER_LOGICAL_KERNEL(greater_equal, CUDA, - paddle::operators::GreaterEqualFunctor); REGISTER_LOGICAL_KERNEL(equal, CUDA, paddle::operators::EqualFunctor); diff --git a/paddle/operators/compare_op.h b/paddle/operators/compare_op.h index 567e89c0a727ad0cdd2add8ec8b2a42c86a58007..9c655d6c0d8e5fe04ee6d85f7e9d9da68105230c 100644 --- a/paddle/operators/compare_op.h +++ b/paddle/operators/compare_op.h @@ -16,6 +16,7 @@ limitations under the License. */ #include #include #include "paddle/framework/op_registry.h" +#include "paddle/operators/elementwise_op_function.h" #include "paddle/platform/transform.h" namespace paddle { @@ -33,18 +34,6 @@ struct LessEqualFunctor { HOSTDEVICE bool operator()(const T& a, const T& b) const { return a <= b; } }; -template -struct GreaterThanFunctor { - using ELEM_TYPE = T; - HOSTDEVICE bool operator()(const T& a, const T& b) const { return a > b; } -}; - -template -struct GreaterEqualFunctor { - using ELEM_TYPE = T; - HOSTDEVICE bool operator()(const T& a, const T& b) const { return a >= b; } -}; - template struct EqualFunctor { using ELEM_TYPE = T; @@ -65,14 +54,7 @@ class CompareOpKernel public: void Compute(const framework::ExecutionContext& context) const override { using T = typename Functor::ELEM_TYPE; - auto* x = context.Input("X"); - auto* y = context.Input("Y"); - auto* out = context.Output("Out"); - Functor binary_func; - platform::Transform trans; - trans(context.template device_context(), x->data(), - x->data() + x->numel(), y->data(), - out->mutable_data(context.GetPlace()), binary_func); + ElementwiseComputeEx(context); } }; diff --git a/paddle/operators/conv_op.cc b/paddle/operators/conv_op.cc index 424eccdb7dc57195d20f75460032a99c76e6adcd..d6882b275b22b9a2a2b6ff8cfb53a3462bbdbefe 100644 --- a/paddle/operators/conv_op.cc +++ b/paddle/operators/conv_op.cc @@ -70,6 +70,13 @@ void ConvOp::InferShape(framework::InferShapeContext* ctx) const { framework::OpKernelType ConvOp::GetExpectedKernelType( const framework::ExecutionContext& ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); + use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); +#ifdef PADDLE_WITH_CUDA + if (platform::is_gpu_place(ctx.GetPlace())) { + auto& dev_ctx = ctx.template device_context(); + use_cudnn &= dev_ctx.cudnn_handle() != nullptr; + } +#endif framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; @@ -283,6 +290,14 @@ void ConvOpGrad::InferShape(framework::InferShapeContext* ctx) const { framework::OpKernelType ConvOpGrad::GetExpectedKernelType( const framework::ExecutionContext& ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); + use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); +#ifdef PADDLE_WITH_CUDA + if (platform::is_gpu_place(ctx.GetPlace())) { + auto& dev_ctx = ctx.template device_context(); + use_cudnn &= dev_ctx.cudnn_handle() != nullptr; + } +#endif + framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; diff --git a/paddle/operators/conv_transpose_op.cc b/paddle/operators/conv_transpose_op.cc index cf4e8c0a303d6888c3b1f2475a483c4bfc90981b..089290a506db10f676c8d7eb92663d2cb56892af 100644 --- a/paddle/operators/conv_transpose_op.cc +++ b/paddle/operators/conv_transpose_op.cc @@ -61,6 +61,13 @@ void ConvTransposeOp::InferShape(framework::InferShapeContext* ctx) const { framework::OpKernelType ConvTransposeOp::GetExpectedKernelType( const framework::ExecutionContext& ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); + use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); +#ifdef PADDLE_WITH_CUDA + if (platform::is_gpu_place(ctx.GetPlace())) { + auto& dev_ctx = ctx.template device_context(); + use_cudnn &= dev_ctx.cudnn_handle() != nullptr; + } +#endif framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; @@ -153,8 +160,8 @@ Example: Output shape: $(N, C_{out}, H_{out}, W_{out})$ Where $$ - H_{out} = (H_{in} - 1) * strides[0] - 2 * paddings[0] + H_f \\ - W_{out} = (W_{in} - 1) * strides[1] - 2 * paddings[1] + W_f + H_{out} = (H_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (H_f - 1) + 1 \\ + W_{out} = (W_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (W_f - 1) + 1 $$ )DOC"); } @@ -242,9 +249,9 @@ Example: Output shape: $(N, C_{out}, D_{out}, H_{out}, W_{out})$ Where $$ - D_{out} = (D_{in} - 1) * strides[0] - 2 * paddings[0] + D_f \\ - H_{out} = (H_{in} - 1) * strides[1] - 2 * paddings[1] + H_f \\ - W_{out} = (W_{in} - 1) * strides[2] - 2 * paddings[2] + W_f + D_{out} = (D_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (D_f - 1) + 1 \\ + H_{out} = (H_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (H_f - 1) + 1 \\ + W_{out} = (W_{in} - 1) * strides[2] - 2 * paddings[2] + dilations[2] * (W_f - 1) + 1 $$ )DOC"); } @@ -263,6 +270,13 @@ void ConvTransposeOpGrad::InferShape(framework::InferShapeContext* ctx) const { framework::OpKernelType ConvTransposeOpGrad::GetExpectedKernelType( const framework::ExecutionContext& ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); + use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); +#ifdef PADDLE_WITH_CUDA + if (platform::is_gpu_place(ctx.GetPlace())) { + auto& dev_ctx = ctx.template device_context(); + use_cudnn &= dev_ctx.cudnn_handle() != nullptr; + } +#endif framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; diff --git a/paddle/operators/conv_transpose_op.h b/paddle/operators/conv_transpose_op.h index a42ade41b165d1bfa00d2db0e45d40cf5d7b00bc..8c0d57afcd21d8622fb6316f7b988d79a45b57fe 100644 --- a/paddle/operators/conv_transpose_op.h +++ b/paddle/operators/conv_transpose_op.h @@ -141,9 +141,9 @@ class GemmConvTransposeKernel : public framework::OpKernel { if (data_dim == 2U) { // col2im: col_matrix -> dy // from (c * k_h * k_w, h * w) to (c, o_h, o_w) - col2im(dev_ctx, col, std::vector{dilations[0], dilations[1]}, - strides, std::vector{paddings[0], paddings[1], paddings[0], - paddings[1]}, + col2im(dev_ctx, col, dilations, strides, + std::vector{paddings[0], paddings[1], paddings[0], + paddings[1]}, &output_batch); } else if (data_dim == 3U) { // col2vol: col_matrix -> dy @@ -247,8 +247,7 @@ class GemmConvTransposeGradKernel : public framework::OpKernel { if (data_dim == 2U) { // im2col: dy -> col matrix // from (c, o_h, o_w) to (c * k_h * k_w, h * w) - im2col(dev_ctx, output_grad_batch, - std::vector{dilations[0], dilations[1]}, strides, + im2col(dev_ctx, output_grad_batch, dilations, strides, std::vector{paddings[0], paddings[1], paddings[0], paddings[1]}, &col); diff --git a/paddle/operators/ctc_align_op.cc b/paddle/operators/ctc_align_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..eeecbd32127d2cf9756432817fc5d36673685aa7 --- /dev/null +++ b/paddle/operators/ctc_align_op.cc @@ -0,0 +1,93 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/ctc_align_op.h" + +namespace paddle { +namespace operators { + +class CTCAlignOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("Input"), + "Input of CTCAlignOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Output"), + "Output of CTCAlignOp should not be null."); + + auto input_dims = ctx->GetInputDim("Input"); + + // TODO(wanghaoshuang): it is tricky to set the wrong dimension here. + ctx->SetOutputDim("Output", input_dims); + } + + protected: + framework::OpKernelType GetExpectedKernelType( + const framework::ExecutionContext& ctx) const override { + return framework::OpKernelType( + framework::ToDataType(ctx.Input("Input")->type()), + ctx.device_context()); + } +}; + +class CTCAlignOpMaker : public framework::OpProtoAndCheckerMaker { + public: + CTCAlignOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("Input", + "(LodTensor, default: LoDTensor), Its shape is " + "[Lp, 1], where Lp is the sum of all input sequences' length."); + AddOutput("Output", "(Tensor, default: Tensor), The align result."); + AddAttr("blank", + "(int, default: 0), the blank label setted in Connectionist " + "Temporal Classification (CTC) op.") + .SetDefault(0); + AddAttr("merge_repeated", + "(bool, default: true), whether to " + "merge repeated elements between two blanks. ") + .SetDefault(true); + AddComment(R"DOC( +CTCAlign op is used to merge repeated elements between two blanks +and then delete all blanks in sequence. + +Given: + Input.data = [0, 1, 2, 2, 0, 4, 0, 4, 5, 0, 6, + 6, 0, 0, 7, 7, 7, 0] + Input.dims = {18, 1} + Input.LoD = [[0, 11, 18]] + +And: + blank = 0 + merge_repeated = True + +Then: + Output.data = [1, 2, 4, 4, 5, 6, + 6, 7] + Output.dims = {8, 1} + Output.LoD = [[0, 6, 8]] + +)DOC"); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OPERATOR(ctc_align, ops::CTCAlignOp, ops::CTCAlignOpMaker, + paddle::framework::EmptyGradOpMaker); +REGISTER_OP_CPU_KERNEL( + ctc_align, ops::CTCAlignKernel, + ops::CTCAlignKernel); diff --git a/paddle/operators/ctc_align_op.cu b/paddle/operators/ctc_align_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..2a970cd9fa965b4126356eaa1519068f9c7a7f34 --- /dev/null +++ b/paddle/operators/ctc_align_op.cu @@ -0,0 +1,90 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include +#include +#include +#include "paddle/operators/ctc_align_op.h" + +namespace paddle { +namespace operators { + +template +__global__ void MergeAndDelCudaKernel(const int64_t num_token, const T* tokens, + const size_t num_seq, size_t* lod0, + const int blank, const int merge_repeated, + size_t* out_lod0, T* output) { + int ouput_idx = 0; + out_lod0[0] = 0; + + for (int i = 0; i < num_seq; ++i) { + T pre_token = -1; + for (int j = lod0[i]; j < lod0[i + 1]; ++j) { + if (tokens[j] != blank && !(merge_repeated && tokens[j] == pre_token)) { + output[ouput_idx] = tokens[j]; + ++ouput_idx; + } + pre_token = tokens[j]; + } + out_lod0[i + 1] = ouput_idx; + } +} + +template +class CTCAlignOpCUDAKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + PADDLE_ENFORCE(platform::is_gpu_place(ctx.GetPlace()), + "It must use CUDAPlace."); + const size_t level = 0; + auto* input = ctx.Input("Input"); + auto* output = ctx.Output("Output"); + auto input_lod = framework::ToAbsOffset(input->lod()); + + const T* tokens = input->data(); + const int64_t num_tokens = input->dims()[0]; + const size_t num_seq = input_lod[level].size() - 1; + + const int blank = ctx.Attr("blank"); + const int merge_repeated = + static_cast(ctx.Attr("merge_repeated")); + + // prepare a lod to record lod information while merging elements + thrust::device_vector dev_out_lod0(input_lod[level].size()); + size_t* dev_out_lod0_ptr = thrust::raw_pointer_cast(dev_out_lod0.data()); + + // merge elements and delete blank + T* output_data = output->mutable_data({num_tokens, 1}, ctx.GetPlace()); + + auto stream = ctx.cuda_device_context().stream(); + MergeAndDelCudaKernel<<<1, 1, 0, stream>>>( + num_tokens, tokens, num_seq, input_lod[level].cuda_data(), blank, + merge_repeated, dev_out_lod0_ptr, output_data); + + // set output lod + std::vector host_out_lod0(dev_out_lod0.begin(), dev_out_lod0.end()); + framework::LoD out_lod; + out_lod.push_back(host_out_lod0); + output->set_lod(out_lod); + + // resize output dims + output->Resize({static_cast(host_out_lod0.back()), 1}); + } +}; + +} // namespace operators +} // namespace paddle + +REGISTER_OP_CUDA_KERNEL(ctc_align, paddle::operators::CTCAlignOpCUDAKernel, + paddle::operators::CTCAlignOpCUDAKernel); diff --git a/paddle/operators/ctc_align_op.h b/paddle/operators/ctc_align_op.h new file mode 100644 index 0000000000000000000000000000000000000000..fed89aa1e899a2450b315f352b9695056ed13aec --- /dev/null +++ b/paddle/operators/ctc_align_op.h @@ -0,0 +1,75 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include +#include "paddle/framework/op_registry.h" +namespace paddle { +namespace operators { + +using Tensor = framework::Tensor; +using LoDTensor = framework::LoDTensor; + +template +class CTCAlignKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + auto* input = ctx.Input("Input"); + auto* output = ctx.Output("Output"); + const size_t level = 0; + auto input_lod = framework::ToAbsOffset(input->lod()); + + // check input dims and lod + auto input_dims = input->dims(); + PADDLE_ENFORCE_EQ(input_dims[0], + static_cast(input_lod[level].back()), + "The first dimension of Input(Input) should be equal to " + "the sum of all sequences' lengths."); + + const size_t num_sequences = input_lod[level].size() - 1; + size_t blank = static_cast(ctx.Attr("blank")); + bool merge_repeated = ctx.Attr("merge_repeated"); + + // merge repeated tokens and delete blank + T* output_data = output->mutable_data(ctx.GetPlace()); + size_t output_idx = 0; + std::vector output_lod0(1, 0); + const T* input_data = input->data(); + for (size_t seq_idx = 0; seq_idx < num_sequences; ++seq_idx) { + T prev_token = -1; + for (size_t i = input_lod[level][seq_idx]; + i < input_lod[level][seq_idx + 1]; ++i) { + if ((unsigned)input_data[i] != blank && + !(merge_repeated && input_data[i] == prev_token)) { + output_data[output_idx] = input_data[i]; + ++output_idx; + } + prev_token = input_data[i]; + } + output_lod0.push_back(output_idx); + } + + // set output lod + framework::LoD output_lod; + output_lod.push_back(output_lod0); + output->set_lod(output_lod); + + // resize output dims + output->Resize({static_cast(output_lod0.back()), 1}); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/detail/grpc_client.cc b/paddle/operators/detail/grpc_client.cc index aee56ffe018aa8d0d2106df24bd9358c930a02ca..9b5f7afc6a48f13ff999f635efeb9e7bf0a76fb5 100644 --- a/paddle/operators/detail/grpc_client.cc +++ b/paddle/operators/detail/grpc_client.cc @@ -13,6 +13,7 @@ See the License for the specific language governing permissions and limitations under the License. */ #include "grpc_client.h" +#include "paddle/framework/threadpool.h" namespace paddle { namespace operators { namespace detail { @@ -22,25 +23,32 @@ bool RPCClient::AsyncSendVariable(const std::string& ep, const framework::Scope& scope, const std::string& var_name, int64_t time_out) { - sendrecv::VariableMessage req; - auto* var = scope.FindVar(var_name); - SerializeToMessage(var_name, var, ctx, &req); - - // varhandle - VarHandle var_h; - var_h.ep = ep; - var_h.scope = &scope; - var_h.name = var_name; - var_h.ctx = &ctx; - - // stub context - auto ch = GetChannel(ep); - SendProcessor* s = new SendProcessor(ch); - s->Prepare(var_h, time_out); - s->response_call_back_ = NULL; - - auto rpc = s->stub_->AsyncSendVariable(s->context_.get(), req, &cq_); - rpc->Finish(&s->reply_, &s->status_, (void*)s); + const platform::DeviceContext* p_ctx = &ctx; + const std::string ep_val = ep; + const std::string var_name_val = var_name; + const framework::Scope* p_scope = &scope; + const auto ch = GetChannel(ep_val); + + framework::Async([var_name_val, p_ctx, ep_val, p_scope, time_out, ch, this] { + auto* var = p_scope->FindVar(var_name_val); + sendrecv::VariableMessage req; + SerializeToMessage(var_name_val, var, *p_ctx, &req); + + // varhandle + VarHandle var_h; + var_h.ep = ep_val; + var_h.scope = p_scope; + var_h.name = var_name_val; + var_h.ctx = p_ctx; + + // stub context + SendProcessor* s = new SendProcessor(ch); + s->Prepare(var_h, time_out); + s->response_call_back_ = NULL; + + auto rpc = s->stub_->AsyncSendVariable(s->context_.get(), req, &cq_); + rpc->Finish(&s->reply_, &s->status_, (void*)s); + }); req_count_++; @@ -50,8 +58,6 @@ bool RPCClient::AsyncSendVariable(const std::string& ep, void ProcGetResponse(const VarHandle& var_h, const sendrecv::VariableMessage& ret_msg) { auto* outvar = var_h.scope->FindVar(var_h.name); - - std::istringstream iss(ret_msg.serialized()); DeserializeFromMessage(ret_msg, *var_h.ctx, outvar); } @@ -60,47 +66,78 @@ bool RPCClient::AsyncGetVariable(const std::string& ep, const framework::Scope& scope, const std::string& var_name, int64_t time_out) { - sendrecv::VariableMessage req; - req.set_varname(var_name); + const platform::DeviceContext* p_ctx = &ctx; + const std::string ep_val = ep; + const std::string var_name_val = var_name; + const framework::Scope* p_scope = &scope; + const auto ch = GetChannel(ep_val); + + framework::Async([var_name_val, ep_val, p_scope, p_ctx, time_out, ch, this] { + sendrecv::VariableMessage req; + req.set_varname(var_name_val); + + // varhandle + VarHandle var_h; + var_h.ep = ep_val; + var_h.scope = p_scope; + var_h.name = var_name_val; + var_h.ctx = p_ctx; + + // stub context + GetProcessor* s = new GetProcessor(ch); + s->Prepare(var_h, time_out); + s->response_call_back_ = ProcGetResponse; + + auto rpc = s->stub_->AsyncGetVariable(s->context_.get(), req, &cq_); + rpc->Finish(&s->reply_, &s->status_, (void*)s); + }); - auto* var = scope.FindVar(var_name); - SerializeToMessage(var_name, var, ctx, &req); + req_count_++; - // varhandle - VarHandle var_h; - var_h.ep = ep; - var_h.scope = &scope; - var_h.name = var_name; - var_h.ctx = &ctx; + return true; +} - // stub context - auto ch = GetChannel(ep); - GetProcessor* s = new GetProcessor(ch); - s->Prepare(var_h, time_out); - s->response_call_back_ = ProcGetResponse; +bool RPCClient::AsyncSendBatchBarrier(const std::string& ep, int64_t time_out) { + const auto ch = GetChannel(ep); - auto rpc = s->stub_->AsyncGetVariable(s->context_.get(), req, &cq_); - rpc->Finish(&s->reply_, &s->status_, (void*)s); + BatchBarrierProcessor* s = new BatchBarrierProcessor(ch); + s->Prepare(time_out); + sendrecv::VariableMessage req; + req.set_varname(BATCH_BARRIER_MESSAGE); + auto rpc = s->stub_->AsyncSendVariable(s->context_.get(), req, &cq_); + rpc->Finish(&s->reply_, &s->status_, (void*)s); req_count_++; return true; } bool RPCClient::Wait() { - bool ok = true; + if (req_count_ <= 0) { + return true; + } + const size_t kReqCnt = req_count_; + bool a[kReqCnt]; + std::vector> waits(req_count_); - while (true) { - if (req_count_ <= 0) { - break; - } + for (int i = 0; i < req_count_; i++) { + waits[i] = framework::Async([i, &a, this] { a[i] = Proceed(); }); + } + + for (int i = 0; i < req_count_; i++) { + waits[i].wait(); + } - if (!Proceed()) { + int last_req_count = req_count_; + req_count_ = 0; + + for (int i = 0; i < last_req_count; i++) { + if (!a[i]) { return false; } } - return ok; + return true; } bool RPCClient::Proceed() { @@ -127,7 +164,6 @@ bool RPCClient::Proceed() { c->Process(); delete c; - req_count_--; return true; } diff --git a/paddle/operators/detail/grpc_client.h b/paddle/operators/detail/grpc_client.h index a62e70a2533ae52d84d010504b19fed5aeb15dc0..f9499f6dc70c541c214e0b659f10b2ed1e8e8581 100644 --- a/paddle/operators/detail/grpc_client.h +++ b/paddle/operators/detail/grpc_client.h @@ -71,6 +71,15 @@ class ClientBase { context_->set_deadline(deadline); } + virtual void Prepare(int64_t time_out) { + context_.reset(new grpc::ClientContext()); + + std::chrono::system_clock::time_point deadline = + std::chrono::system_clock::now() + std::chrono::milliseconds(time_out); + + context_->set_deadline(deadline); + } + virtual void Process() = 0; std::unique_ptr stub_; @@ -117,6 +126,17 @@ class GetProcessor : public ClientBase { RequestGetCallBack response_call_back_ = ProcGetResponse; }; +class BatchBarrierProcessor : public ClientBase { + public: + explicit BatchBarrierProcessor(std::shared_ptr ch) + : ClientBase(ch) {} + + virtual ~BatchBarrierProcessor() {} + + virtual void Process() {} + sendrecv::VoidMessage reply_; +}; + class RPCClient { public: bool AsyncSendVariable(const std::string& ep, @@ -130,6 +150,10 @@ class RPCClient { const framework::Scope& scope, const std::string& var_name, int64_t time_out = 600 * 1000); + + bool AsyncSendBatchBarrier(const std::string& ep, + int64_t time_out = 600 * 1000); + bool Wait(); private: diff --git a/paddle/operators/detail/grpc_server.cc b/paddle/operators/detail/grpc_server.cc index c0b94746a0b7f6ffb657bbf5af18360426933858..4f94e1315fbd2810a05354f7c3fc54ea30967e8a 100644 --- a/paddle/operators/detail/grpc_server.cc +++ b/paddle/operators/detail/grpc_server.cc @@ -36,7 +36,10 @@ class RequestBase { CallStatus Status() { return status_; } void SetStatus(CallStatus status) { status_ = status; } - virtual std::string GetReqName() { assert(false); } + virtual std::string GetReqName() { + assert(false); + return ""; + } protected: grpc::ServerContext ctx_; @@ -80,11 +83,13 @@ class RequestGet final : public RequestBase { public: explicit RequestGet(sendrecv::SendRecvService::AsyncService* service, grpc::ServerCompletionQueue* cq, framework::Scope* scope, - const platform::DeviceContext* dev_ctx) + const platform::DeviceContext* dev_ctx, + SimpleBlockQueue* queue) : RequestBase(service, cq), responder_(&ctx_), scope_(scope), - dev_ctx_(dev_ctx) { + dev_ctx_(dev_ctx), + queue_(queue) { service_->RequestGetVariable(&ctx_, &request_, &responder_, cq_, cq_, this); } @@ -100,6 +105,7 @@ class RequestGet final : public RequestBase { // TODO(gongwb): check var's info. responder_.Finish(reply_, grpc::Status::OK, this); status_ = FINISH; + queue_->Push('c'); } protected: @@ -108,8 +114,15 @@ class RequestGet final : public RequestBase { ServerAsyncResponseWriter responder_; framework::Scope* scope_; const platform::DeviceContext* dev_ctx_; + SimpleBlockQueue* queue_; }; +void AsyncGRPCServer::WaitClientGet(int count) { + for (int i = 0; i < count; ++i) { + var_get_queue_.Pop(); + } +} + void AsyncGRPCServer::RunSyncUpdate() { grpc::ServerBuilder builder; builder.AddListeningPort(address_, grpc::InsecureServerCredentials()); @@ -119,6 +132,7 @@ void AsyncGRPCServer::RunSyncUpdate() { cq_send_ = builder.AddCompletionQueue(); cq_get_ = builder.AddCompletionQueue(); + server_ = builder.BuildAndStart(); LOG(INFO) << "Server listening on " << address_ << std::endl; @@ -128,11 +142,11 @@ void AsyncGRPCServer::RunSyncUpdate() { std::bind(&AsyncGRPCServer::TryToRegisterNewGetOne, this); t_send_.reset( - new std::thread(std::bind(&AsyncGRPCServer::HandleRequest, this, false, + new std::thread(std::bind(&AsyncGRPCServer::HandleRequest, this, cq_send_.get(), "cq_send", send_register))); t_get_.reset( - new std::thread(std::bind(&AsyncGRPCServer::HandleRequest, this, true, + new std::thread(std::bind(&AsyncGRPCServer::HandleRequest, this, cq_get_.get(), "cq_get", get_register))); // wait server @@ -149,7 +163,6 @@ void AsyncGRPCServer::ShutdownQueue() { } // This URL explains why shutdown is complicate: -// https://stackoverflow.com/questions/35708348/grpc-what-is-the-recommended-way-to-shut-down-an-asynchronous-server-in-c void AsyncGRPCServer::ShutDown() { server_->Shutdown(); ShutdownQueue(); @@ -162,7 +175,7 @@ void AsyncGRPCServer::TryToRegisterNewSendOne() { } RequestSend* send = new RequestSend(&service_, cq_send_.get(), &var_recv_queue_); - VLOG(4) << "create RequestSend status:" << send->Status(); + VLOG(4) << "Create RequestSend status:" << send->Status(); } void AsyncGRPCServer::TryToRegisterNewGetOne() { @@ -170,11 +183,13 @@ void AsyncGRPCServer::TryToRegisterNewGetOne() { if (is_shut_down_) { return; } - RequestGet* get = new RequestGet(&service_, cq_get_.get(), scope_, dev_ctx_); - VLOG(4) << "create Requestget status:" << get->Status(); + RequestGet* get = new RequestGet(&service_, cq_get_.get(), scope_, dev_ctx_, + &var_get_queue_); + VLOG(4) << "Create RequestGet status:" << get->Status(); } -void AsyncGRPCServer::HandleRequest(bool wait, grpc::ServerCompletionQueue* cq, +// FIXME(typhoonzero): change cq_name to enum. +void AsyncGRPCServer::HandleRequest(grpc::ServerCompletionQueue* cq, std::string cq_name, std::function TryToRegisterNewOne) { TryToRegisterNewOne(); @@ -188,9 +203,9 @@ void AsyncGRPCServer::HandleRequest(bool wait, grpc::ServerCompletionQueue* cq, } PADDLE_ENFORCE(tag); - if (wait && !done_) { - Wait(); - } + // FIXME(typhoonzero): de-couple the barriers with recv_op + if (cq_name == "cq_get") WaitCond(1); + if (cq_name == "cq_send") WaitCond(0); RequestBase* base = (RequestBase*)tag; // reference: @@ -222,22 +237,18 @@ void AsyncGRPCServer::HandleRequest(bool wait, grpc::ServerCompletionQueue* cq, } } -void AsyncGRPCServer::Wait() { - std::unique_lock lock(this->mutex_); - condition_.wait(lock, [=] { return this->done_ == true; }); -} - -void AsyncGRPCServer::Reset() { - std::lock_guard lock(this->mutex_); - done_ = false; +void AsyncGRPCServer::WaitCond(int cond) { + std::unique_lock lock(this->barrier_mutex_); + barrier_condition_.wait(lock, + [=] { return this->barrier_cond_step_ == cond; }); } -void AsyncGRPCServer::Done() { +void AsyncGRPCServer::SetCond(int cond) { { - std::lock_guard lock(this->mutex_); - done_ = true; + std::lock_guard lock(this->barrier_mutex_); + barrier_cond_step_ = cond; } - condition_.notify_all(); + barrier_condition_.notify_all(); } } // namespace detail diff --git a/paddle/operators/detail/grpc_server.h b/paddle/operators/detail/grpc_server.h index 2c078b77771656dc7fc0342ecf21b8d33dc11817..3f8b9d93176148619d6820f6a365d9da2e73b10d 100644 --- a/paddle/operators/detail/grpc_server.h +++ b/paddle/operators/detail/grpc_server.h @@ -41,9 +41,10 @@ class AsyncGRPCServer final : public sendrecv::SendRecvService::Service { void RunSyncUpdate(); - void Reset(); - - void Done(); + // functions to sync server barrier status. + void WaitCond(int cond); + void SetCond(int cond); + void WaitClientGet(int count); void SetScope(framework::Scope *scope) { scope_ = scope; } @@ -56,9 +57,7 @@ class AsyncGRPCServer final : public sendrecv::SendRecvService::Service { void ShutDown(); protected: - void Wait(); - void HandleRequest(bool wait, grpc::ServerCompletionQueue *cq, - std::string cq_name, + void HandleRequest(grpc::ServerCompletionQueue *cq, std::string cq_name, std::function TryToRegisterNewOne); void TryToRegisterNewSendOne(); void TryToRegisterNewGetOne(); @@ -78,11 +77,12 @@ class AsyncGRPCServer final : public sendrecv::SendRecvService::Service { const platform::DeviceContext *dev_ctx_; // received variable from RPC, operators fetch variable from this queue. SimpleBlockQueue var_recv_queue_; + SimpleBlockQueue var_get_queue_; // condition of the sub program - std::mutex mutex_; - volatile mutable bool done_; - std::condition_variable condition_; + std::mutex barrier_mutex_; + mutable int barrier_cond_step_; + std::condition_variable barrier_condition_; std::unique_ptr t_send_; std::unique_ptr t_get_; diff --git a/paddle/operators/detail/sendrecvop_utils.h b/paddle/operators/detail/sendrecvop_utils.h index bc6581afab93c626c7c2439d699c6c2d858df9fa..8e66f7299c7b4d30bc5a6fe6a18b7cb3ae3827a5 100644 --- a/paddle/operators/detail/sendrecvop_utils.h +++ b/paddle/operators/detail/sendrecvop_utils.h @@ -30,6 +30,9 @@ namespace paddle { namespace operators { namespace detail { +#define LISTEN_TERMINATE_MESSAGE "TERMINATE@RECV" +#define BATCH_BARRIER_MESSAGE "BATCH_BARRIER@RECV" + void SerializeToMessage(const std::string& name, const framework::Variable* var, const platform::DeviceContext& ctx, sendrecv::VariableMessage* msg); diff --git a/paddle/operators/dropout_op.cc b/paddle/operators/dropout_op.cc index 35cb18797ff66cb87a6658e73ce02b0bfae29baa..5274aa204e6629c9c5ea850c433e0948c89015bd 100644 --- a/paddle/operators/dropout_op.cc +++ b/paddle/operators/dropout_op.cc @@ -51,6 +51,13 @@ class DropoutOpMaker : public framework::OpProtoAndCheckerMaker { "'dropout_prob' must be between 0.0 and 1.0."); }); AddAttr("is_test", "True if in test phase.").SetDefault(false); + AddAttr("fix_seed", + "A flag indicating whether to use a fixed seed to generate " + "random mask. NOTE: DO NOT set this flag to true in " + "training. Setting this flag to true is only useful in " + "unittest or for debug that always the same output units " + "will be dropped.") + .SetDefault(false); AddAttr("seed", "Dropout random seed.").SetDefault(0); AddComment(R"DOC( diff --git a/paddle/operators/dropout_op.cu b/paddle/operators/dropout_op.cu index c56930336e865079f1b96df0f35b0a051fe63a27..84d78445a4fa340ba3c066bb48b96b2a890db652 100644 --- a/paddle/operators/dropout_op.cu +++ b/paddle/operators/dropout_op.cu @@ -62,7 +62,11 @@ class GPUDropoutKernel : public framework::OpKernel { auto* mask = context.Output("Mask"); auto* mask_data = mask->mutable_data(context.GetPlace()); int size = framework::product(mask->dims()); - int seed = context.Attr("seed"); + + std::random_device rnd; + int seed = + context.Attr("fix_seed") ? context.Attr("seed") : rnd(); + thrust::counting_iterator index_sequence_begin(0); thrust::transform(index_sequence_begin, index_sequence_begin + size, thrust::device_ptr(mask_data), diff --git a/paddle/operators/dropout_op.h b/paddle/operators/dropout_op.h index c90b8d277eb78048c001d36a367287146b51c636..46e5dbc64ff9ad3d04a9c1c07f4226932f661baf 100644 --- a/paddle/operators/dropout_op.h +++ b/paddle/operators/dropout_op.h @@ -38,9 +38,15 @@ class CPUDropoutKernel : public framework::OpKernel { if (!context.Attr("is_test")) { auto* mask = context.Output("Mask"); auto* mask_data = mask->mutable_data(context.GetPlace()); - int seed = context.Attr("seed"); + + // NOTE: fixed seed should only be used in unittest or for debug. + // Guarantee to use random seed in training. + std::random_device rnd; std::minstd_rand engine; + int seed = + context.Attr("fix_seed") ? context.Attr("seed") : rnd(); engine.seed(seed); + std::uniform_real_distribution dist(0, 1); size_t size = framework::product(mask->dims()); for (size_t i = 0; i < size; ++i) { diff --git a/paddle/operators/edit_distance_op.cc b/paddle/operators/edit_distance_op.cc index e383f07fa9b53a3def10f6405a0d36f48f52ff08..7e7dfc79eba5c9a75366415e5f4b3183653a5cc6 100644 --- a/paddle/operators/edit_distance_op.cc +++ b/paddle/operators/edit_distance_op.cc @@ -25,6 +25,8 @@ class EditDistanceOp : public framework::OperatorWithKernel { PADDLE_ENFORCE(ctx->HasInput("Hyps"), "Input(Hyps) shouldn't be null."); PADDLE_ENFORCE(ctx->HasInput("Refs"), "Input(Refs) shouldn't be null."); PADDLE_ENFORCE(ctx->HasOutput("Out"), "Output(Out) shouldn't be null."); + PADDLE_ENFORCE(ctx->HasOutput("SequenceNum"), + "Output(SequenceNum) shouldn't be null."); auto hyp_dims = ctx->GetInputDim("Hyps"); auto ref_dims = ctx->GetInputDim("Refs"); PADDLE_ENFORCE(hyp_dims.size() == 2 && hyp_dims[1] == 1, @@ -34,6 +36,7 @@ class EditDistanceOp : public framework::OperatorWithKernel { "Input(Refs) must be a 2-D LoDTensor with the 2nd dimension " "equal to 1."); ctx->SetOutputDim("Out", ctx->GetInputDim("Refs")); + ctx->SetOutputDim("SequenceNum", {1}); } protected: @@ -49,11 +52,12 @@ class EditDistanceOpMaker : public framework::OpProtoAndCheckerMaker { EditDistanceOpMaker(OpProto *proto, OpAttrChecker *op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { AddInput("Hyps", - "(2-D LoDTensor, 2nd dim. equal to 1) " + "(2-D LoDTensor, 2nd dim. equal to 1) " "The indices for hypothesis strings."); AddInput("Refs", - "(2-D LoDTensor, 2nd dim. equal to 1) " + "(2-D LoDTensor, 2nd dim. equal to 1) " "The indices for reference strings."); + AddOutput("SequenceNum", "The sequence count of current batch"); AddAttr("normalized", "(bool, default false) Indicated whether to normalize " "the edit distance by the length of reference string.") @@ -66,22 +70,22 @@ class EditDistanceOpMaker : public framework::OpProtoAndCheckerMaker { EditDistance operator computes the edit distances between a batch of hypothesis strings and their references. -Edit distance, also called Levenshtein distance, measures how dissimilar two strings -are by counting the minimum number of operations to transform one string into anthor. -Here the operations include insertion, deletion, and substitution. For example, -given hypothesis string A = "kitten" and reference B = "sitting", the edit distance -is 3 for A will be transformed into B at least after two substitutions and one +Edit distance, also called Levenshtein distance, measures how dissimilar two strings +are by counting the minimum number of operations to transform one string into anthor. +Here the operations include insertion, deletion, and substitution. For example, +given hypothesis string A = "kitten" and reference B = "sitting", the edit distance +is 3 for A will be transformed into B at least after two substitutions and one insertion: - + "kitten" -> "sitten" -> "sittin" -> "sitting" -Input(Hyps) is a LoDTensor consisting of all the hypothesis strings with the total -number denoted by `batch_size`, and the separation is specified by the LoD information. -And the `batch_size` reference strings are arranged in order in the same way in the +Input(Hyps) is a LoDTensor consisting of all the hypothesis strings with the total +number denoted by `batch_size`, and the separation is specified by the LoD information. +And the `batch_size` reference strings are arranged in order in the same way in the LoDTensor Input(Refs). -Output(Out) contains the `batch_size` results and each stands for the edit stance -for a pair of strings respectively. If Attr(normalized) is true, the edit distance +Output(Out) contains the `batch_size` results and each stands for the edit stance +for a pair of strings respectively. If Attr(normalized) is true, the edit distance will be divided by the length of reference string. )DOC"); } diff --git a/paddle/operators/edit_distance_op.cu b/paddle/operators/edit_distance_op.cu index cf5ebc5c38fd006d10de790e45e9bff3409bd20c..c3e116af086627d576c7a788caebd45667d70017 100644 --- a/paddle/operators/edit_distance_op.cu +++ b/paddle/operators/edit_distance_op.cu @@ -14,6 +14,7 @@ limitations under the License. */ #include #include "paddle/framework/op_registry.h" +#include "paddle/operators/math/math_function.h" #include "paddle/platform/cuda_helper.h" #include "paddle/platform/gpu_info.h" @@ -39,8 +40,8 @@ __global__ void FillFirstColumn(T* dist, const int M, const int N) { } template -__global__ void Levenshtein(T* dist, const int* x1, const int* x2, const int M, - const int N, const int start) { +__global__ void Levenshtein(T* dist, const int64_t* x1, const int64_t* x2, + const int M, const int N, const int start) { int idx = blockDim.x * blockIdx.x + threadIdx.x; int offset = N; int index = start + idx * offset; @@ -72,6 +73,8 @@ class EditDistanceGPUKernel : public framework::OpKernel { auto* x1_t = ctx.Input("Hyps"); auto* x2_t = ctx.Input("Refs"); + auto* sequence_num = ctx.Output("SequenceNum"); + sequence_num->mutable_data(ctx.GetPlace()); auto normalized = ctx.Attr("normalized"); auto stream = reinterpret_cast( @@ -88,7 +91,11 @@ class EditDistanceGPUKernel : public framework::OpKernel { "Reference string %d is empty.", i); } - auto num_strs = hyp_lod.size() - 1; + const size_t num_strs = hyp_lod.size() - 1; + math::SetConstant set_constant; + set_constant(ctx.template device_context(), + sequence_num, static_cast(num_strs)); + out_t->Resize({static_cast(num_strs), 1}); out_t->mutable_data(ctx.GetPlace()); auto out = out_t->data(); @@ -113,8 +120,8 @@ class EditDistanceGPUKernel : public framework::OpKernel { dist_t.Resize({m + 1, n + 1}); dist_t.mutable_data(ctx.GetPlace()); auto dist = dist_t.data(); - auto x1 = x1_t->data() + hyp_lod[num]; - auto x2 = x2_t->data() + ref_lod[num]; + auto x1 = x1_t->data() + hyp_lod[num]; + auto x2 = x2_t->data() + ref_lod[num]; FillFirstColumn<<<1 + m / PADDLE_CUDA_NUM_THREADS, PADDLE_CUDA_NUM_THREADS, 0, stream>>>(dist, m, n); diff --git a/paddle/operators/edit_distance_op.h b/paddle/operators/edit_distance_op.h index 537e70281a5a750db480468a8f8e3c0465de6c5a..974299e604d22422e9024382e85c843ad831575d 100644 --- a/paddle/operators/edit_distance_op.h +++ b/paddle/operators/edit_distance_op.h @@ -16,7 +16,6 @@ limitations under the License. */ #include #include "paddle/framework/eigen.h" #include "paddle/framework/op_registry.h" - namespace paddle { namespace operators { @@ -28,6 +27,8 @@ class EditDistanceKernel : public framework::OpKernel { auto* x1_t = ctx.Input("Hyps"); auto* x2_t = ctx.Input("Refs"); + auto* sequence_num = ctx.Output("SequenceNum"); + int64_t* seq_num_data = sequence_num->mutable_data(ctx.GetPlace()); auto normalized = ctx.Attr("normalized"); @@ -41,6 +42,7 @@ class EditDistanceKernel : public framework::OpKernel { "Reference string %d is empty.", i); } auto num_strs = hyp_lod.size() - 1; + *seq_num_data = static_cast(num_strs); out_t->Resize({static_cast(num_strs), 1}); out_t->mutable_data(ctx.GetPlace()); @@ -60,8 +62,8 @@ class EditDistanceKernel : public framework::OpKernel { dist_t.Resize({m + 1, n + 1}); dist_t.mutable_data(ctx.GetPlace()); auto dist = dist_t.data(); - auto x1 = x1_t->data() + hyp_lod[num]; - auto x2 = x2_t->data() + ref_lod[num]; + auto x1 = x1_t->data() + hyp_lod[num]; + auto x2 = x2_t->data() + ref_lod[num]; for (int64_t i = 0; i < m + 1; ++i) { dist[i * (n + 1)] = i; } diff --git a/paddle/operators/elementwise_add_op.h b/paddle/operators/elementwise_add_op.h index 6478e1e0c2e1cfc8a1be5e8842113ec8ca33d762..a8389429f26c17ceab1db22175c90888546ead6f 100644 --- a/paddle/operators/elementwise_add_op.h +++ b/paddle/operators/elementwise_add_op.h @@ -28,39 +28,7 @@ template class ElementwiseAddKernel : public framework::OpKernel { public: void Compute(const framework::ExecutionContext& ctx) const override { - using Tensor = framework::Tensor; - - auto* x = ctx.Input("X"); - auto* y = ctx.Input("Y"); - auto* z = ctx.Output("Out"); - z->mutable_data(ctx.GetPlace()); - TransformFunctor, T, DeviceContext> functor( - x, y, z, ctx.template device_context(), AddFunctor()); - - auto x_dims = x->dims(); - auto y_dims = y->dims(); - PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(), - "Rank of first input must >= rank of second input."); - - if (x_dims == y_dims) { - functor.Run(); - return; - } - - int axis = ctx.Attr("axis"); - axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis); - PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(), - "Axis should be in range [0, x_dims)"); - - int pre, n, post; - get_mid_dims(x_dims, y_dims, axis, pre, n, post); - if (post == 1) { - functor.RunRowWise(n, pre); - return; - } else { - functor.RunMidWise(n, pre, post); - return; - } + ElementwiseComputeEx, DeviceContext, T>(ctx); } }; diff --git a/paddle/operators/elementwise_div_op.h b/paddle/operators/elementwise_div_op.h index 7783875e24ef0d94e50f49b3598c93ff8bd5d73f..ef26cb6c914f50ded07cc9d0d8de3f49f2151129 100644 --- a/paddle/operators/elementwise_div_op.h +++ b/paddle/operators/elementwise_div_op.h @@ -19,11 +19,16 @@ limitations under the License. */ namespace paddle { namespace operators { +template +struct DivFunctor { + inline HOSTDEVICE T operator()(T a, T b) const { return a / b; } +}; + template class ElementwiseDivKernel : public framework::OpKernel { public: void Compute(const framework::ExecutionContext& ctx) const override { - ElementwiseCompute(ctx); + ElementwiseComputeEx, DeviceContext, T>(ctx); } }; diff --git a/paddle/operators/elementwise_max_op.cc b/paddle/operators/elementwise_max_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..53c27ae5be4cbfe85ce61aa27196594ae152eea4 --- /dev/null +++ b/paddle/operators/elementwise_max_op.cc @@ -0,0 +1,45 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/elementwise_max_op.h" +#include "paddle/operators/elementwise_op.h" + +namespace paddle { +namespace operators { +class ElementwiseMaxOpMaker : public ElementwiseOpMaker { + public: + ElementwiseMaxOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : ElementwiseOpMaker(proto, op_checker) { + SetComment("Max", "Out = max(X, Y)"); + AddComment(comment_); + } +}; +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP(elementwise_max, ops::ElementwiseOp, ops::ElementwiseMaxOpMaker, + elementwise_max_grad, ops::ElementwiseOpGrad); +REGISTER_OP_CPU_KERNEL( + elementwise_max, + ops::ElementwiseMaxKernel, + ops::ElementwiseMaxKernel, + ops::ElementwiseMaxKernel, + ops::ElementwiseMaxKernel); +REGISTER_OP_CPU_KERNEL( + elementwise_max_grad, + ops::ElementwiseMaxGradKernel, + ops::ElementwiseMaxGradKernel, + ops::ElementwiseMaxGradKernel, + ops::ElementwiseMaxGradKernel); diff --git a/paddle/operators/elementwise_max_op.cu b/paddle/operators/elementwise_max_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..5ff4af17477cbd35b765cc00d46c95fda620e2df --- /dev/null +++ b/paddle/operators/elementwise_max_op.cu @@ -0,0 +1,32 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#define EIGEN_USE_GPU +#include "paddle/operators/elementwise_max_op.h" + +namespace ops = paddle::operators; + +REGISTER_OP_CUDA_KERNEL( + elementwise_max, + ops::ElementwiseMaxKernel, + ops::ElementwiseMaxKernel, + ops::ElementwiseMaxKernel, + ops::ElementwiseMaxKernel); +REGISTER_OP_CUDA_KERNEL( + elementwise_max_grad, + ops::ElementwiseMaxGradKernel, + ops::ElementwiseMaxGradKernel, + ops::ElementwiseMaxGradKernel, + ops::ElementwiseMaxGradKernel); diff --git a/paddle/operators/elementwise_max_op.h b/paddle/operators/elementwise_max_op.h new file mode 100644 index 0000000000000000000000000000000000000000..255728e8e620665a7de225b228c19d6c510da1c8 --- /dev/null +++ b/paddle/operators/elementwise_max_op.h @@ -0,0 +1,120 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include "paddle/operators/elementwise_op_function.h" + +namespace paddle { +namespace operators { + +template +struct MaxFunctor { + inline HOSTDEVICE T operator()(T a, T b) const { return a > b ? a : b; } +}; + +template +class ElementwiseMaxKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + ElementwiseComputeEx, DeviceContext, T>(ctx); + } +}; + +template +struct ElementwiseMaxGradFunctor { + template + void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz) { + auto x_e = framework::EigenVector::Flatten(*x); + auto y_e = framework::EigenVector::Flatten(*y); + auto dz_e = framework::EigenVector::Flatten(*dz); + + if (dx) { + auto dx_e = framework::EigenVector::Flatten(*dx); + dx_e.device(d) = (x_e > y_e).template cast() * dz_e; + } + if (dy) { + auto dy_e = framework::EigenVector::Flatten(*dy); + dy_e.device(d) = (x_e <= y_e).template cast() * dz_e; + } + } +}; + +template +struct ElementwiseMaxBroadCastGradFunctor { + template + void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n) { + auto x_e = framework::EigenVector::Flatten(*x); + auto y_e = framework::EigenVector::Flatten(*y); + auto dz_e = framework::EigenVector::Flatten(*dz); + + auto y_e_bcast = y_e.reshape(Eigen::DSizes(1, n)) + .broadcast(Eigen::DSizes(pre, 1)) + .reshape(Eigen::DSizes(x_e.size())); + + if (dx) { + auto dx_e = framework::EigenVector::Flatten(*dx); + dx_e.device(d) = (x_e > y_e_bcast).template cast() * dz_e; + } + + if (dy) { + auto dy_e = framework::EigenVector::Flatten(*dy); + dy_e.device(d) = ((x_e <= y_e_bcast).template cast() * dz_e) + .reshape(Eigen::DSizes(pre, n)) + .sum(Eigen::array{{0}}); + } + } +}; + +template +struct ElementwiseMaxBroadCast2GradFunctor { + template + void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n, + Post post) { + auto x_e = framework::EigenVector::Flatten(*x); + auto y_e = framework::EigenVector::Flatten(*y); + auto dz_e = framework::EigenVector::Flatten(*dz); + + auto y_e_bcast = y_e.reshape(Eigen::DSizes(1, n, 1)) + .broadcast(Eigen::DSizes(pre, 1, post)) + .reshape(Eigen::DSizes(x_e.size())); + if (dx) { + auto dx_e = framework::EigenVector::Flatten(*dx); + dx_e.device(d) = (x_e > y_e_bcast).template cast() * dz_e; + } + + if (dy) { + auto dy_e = framework::EigenVector::Flatten(*dy); + dy_e.device(d) = ((x_e <= y_e_bcast).template cast() * dz_e) + .reshape(Eigen::DSizes(pre, n, post)) + .sum(Eigen::array{{0, 2}}); + } + } +}; + +template +class ElementwiseMaxGradKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + ElementwiseGradCompute, + ElementwiseMaxBroadCastGradFunctor, + ElementwiseMaxBroadCast2GradFunctor>(ctx); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/elementwise_min_op.cc b/paddle/operators/elementwise_min_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..99482e1bf60c88062087c5fe0105e90aa0a8677c --- /dev/null +++ b/paddle/operators/elementwise_min_op.cc @@ -0,0 +1,45 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/elementwise_min_op.h" +#include "paddle/operators/elementwise_op.h" + +namespace paddle { +namespace operators { +class ElementwiseMinOpMaker : public ElementwiseOpMaker { + public: + ElementwiseMinOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : ElementwiseOpMaker(proto, op_checker) { + SetComment("Max", "Out = min(X, Y)"); + AddComment(comment_); + } +}; +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP(elementwise_min, ops::ElementwiseOp, ops::ElementwiseMinOpMaker, + elementwise_min_grad, ops::ElementwiseOpGrad); +REGISTER_OP_CPU_KERNEL( + elementwise_min, + ops::ElementwiseMinKernel, + ops::ElementwiseMinKernel, + ops::ElementwiseMinKernel, + ops::ElementwiseMinKernel); +REGISTER_OP_CPU_KERNEL( + elementwise_min_grad, + ops::ElementwiseMinGradKernel, + ops::ElementwiseMinGradKernel, + ops::ElementwiseMinGradKernel, + ops::ElementwiseMinGradKernel); diff --git a/paddle/operators/elementwise_min_op.cu b/paddle/operators/elementwise_min_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..3547e6ccb77177002b1ecbee4e4604b602f72209 --- /dev/null +++ b/paddle/operators/elementwise_min_op.cu @@ -0,0 +1,32 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#define EIGEN_USE_GPU +#include "paddle/operators/elementwise_min_op.h" + +namespace ops = paddle::operators; + +REGISTER_OP_CUDA_KERNEL( + elementwise_min, + ops::ElementwiseMinKernel, + ops::ElementwiseMinKernel, + ops::ElementwiseMinKernel, + ops::ElementwiseMinKernel); +REGISTER_OP_CUDA_KERNEL( + elementwise_min_grad, + ops::ElementwiseMinGradKernel, + ops::ElementwiseMinGradKernel, + ops::ElementwiseMinGradKernel, + ops::ElementwiseMinGradKernel); diff --git a/paddle/operators/elementwise_min_op.h b/paddle/operators/elementwise_min_op.h new file mode 100644 index 0000000000000000000000000000000000000000..e6627a0f1bb468c8e4661b83489cb964b72dddb0 --- /dev/null +++ b/paddle/operators/elementwise_min_op.h @@ -0,0 +1,120 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include "paddle/operators/elementwise_op_function.h" + +namespace paddle { +namespace operators { + +template +struct MinFunctor { + inline HOSTDEVICE T operator()(T a, T b) const { return a < b ? a : b; } +}; + +template +class ElementwiseMinKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + ElementwiseComputeEx, DeviceContext, T>(ctx); + } +}; + +template +struct ElementwiseMinGradFunctor { + template + void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz) { + auto x_e = framework::EigenVector::Flatten(*x); + auto y_e = framework::EigenVector::Flatten(*y); + auto dz_e = framework::EigenVector::Flatten(*dz); + + if (dx) { + auto dx_e = framework::EigenVector::Flatten(*dx); + dx_e.device(d) = (x_e < y_e).template cast() * dz_e; + } + if (dy) { + auto dy_e = framework::EigenVector::Flatten(*dy); + dy_e.device(d) = (x_e >= y_e).template cast() * dz_e; + } + } +}; + +template +struct ElementwiseMinBroadCastGradFunctor { + template + void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n) { + auto x_e = framework::EigenVector::Flatten(*x); + auto y_e = framework::EigenVector::Flatten(*y); + auto dz_e = framework::EigenVector::Flatten(*dz); + + auto y_e_bcast = y_e.reshape(Eigen::DSizes(1, n)) + .broadcast(Eigen::DSizes(pre, 1)) + .reshape(Eigen::DSizes(x_e.size())); + + if (dx) { + auto dx_e = framework::EigenVector::Flatten(*dx); + dx_e.device(d) = (x_e < y_e_bcast).template cast() * dz_e; + } + + if (dy) { + auto dy_e = framework::EigenVector::Flatten(*dy); + dy_e.device(d) = ((x_e >= y_e_bcast).template cast() * dz_e) + .reshape(Eigen::DSizes(pre, n)) + .sum(Eigen::array{{0}}); + } + } +}; + +template +struct ElementwiseMinBroadCast2GradFunctor { + template + void operator()(Device d, X x, Y y, Z z, dX dx, dY dy, dZ dz, Pre pre, N n, + Post post) { + auto x_e = framework::EigenVector::Flatten(*x); + auto y_e = framework::EigenVector::Flatten(*y); + auto dz_e = framework::EigenVector::Flatten(*dz); + + auto y_e_bcast = y_e.reshape(Eigen::DSizes(1, n, 1)) + .broadcast(Eigen::DSizes(pre, 1, post)) + .reshape(Eigen::DSizes(x_e.size())); + if (dx) { + auto dx_e = framework::EigenVector::Flatten(*dx); + dx_e.device(d) = (x_e < y_e_bcast).template cast() * dz_e; + } + + if (dy) { + auto dy_e = framework::EigenVector::Flatten(*dy); + dy_e.device(d) = ((x_e >= y_e_bcast).template cast() * dz_e) + .reshape(Eigen::DSizes(pre, n, post)) + .sum(Eigen::array{{0, 2}}); + } + } +}; + +template +class ElementwiseMinGradKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + ElementwiseGradCompute, + ElementwiseMinBroadCastGradFunctor, + ElementwiseMinBroadCast2GradFunctor>(ctx); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/elementwise_mul_op.h b/paddle/operators/elementwise_mul_op.h index 0e6559eacc0f90fec78397ce66372071370291e5..4b86b00b5a095ae898f9ce0c17cde2cc91060ba9 100644 --- a/paddle/operators/elementwise_mul_op.h +++ b/paddle/operators/elementwise_mul_op.h @@ -18,11 +18,16 @@ limitations under the License. */ namespace paddle { namespace operators { +template +struct MulFunctor { + inline HOSTDEVICE T operator()(T a, T b) const { return a * b; } +}; + template class ElementwiseMulKernel : public framework::OpKernel { public: void Compute(const framework::ExecutionContext& ctx) const override { - ElementwiseCompute(ctx); + ElementwiseComputeEx, DeviceContext, T>(ctx); } }; diff --git a/paddle/operators/elementwise_op.h b/paddle/operators/elementwise_op.h index a342595b546bfca1a344cf8a549597df6a29adec..1a0131d8b943da3deebd0c461f78cb02b34e6dc2 100644 --- a/paddle/operators/elementwise_op.h +++ b/paddle/operators/elementwise_op.h @@ -26,9 +26,9 @@ class ElementwiseOp : public framework::OperatorWithKernel { using Tensor = framework::Tensor; void InferShape(framework::InferShapeContext* ctx) const override { PADDLE_ENFORCE(ctx->HasInput("X"), - "Input(X) of elementwise op should not be null"); + "Input(X) of elementwise op should not be null."); PADDLE_ENFORCE(ctx->HasInput("Y"), - "Input(Y) of elementwise op should not be null"); + "Input(Y) of elementwise op should not be null."); PADDLE_ENFORCE(ctx->HasOutput("Out"), "Output(Out) of elementwise op should not be null."); @@ -45,12 +45,12 @@ class ElementwiseOpMaker : public framework::OpProtoAndCheckerMaker { public: ElementwiseOpMaker(OpProto* proto, OpAttrChecker* op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { - AddInput("X", "(Tensor) The first input tensor of elementwise op"); - AddInput("Y", "(Tensor) The second input tensor of elementwise op"); - AddOutput("Out", "The output of elementwise op"); + AddInput("X", "(Tensor), The first input tensor of elementwise op."); + AddInput("Y", "(Tensor), The second input tensor of elementwise op."); + AddOutput("Out", "The output of elementwise op."); AddAttr("axis", - "(int, default -1) The starting dimension index " - "for broadcasting Y onto X") + "(int, default -1). The start dimension index " + "for broadcasting Y onto X.") .SetDefault(-1) .EqualGreaterThan(-1); comment_ = R"DOC( @@ -58,19 +58,18 @@ Limited Elementwise {name} Operator. The equation is: -.. math:: - {equation} +$${equation}$$ -X is a tensor of any dimension and the dimensions of tensor Y must be smaller than -or equal to the dimensions of X. +$X$ is a tensor of any dimension and the dimensions of tensor $Y$ must be +smaller than or equal to the dimensions of $X$. There are two cases for this operator: -1. The shape of Y is same with X; -2. The shape of Y is a subset of X. +1. The shape of $Y$ is same with $X$; +2. The shape of $Y$ is a subset of $X$. For case 2: -Y will be broadcasted to match the shape of X and axis should be -the starting dimension index for broadcasting Y onto X. +$Y$ will be broadcasted to match the shape of $X$ and axis should be +set to index of the start dimension to broadcast $Y$ onto $X$. For example .. code-block:: python @@ -81,7 +80,8 @@ For example shape(X) = (2, 3, 4, 5), shape(Y) = (3, 4), with axis=1 shape(X) = (2, 3, 4, 5), shape(Y) = (2), with axis=0 -Either of the inputs X and Y or none can carry the LoD (Level of Details) information. However, the output only shares the LoD information with input X. +Either of the inputs $X$ and $Y$ or none can carry the LoD (Level of Details) +information. However, the output only shares the LoD information with input $X$. )DOC"; AddComment(comment_); diff --git a/paddle/operators/elementwise_op_function.h b/paddle/operators/elementwise_op_function.h index 0c75276b03140473cee4b57a4022ff3c6989ab4c..d749b8e8757d0d433be05876779ccc22b95ca80b 100644 --- a/paddle/operators/elementwise_op_function.h +++ b/paddle/operators/elementwise_op_function.h @@ -176,14 +176,15 @@ class MidWiseTransformIterator }; #endif -template +template class TransformFunctor { public: TransformFunctor(const framework::Tensor* x, const framework::Tensor* y, framework::Tensor* z, const DeviceContext& ctx, Functor func) : x_(x->data()), y_(y->data()), - z_(z->mutable_data(ctx.GetPlace())), + z_(z->mutable_data(ctx.GetPlace())), nx_(x->numel()), ctx_(ctx), func_(func) {} @@ -208,7 +209,7 @@ class TransformFunctor { private: const T* x_; const T* y_; - T* z_; + OutType* z_; int64_t nx_; const DeviceContext& ctx_; Functor func_; @@ -340,6 +341,13 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) { return; } + if (y_dims.size() == 1 && y_dims[0] == 1) { + // y is a scalar + auto extended_dims = framework::vectorize(x_dims); + extended_dims.push_back(1); + x_dims = framework::make_ddim(extended_dims); + } + int axis = ctx.Attr("axis"); axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis); @@ -356,5 +364,51 @@ void ElementwiseGradCompute(const framework::ExecutionContext& ctx) { return; } } + +template +void ElementwiseComputeEx(const framework::ExecutionContext& ctx) { + using Tensor = framework::Tensor; + + auto* x = ctx.Input("X"); + auto* y = ctx.Input("Y"); + auto* z = ctx.Output("Out"); + z->mutable_data(ctx.GetPlace()); + TransformFunctor functor( + x, y, z, ctx.template device_context(), Functor()); + + auto x_dims = x->dims(); + auto y_dims = y->dims(); + PADDLE_ENFORCE_GE(x_dims.size(), y_dims.size(), + "Rank of first input must >= rank of second input."); + + if (x_dims == y_dims) { + functor.Run(); + return; + } + + if (y_dims.size() == 1 && y_dims[0] == 1) { + // y is a scalar + auto extended_dims = framework::vectorize(x_dims); + extended_dims.push_back(1); + x_dims = framework::make_ddim(extended_dims); + } + + int axis = ctx.Attr("axis"); + axis = (axis == -1 ? x_dims.size() - y_dims.size() : axis); + PADDLE_ENFORCE(axis >= 0 && axis < x_dims.size(), + "Axis should be in range [0, x_dims)"); + + int pre, n, post; + get_mid_dims(x_dims, y_dims, axis, pre, n, post); + if (post == 1) { + functor.RunRowWise(n, pre); + return; + } else { + functor.RunMidWise(n, pre, post); + return; + } +} + } // namespace operators } // namespace paddle diff --git a/paddle/operators/elementwise_pow_op.cc b/paddle/operators/elementwise_pow_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..5293cc7dd34ccee860c50e964516da9b4d42d29c --- /dev/null +++ b/paddle/operators/elementwise_pow_op.cc @@ -0,0 +1,37 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/elementwise_pow_op.h" +#include "paddle/operators/elementwise_op.h" + +namespace paddle { +namespace operators { +class ElementwisePowOpMaker : public ElementwiseOpMaker { + public: + ElementwisePowOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : ElementwiseOpMaker(proto, op_checker) { + SetComment("Pow", "Out = X ^ Y"); + AddComment(comment_); + } +}; +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP_WITHOUT_GRADIENT(elementwise_pow, ops::ElementwiseOp, + ops::ElementwisePowOpMaker); +REGISTER_OP_CPU_KERNEL( + elementwise_pow, + ops::ElementwisePowKernel, + ops::ElementwisePowKernel); diff --git a/paddle/operators/elementwise_pow_op.cu b/paddle/operators/elementwise_pow_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..643c978e635bc8e9671b47774c2eac5b713f59c2 --- /dev/null +++ b/paddle/operators/elementwise_pow_op.cu @@ -0,0 +1,20 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + http://www.apache.org/licenses/LICENSE-2.0 +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#define EIGEN_USE_GPU +#include "paddle/operators/elementwise_pow_op.h" + +namespace ops = paddle::operators; + +REGISTER_OP_CUDA_KERNEL( + elementwise_pow, + ops::ElementwisePowKernel, + ops::ElementwisePowKernel); diff --git a/paddle/operators/elementwise_pow_op.h b/paddle/operators/elementwise_pow_op.h new file mode 100644 index 0000000000000000000000000000000000000000..6019e709e0db0fd62b4d3350bb768095f87ef241 --- /dev/null +++ b/paddle/operators/elementwise_pow_op.h @@ -0,0 +1,37 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include +#include "paddle/operators/elementwise_op_function.h" + +namespace paddle { +namespace operators { + +template +struct PowFunctor { + inline HOSTDEVICE T operator()(T a, T b) const { return std::pow(a, b); } +}; + +template +class ElementwisePowKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + ElementwiseComputeEx, DeviceContext, T>(ctx); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/elementwise_sub_op.h b/paddle/operators/elementwise_sub_op.h index 347e92f87c7a46d3935f69864718d63661bef05f..a2aca793026189ec87e00b52d7c351689f870400 100644 --- a/paddle/operators/elementwise_sub_op.h +++ b/paddle/operators/elementwise_sub_op.h @@ -18,11 +18,16 @@ limitations under the License. */ namespace paddle { namespace operators { +template +struct SubFunctor { + inline HOSTDEVICE T operator()(T a, T b) const { return a - b; } +}; + template class ElementwiseSubKernel : public framework::OpKernel { public: void Compute(const framework::ExecutionContext& ctx) const override { - ElementwiseCompute(ctx); + ElementwiseComputeEx, DeviceContext, T>(ctx); } }; diff --git a/paddle/operators/expand_op.cc b/paddle/operators/expand_op.cc index 08fa91ed72aa41ed2f513c090b9085410bb5cc47..043c93654d33f7c105c89960e18ec72d3557237d 100644 --- a/paddle/operators/expand_op.cc +++ b/paddle/operators/expand_op.cc @@ -58,21 +58,21 @@ class ExpandOpMaker : public framework::OpProtoAndCheckerMaker { ExpandOpMaker(OpProto* proto, OpAttrChecker* op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { AddInput("X", - "(Tensor, default Tensor) A tensor with rank in [1, 6]." - "X is the input tensor to be expanded."); + "(Tensor, default Tensor). A tensor with rank in [1, 6]." + "X is the input to be expanded."); AddOutput("Out", - "(Tensor, default Tensor) A tensor with rank in [1, 6]." - "The rank of Output(Out) is same as Input(X) except that each " - "dimension size of Output(Out) is equal to corresponding " - "dimension size of Input(X) multiplying corresponding value of " - "Attr(expand_times)."); + "(Tensor, default Tensor). A tensor with rank in [1, 6]." + "The rank of Output(Out) have the same with Input(X). " + "After expanding, size of each dimension of Output(Out) is equal " + "to size of the corresponding dimension of Input(X) multiplying " + "the corresponding value given by Attr(expand_times)."); AddAttr>("expand_times", "Expand times number for each dimension."); AddComment(R"DOC( Expand operator tiles the input by given times number. You should set times number for each dimension by providing attribute 'expand_times'. The rank of X -should be in [1, 6]. Please notice that size of 'expand_times' must be same with -X's rank. Following is a using case: +should be in [1, 6]. Please note that size of 'expand_times' must be the same +with X's rank. Following is a using case: Input(X) is a 3-D tensor with shape [2, 3, 1]: diff --git a/paddle/operators/gru_op.cc b/paddle/operators/gru_op.cc index 76f2adefede3b4bc4035f86f8f8663eed29343ae..fb901b639492a179925ff852f9030fc6674d1f63 100644 --- a/paddle/operators/gru_op.cc +++ b/paddle/operators/gru_op.cc @@ -135,14 +135,14 @@ class GRUOpMaker : public framework::OpProtoAndCheckerMaker { AddComment(R"DOC( GRU Operator implements part calculations of the complete GRU as following: -\f[ -update \ gate: u_t = actGate(xu_t + W_u * h_{t-1} + b_u) \\ -reset \ gate: r_t = actGate(xr_t + W_r * h_{t-1} + b_r) \\ -output \ candidate: {h}_t = actNode(xc_t + W_c * dot(r_t, h_{t-1}) + b_c) \\ +$$ +update\_gate: u_t = actGate(xu_t + W_u * h_{t-1} + b_u) \\ +reset\_gate: r_t = actGate(xr_t + W_r * h_{t-1} + b_r) \\ +output\_candidate: {h}_t = actNode(xc_t + W_c * dot(r_t, h_{t-1}) + b_c) \\ output: h_t = dot((1 - u_t), h_{t-1}) + dot(u_t, {h}_t) -\f] +$$ -@note To implement the complete GRU, fully-connected operator must be used +@note To implement the complete GRU, fully-connected operator must be used before to feed xu, xr and xc as the Input of GRU operator. )DOC"); } diff --git a/paddle/operators/gru_op.h b/paddle/operators/gru_op.h index b1957fb9ce6add8628cb206abf2c569d3f615c85..a08bd4233b02d021aaa64bafe4b855f11a60d338 100644 --- a/paddle/operators/gru_op.h +++ b/paddle/operators/gru_op.h @@ -30,11 +30,12 @@ using Tensor = framework::Tensor; template inline void ReorderInitState(const DeviceContext& ctx, - const framework::Tensor& src, const size_t* index, + const framework::Tensor& src, + framework::Vector index_lod, framework::Tensor* dst, bool indexed_src) { math::CopyMatrixRowsFunctor row_shuffle; dst->mutable_data(src.dims(), ctx.GetPlace()); - row_shuffle(ctx, src, index, *dst, indexed_src); + row_shuffle(ctx, src, index_lod, *dst, indexed_src); } template @@ -76,7 +77,9 @@ class GRUKernel : public framework::OpKernel { gru_value.state_weight = const_cast(weight_data + 2 * frame_size * frame_size); Tensor ordered_h0; - const size_t* order = batch_gate->lod()[2].data(); + + framework::Vector order(batch_gate->lod()[2]); + if (h0) { // Since the batch computing for GRU reorders the input sequences // according to their length. The initialized cell state also needs @@ -159,7 +162,9 @@ class GRUGradKernel : public framework::OpKernel { zero(dev_ctx, &batch_reset_hidden_prev_grad, static_cast(0.0)); Tensor ordered_h0, ordered_h0_grad; - const size_t* order = batch_gate->lod()[2].data(); + + framework::Vector order(batch_gate->lod()[2]); + if (h0) { ReorderInitState(dev_ctx, *h0, order, &ordered_h0, true); diff --git a/paddle/operators/im2sequence_op.cc b/paddle/operators/im2sequence_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..31baaedf6914b1a6939fc762491ef35013db4bb6 --- /dev/null +++ b/paddle/operators/im2sequence_op.cc @@ -0,0 +1,157 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/im2sequence_op.h" + +namespace paddle { +namespace operators { + +class Im2SequenceOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + protected: + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of Im2SequenceOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Out"), + "Output(Out) of Im2SequenceOp op should not be null."); + + auto in_dim = ctx->GetInputDim("X"); + PADDLE_ENFORCE_EQ(in_dim.size(), 4, + "Input(X) format must be 4D tensor, eg., NCHW."); + + auto kernels = ctx->Attrs().Get>("kernels"); + auto strides = ctx->Attrs().Get>("strides"); + auto paddings = ctx->Attrs().Get>("paddings"); + + int batch_size = in_dim[0]; + int img_channels = in_dim[1]; + int img_height = in_dim[2]; + int img_width = in_dim[3]; + + int output_height = OutputSize(img_height, kernels[0], paddings[0], + paddings[2], strides[0]); + int output_width = + OutputSize(img_width, kernels[1], paddings[1], paddings[3], strides[1]); + + ctx->SetOutputDim("Out", {batch_size * output_height * output_width, + img_channels * kernels[0] * kernels[1]}); + } +}; + +class Im2SequenceOpMaker : public framework::OpProtoAndCheckerMaker { + public: + Im2SequenceOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", + "(Tensor) The input tensor has NCHW format." + "N: batch size" + "C: channels" + "H: height" + "W: width"); + AddOutput("Out", "(LodTensor) The output data of im2sequence op,"); + AddAttr>("kernels", + "(vector), the " + "kernels(kernel_height, kernel_width)"); + AddAttr>("strides", + "(vector default:{1, 1}), the " + "strides(h_stride, w_stride)") + .SetDefault({1, 1}); + AddAttr>("paddings", + "(vector default:{0, 0, 0, 0}), the " + "paddings(up_pad, left_pad, down_pad, right_pad)") + .SetDefault({0, 0, 0, 0}); + AddComment(R"DOC( +This op uses kernels to scan images and converts these images to sequences. +After expanding, The number of time steps are output_height * output_width +and the dimension of each time step is kernel_height * kernel_width * channels, +in which: + +output_height = + 1 + (padding_height + padding_down + img_height - kernel_height + stride_height - 1) / + stride_height; +output_width = + 1 + (padding_left + padding+right + img_width - kernel_width + stride_width - 1) / + stride_width; + +This op can be used after convolution neural network, and before recurrent neural network. + +Given: + +x = [[[[ 6. 2. 1.] + [ 8. 3. 5.] + [ 0. 2. 6.]] + + [[ 2. 4. 4.] + [ 6. 3. 0.] + [ 6. 4. 7.]]] + + [[[ 6. 7. 1.] + [ 5. 7. 9.] + [ 2. 4. 8.]] + + [[ 1. 2. 1.] + [ 1. 3. 5.] + [ 9. 0. 8.]]]] +x.dims = {2, 2, 3, 3} + +And: + +kernels = [2, 2] +strides = [1, 1] +paddings = [0, 0, 0, 0] + +Then: + +output.data = [[ 6. 2. 8. 3. 2. 4. 6. 3.] + [ 2. 1. 3. 5. 4. 4. 3. 0.] + [ 8. 3. 0. 2. 6. 3. 6. 4.] + [ 3. 5. 2. 6. 3. 0. 4. 7.] + [ 6. 7. 5. 7. 1. 2. 1. 3.] + [ 7. 1. 7. 9. 2. 1. 3. 5.] + [ 5. 7. 2. 4. 1. 3. 9. 0.] + [ 7. 9. 4. 8. 3. 5. 0. 8.]] +output.dims = {8, 9} +output.lod = [[0, 4, 8]] + +)DOC"); + } +}; + +class Im2SequenceGradOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + protected: + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), "Input(X) should not be null"); + PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Out")), + "Input(Out@GRAD) shouldn't be null."); + ctx->SetOutputDim(framework::GradVarName("X"), ctx->GetInputDim("X")); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP(im2sequence, ops::Im2SequenceOp, ops::Im2SequenceOpMaker, + im2sequence_grad, ops::Im2SequenceGradOp); +REGISTER_OP_CPU_KERNEL( + im2sequence, + ops::Im2SequenceKernel); +REGISTER_OP_CPU_KERNEL( + im2sequence_grad, + ops::Im2SequenceGradKernel); diff --git a/paddle/operators/im2sequence_op.cu b/paddle/operators/im2sequence_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..9db7529112f2710d6ff4af2b444e304543486de3 --- /dev/null +++ b/paddle/operators/im2sequence_op.cu @@ -0,0 +1,25 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. */ + +#define EIGEN_USE_GPU +#include "paddle/operators/im2sequence_op.h" + +namespace ops = paddle::operators; + +REGISTER_OP_CUDA_KERNEL( + im2sequence, + ops::Im2SequenceKernel); +REGISTER_OP_CUDA_KERNEL( + im2sequence_grad, + ops::Im2SequenceGradKernel); diff --git a/paddle/operators/im2sequence_op.h b/paddle/operators/im2sequence_op.h new file mode 100644 index 0000000000000000000000000000000000000000..f33aec71a92a65ec0e4114530d70e36c9dc1be04 --- /dev/null +++ b/paddle/operators/im2sequence_op.h @@ -0,0 +1,135 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + + Licensed under the Apache License, Version 2.0 (the "License"); + You may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. */ + +#pragma once + +#include "paddle/framework/data_layout.h" +#include "paddle/framework/eigen.h" +#include "paddle/framework/op_registry.h" +#include "paddle/operators/math/im2col.h" +#include "paddle/operators/math/math_function.h" + +namespace paddle { +namespace operators { + +using Tensor = framework::Tensor; +using LoDTensor = framework::LoDTensor; + +inline int OutputSize(int input_size, int filter_size, int padding_0, + int padding_1, int stride) { + const int output_size = + (input_size + padding_0 + padding_1 - filter_size) / stride + 1; + return output_size; +} + +template +class Im2SequenceKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + const Tensor* in = ctx.Input("X"); + LoDTensor* out = ctx.Output("Out"); + out->mutable_data(ctx.GetPlace()); + // TODO(wanghaoshuang): Add layout checker after 'set_layout' + // being available for python API + // PADDLE_ENFORCE_EQ(in->layout(), framework::DataLayout::kNCHW, + // "Input(X) layout must be NCHW"); + auto in_dim = in->dims(); + int batch_size = in_dim[0]; + int img_channels = in_dim[1]; + int img_height = in_dim[2]; + int img_width = in_dim[3]; + + auto kernels = ctx.Attr>("kernels"); + auto strides = ctx.Attr>("strides"); + auto paddings = ctx.Attr>("paddings"); + int output_height = OutputSize(img_height, kernels[0], paddings[0], + paddings[2], strides[0]); + int output_width = + OutputSize(img_width, kernels[1], paddings[1], paddings[3], strides[1]); + + const std::vector dilations({1, 1}); + + auto out_dims = out->dims(); + out->Resize({batch_size, out->numel() / batch_size}); + for (int i = 0; i < batch_size; i++) { + const Tensor src = + in->Slice(i, i + 1).Resize({img_channels, img_height, img_width}); + Tensor dst = out->Slice(i, i + 1).Resize( + {output_height, output_width, img_channels, kernels[0], kernels[1]}); + + math::Im2ColFunctor f; + auto& dev_ctx = ctx.template device_context(); + f(dev_ctx, src, dilations, strides, paddings, &dst); + } + out->Resize(out_dims); + + // set lod information + // TODO(wanghaoshuang): Move this to InferShape + framework::LoD lod(1); + lod[0].reserve(batch_size + 1); + for (int i = 0, offset = 0; i < batch_size + 1; ++i) { + lod[0].push_back(offset); + offset += output_height * output_width; + } + out->set_lod(lod); + } +}; + +template +class Im2SequenceGradKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + auto* in = ctx.Input("X"); + Tensor* d_out = + const_cast(ctx.Input(framework::GradVarName("Out"))); + auto* d_x = ctx.Output(framework::GradVarName("X")); + d_x->mutable_data(ctx.GetPlace()); + + auto x_v = framework::EigenVector::Flatten(*d_x); + auto& place = *ctx.template device_context().eigen_device(); + x_v.device(place) = x_v.constant(0.0); + + auto in_dim = in->dims(); + int batch_size = in_dim[0]; + int img_channels = in_dim[1]; + int img_height = in_dim[2]; + int img_width = in_dim[3]; + + auto kernels = ctx.Attr>("kernels"); + auto strides = ctx.Attr>("strides"); + auto paddings = ctx.Attr>("paddings"); + int output_height = OutputSize(img_height, kernels[0], paddings[0], + paddings[2], strides[0]); + int output_width = + OutputSize(img_width, kernels[1], paddings[1], paddings[3], strides[1]); + + const std::vector dilations({1, 1}); + + auto d_out_dims = d_out->dims(); + d_out->Resize({batch_size, d_out->numel() / batch_size}); + for (int i = 0; i < batch_size; i++) { + Tensor dst = + d_x->Slice(i, i + 1).Resize({img_channels, img_height, img_width}); + const Tensor src = d_out->Slice(i, i + 1).Resize( + {output_height, output_width, img_channels, kernels[0], kernels[1]}); + math::Col2ImFunctor f; + auto& dev_ctx = ctx.template device_context(); + f(dev_ctx, src, dilations, strides, paddings, &dst); + } + d_out->Resize(d_out_dims); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/iou_similarity_op.cc b/paddle/operators/iou_similarity_op.cc new file mode 100755 index 0000000000000000000000000000000000000000..c520b28b83e66dbf53d2e19985370be4a2f69e23 --- /dev/null +++ b/paddle/operators/iou_similarity_op.cc @@ -0,0 +1,96 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/iou_similarity_op.h" + +namespace paddle { +namespace operators { + +class IOUSimilarityOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + protected: + void InferShape(framework::InferShapeContext *ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of IOUSimilarityOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Y"), + "Input(Y) of IOUSimilarityOp should not be null."); + auto x_dims = ctx->GetInputDim("X"); + auto y_dims = ctx->GetInputDim("Y"); + + PADDLE_ENFORCE_EQ(x_dims.size(), 2UL, "The rank of Input(X) must be 2."); + PADDLE_ENFORCE_EQ(x_dims[1], 4UL, "The shape of X is [N, 4]"); + PADDLE_ENFORCE_EQ(y_dims.size(), 2UL, "The rank of Input(Y) must be 2."); + PADDLE_ENFORCE_EQ(y_dims[1], 4UL, "The shape of Y is [M, 4]"); + + ctx->ShareLoD("X", /*->*/ "Out"); + ctx->SetOutputDim("Out", framework::make_ddim({x_dims[0], y_dims[0]})); + } +}; + +class IOUSimilarityOpMaker : public framework::OpProtoAndCheckerMaker { + public: + IOUSimilarityOpMaker(OpProto *proto, OpAttrChecker *op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", + "(LoDTensor, default LoDTensor) " + "Box list X is a 2-D LoDTensor with shape [N, 4] holds N boxes, " + "each box is represented as [xmin, ymin, xmax, ymax], " + "the shape of X is [N, 4]. [xmin, ymin] is the left top " + "coordinate of the box if the input is image feature map, they " + "are close to the origin of the coordinate system. " + "[xmax, ymax] is the right bottom coordinate of the box. " + "This tensor can contain LoD information to represent a batch " + "of inputs. One instance of this batch can contain different " + "numbers of entities."); + AddInput("Y", + "(Tensor, default Tensor) " + "Box list Y holds M boxes, each box is represented as " + "[xmin, ymin, xmax, ymax], the shape of X is [N, 4]. " + "[xmin, ymin] is the left top coordinate of the box if the " + "input is image feature map, and [xmax, ymax] is the right " + "bottom coordinate of the box."); + + AddOutput("Out", + "(LoDTensor, the lod is same as input X) The output of " + "iou_similarity op, a tensor with shape [N, M] " + "representing pairwise iou scores."); + + AddComment(R"DOC( +IOU Similarity Operator. +Computes intersection-over-union (IOU) between two box lists. + Box list 'X' should be a LoDTensor and 'Y' is a common Tensor, + boxes in 'Y' are shared by all instance of the batched inputs of X. + Given two boxes A and B, the calculation of IOU is as follows: + +$$ +IOU(A, B) = +\frac{area(A\cap B)}{area(A)+area(B)-area(A\cap B)} +$$ + +)DOC"); + } +}; +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP_WITHOUT_GRADIENT(iou_similarity, ops::IOUSimilarityOp, + ops::IOUSimilarityOpMaker); + +REGISTER_OP_CPU_KERNEL( + iou_similarity, + ops::IOUSimilarityKernel, + ops::IOUSimilarityKernel); diff --git a/paddle/operators/iou_similarity_op.cu b/paddle/operators/iou_similarity_op.cu new file mode 100755 index 0000000000000000000000000000000000000000..fa5052624618c35875b241419946f69b776c81d4 --- /dev/null +++ b/paddle/operators/iou_similarity_op.cu @@ -0,0 +1,21 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/iou_similarity_op.h" + +namespace ops = paddle::operators; +REGISTER_OP_CUDA_KERNEL( + iou_similarity, + ops::IOUSimilarityKernel, + ops::IOUSimilarityKernel); diff --git a/paddle/operators/iou_similarity_op.h b/paddle/operators/iou_similarity_op.h new file mode 100644 index 0000000000000000000000000000000000000000..e36177069d7b18ea23759f99c4679218fbfd32b8 --- /dev/null +++ b/paddle/operators/iou_similarity_op.h @@ -0,0 +1,90 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include "paddle/framework/op_registry.h" +#include "paddle/platform/for_range.h" + +template +inline HOSTDEVICE T IOUSimilarity(T xmin1, T ymin1, T xmax1, T ymax1, T xmin2, + T ymin2, T xmax2, T ymax2) { + constexpr T zero = static_cast(0); + T area1 = (ymax1 - ymin1) * (xmax1 - xmin1); + T area2 = (ymax2 - ymin2) * (xmax2 - xmin2); + T inter_xmax = xmax1 > xmax2 ? xmax2 : xmax1; + T inter_ymax = ymax1 > ymax2 ? ymax2 : ymax1; + T inter_xmin = xmin1 > xmin2 ? xmin1 : xmin2; + T inter_ymin = ymin1 > ymin2 ? ymin1 : ymin2; + T inter_height = inter_ymax - inter_ymin; + T inter_width = inter_xmax - inter_xmin; + inter_height = inter_height > zero ? inter_height : zero; + inter_width = inter_width > zero ? inter_width : zero; + T inter_area = inter_width * inter_height; + T union_area = area1 + area2 - inter_area; + T sim_score = inter_area / union_area; + return sim_score; +} + +template +struct IOUSimilarityFunctor { + IOUSimilarityFunctor(const T* x, const T* y, T* z, int cols) + : x_(x), y_(y), z_(z), cols_(static_cast(cols)) {} + + inline HOSTDEVICE void operator()(size_t row_id) const { + T x_min1 = x_[row_id * 4]; + T y_min1 = x_[row_id * 4 + 1]; + T x_max1 = x_[row_id * 4 + 2]; + T y_max1 = x_[row_id * 4 + 3]; + for (size_t i = 0; i < cols_; ++i) { + T x_min2 = y_[i * 4]; + T y_min2 = y_[i * 4 + 1]; + T x_max2 = y_[i * 4 + 2]; + T y_max2 = y_[i * 4 + 3]; + + T sim = IOUSimilarity(x_min1, y_min1, x_max1, y_max1, x_min2, y_min2, + x_max2, y_max2); + + z_[row_id * cols_ + i] = sim; + } + } + const T* x_; + const T* y_; + T* z_; + const size_t cols_; +}; + +namespace paddle { +namespace operators { + +template +class IOUSimilarityKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + const framework::LoDTensor* in_x = ctx.Input("X"); + const framework::Tensor* in_y = ctx.Input("Y"); + framework::LoDTensor* out = ctx.Output("Out"); + + int x_n = in_x->dims()[0]; + int y_n = in_y->dims()[0]; + IOUSimilarityFunctor functor(in_x->data(), in_y->data(), + out->mutable_data(ctx.GetPlace()), y_n); + + platform::ForRange for_range( + static_cast(ctx.device_context()), x_n); + for_range(functor); + } +}; // namespace operators + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/label_smooth_op.cc b/paddle/operators/label_smooth_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..c89082f44b360cbd171eccb212674040b8688a46 --- /dev/null +++ b/paddle/operators/label_smooth_op.cc @@ -0,0 +1,128 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/label_smooth_op.h" + +namespace paddle { +namespace operators { + +class LabelSmoothOp : public framework::OperatorWithKernel { + public: + LabelSmoothOp(const std::string &type, + const framework::VariableNameMap &inputs, + const framework::VariableNameMap &outputs, + const framework::AttributeMap &attrs) + : OperatorWithKernel(type, inputs, outputs, attrs) {} + + void InferShape(framework::InferShapeContext *ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of LabelSmoothOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Out"), + "Output(Out) of LabelSmoothOp should not be null."); + auto in_dims = ctx->GetInputDim("X"); + if (ctx->HasInput("PriorDist")) { + auto noise_dims = ctx->GetInputDim("PriorDist"); + auto noise_numel = paddle::framework::product(noise_dims); + PADDLE_ENFORCE( + in_dims[1] == noise_numel, + "The number of elements in Input(PriorDist) must be equal to the " + "dimension of each label."); + } + ctx->ShareLoD("X", /*->*/ "Out"); + ctx->SetOutputDim("Out", in_dims); + } +}; + +class LabelSmoothOpMaker : public framework::OpProtoAndCheckerMaker { + public: + LabelSmoothOpMaker(OpProto *proto, OpAttrChecker *op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", + "(LoDTensor) The input labels of LabelSmooth operator. This " + "input can be batched labels in one-hot encoding or output from " + "softmax, with shape [N x K], where N is the batch size and K is " + "the number of classes"); + AddInput("PriorDist", + "(Tensor, optional)" + "The prior distribution to be added to the smoothed label. It is " + "fixed during training and the number of elements should be equal " + "to the dimension K of each label. Default is uniform " + "distribution and each element will be set to 1/K if not provided " + "in input.") + .AsDispensable(); + AddOutput("Out", + "(loDTensor) The smoothed label of LabelSmooth operator. It has" + "the same shape and LoD with the Input(LoDTensor)."); + AddAttr("epsilon", + "(float, default 0.0f)" + "The smoothing parameter of LabelSmooth operator.") + .SetDefault(0.0f); + AddComment(R"DOC( +LabelSmooth Operator. + +Label smoothing is a mechanism to regularize the classifier layer. In machine +learning, optimizing the log-likelihood of the correct label directly may +cause two problems. First, it may result in overfitting: if the model learns +to assign full probability to the ground-truth label for each training example, +it is not guaranteed to generalize. Second, it encourages the differences +between the largest logit and all others to become large, reducing the ability +of the model to adapt. Label smoothing is proposed to encourage the model to +be less confident, which replaces the ground-truth label $y$ with the weighted +sum of itself and some fixed distribution $\mu$, i.e. + +$$ + \tilde{y} = (1 - \epsilon) * y + \epsilon * \mu, +$$ + +where $(1 - \epsilon)$ and $\epsilon$ are the weights respectively, and +$\tilde{y}$ is the smoothed label. Usually uniform distribution is used for +$\mu$. This change in the ground-truth label is called label-smoothing +regularization or LSR. + +See more details about label smoothing in https://arxiv.org/abs/1512.00567. + +)DOC"); + } +}; + +class LabelSmoothGradOp : public framework::OperatorWithKernel { + public: + LabelSmoothGradOp(const std::string &type, + const framework::VariableNameMap &inputs, + const framework::VariableNameMap &outputs, + const framework::AttributeMap &attrs) + : OperatorWithKernel(type, inputs, outputs, attrs) {} + + void InferShape(framework::InferShapeContext *ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), "Input(X) shouldn't be null."); + PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Out")), + "Input(Out@GRAD) shouldn't be null."); + ctx->SetOutputDim(framework::GradVarName("X"), ctx->GetInputDim("X")); + } +}; + +} // namespace operators +} // namespace paddle +namespace ops = paddle::operators; + +REGISTER_OP(label_smooth, ops::LabelSmoothOp, ops::LabelSmoothOpMaker, + label_smooth_grad, ops::LabelSmoothGradOp); +REGISTER_OP_CPU_KERNEL( + label_smooth, + ops::LabelSmoothKernel, + ops::LabelSmoothKernel); +REGISTER_OP_CPU_KERNEL( + label_smooth_grad, + ops::LabelSmoothGradKernel, + ops::LabelSmoothGradKernel); diff --git a/paddle/operators/label_smooth_op.cu b/paddle/operators/label_smooth_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..5a0cec12bc58a56e4b0c3bd6fbc6c4754ef81fa4 --- /dev/null +++ b/paddle/operators/label_smooth_op.cu @@ -0,0 +1,26 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/label_smooth_op.h" + +namespace ops = paddle::operators; + +REGISTER_OP_CUDA_KERNEL( + label_smooth, + ops::LabelSmoothKernel, + ops::LabelSmoothKernel); +REGISTER_OP_CUDA_KERNEL( + label_smooth_grad, + ops::LabelSmoothGradKernel, + ops::LabelSmoothGradKernel); diff --git a/paddle/operators/label_smooth_op.h b/paddle/operators/label_smooth_op.h new file mode 100644 index 0000000000000000000000000000000000000000..87bc9f793e3b4e249142710243c45d51f3a913b2 --- /dev/null +++ b/paddle/operators/label_smooth_op.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include "paddle/framework/eigen.h" +#include "paddle/framework/op_registry.h" + +namespace paddle { +namespace operators { + +template +class LabelSmoothKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const { + auto* out_t = ctx.Output("Out"); + auto* in_t = ctx.Input("X"); + auto* dist_t = ctx.Input("PriorDist"); + auto label_dim = in_t->dims()[1]; + out_t->mutable_data(ctx.GetPlace()); + + auto epsilon = ctx.Attr("epsilon"); + auto out = framework::EigenVector::Flatten(*out_t); + auto in = framework::EigenVector::Flatten(*in_t); + auto& dev = *ctx.template device_context().eigen_device(); + if (dist_t) { + auto dist = framework::EigenVector::Flatten(*dist_t); + out.device(dev) = + static_cast(1 - epsilon) * in + + epsilon * dist.broadcast(Eigen::DSizes(in_t->numel())); + } else { + out.device(dev) = static_cast(1 - epsilon) * in + + static_cast(epsilon / label_dim); + } + } +}; + +template +class LabelSmoothGradKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const { + auto* d_out_t = ctx.Input(framework::GradVarName("Out")); + auto* d_in_t = ctx.Output(framework::GradVarName("X")); + d_in_t->mutable_data(ctx.GetPlace()); + + auto d_out = framework::EigenVector::Flatten(*d_out_t); + auto d_in = framework::EigenVector::Flatten(*d_in_t); + + auto epsilon = ctx.Attr("epsilon"); + auto& dev = *ctx.template device_context().eigen_device(); + d_in.device(dev) = static_cast(1 - epsilon) * d_out; + } +}; +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/layer_norm_op.cc b/paddle/operators/layer_norm_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..1c6d2ae4d05becaeed34d66cad398cc90f9d3ece --- /dev/null +++ b/paddle/operators/layer_norm_op.cc @@ -0,0 +1,370 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/layer_norm_op.h" + +namespace paddle { +namespace operators { + +using Tensor = framework::Tensor; +using LoDTensor = framework::LoDTensor; +using DataLayout = framework::DataLayout; + +template +using EigenMatrixMapRowMajor = Eigen::Map< + Eigen::Matrix>; +template +using ConstEigenMatrixMapRowMajor = Eigen::Map< + const Eigen::Matrix>; + +class LayerNormOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext *ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Y"), + "Output(Y) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Mean"), + "Output(Mean) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Variance"), + "Output(Variance) of LayerNormOp should not be null."); + + auto x_dim = ctx->GetInputDim("X"); + auto begin_norm_axis = ctx->Attrs().Get("begin_norm_axis"); + PADDLE_ENFORCE_LT(begin_norm_axis, x_dim.size(), + "'begin_norm_axis' must be less than the rank of X."); + + auto matrix_dim = framework::flatten_to_2d(x_dim, begin_norm_axis); + int left = static_cast(matrix_dim[0]); + int right = static_cast(matrix_dim[1]); + if (ctx->HasInput("Scale")) { + PADDLE_ENFORCE_EQ(ctx->GetInputDim("Scale").size(), 1UL); + PADDLE_ENFORCE_EQ(ctx->GetInputDim("Scale")[0], right); + } + if (ctx->HasInput("Bias")) { + PADDLE_ENFORCE_EQ(ctx->GetInputDim("Bias").size(), 1UL); + PADDLE_ENFORCE_EQ(ctx->GetInputDim("Bias")[0], right); + } + + ctx->SetOutputDim("Y", ctx->GetInputDim("X")); + ctx->SetOutputDim("Mean", {left}); + ctx->SetOutputDim("Variance", {left}); + ctx->ShareLoD("X", "Y"); + } +}; + +class LayerNormOpMaker : public framework::OpProtoAndCheckerMaker { + public: + LayerNormOpMaker(OpProto *proto, OpAttrChecker *op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", "(LoDTensor) The input tensor."); + AddInput("Scale", + "(Tensor, optional) Scale is a 1-dimensional tensor of size " + "H(`begin_norm_axis` splits the tensor(`X`) to a matrix [N,H])." + "It is applied to the output.") + .AsDispensable(); + AddInput("Bias", + "(Tensor, optional) Bias is a 1-dimensional tensor of size " + "H(`begin_norm_axis` splits the tensor(`X`) to a matrix [N,H])." + "It is applied to the output.") + .AsDispensable(); + AddOutput("Y", "(LoDTensor) Result after normalization."); + AddOutput("Mean", "(Tensor) Mean of the current mini batch.") + .AsIntermediate(); + AddOutput("Variance", "(Tensor) Variance of the current mini batch.") + .AsIntermediate(); + + AddAttr("epsilon", + "(float, default 1e-5) Constant for " + "numerical stability") + .SetDefault(1e-5) + .AddCustomChecker([](const float &epsilon) { + PADDLE_ENFORCE(epsilon >= 0.0f && epsilon <= 0.001f, + "'epsilon' should be between 0.0 and 0.001."); + }); + AddAttr("begin_norm_axis", + "(int default:1), the " + "axis of `begin_norm_axis ... Rank(X) - 1` will be " + "normalized. `begin_norm_axis` splits the tensor(`X`) to a " + "matrix [N,H].") + .SetDefault(1) + .AddCustomChecker([](const int &begin_norm_axis) { + PADDLE_ENFORCE_GT(begin_norm_axis, 0, + "'begin_norm_axis' should be greater than zero."); + }); + + AddComment(R"DOC( +Layer Normalization. + +Layer Norm has been implemented as discussed in the paper: +https://arxiv.org/abs/1607.06450 +... +)DOC"); + } +}; + +template +class LayerNormKernel + : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext &ctx) const override { + const float epsilon = ctx.Attr("epsilon"); + const auto *scale = ctx.Input("Scale"); + const auto *bias = ctx.Input("Bias"); + const auto *x = ctx.Input("X"); + const auto &x_dims = x->dims(); + const auto begin_norm_axis = ctx.Attr("begin_norm_axis"); + + auto *output = ctx.Output("Y"); + auto *mean = ctx.Output("Mean"); + auto *var = ctx.Output("Variance"); + output->mutable_data(ctx.GetPlace()); + mean->mutable_data(ctx.GetPlace()); + var->mutable_data(ctx.GetPlace()); + + auto matrix_dim = framework::flatten_to_2d(x_dims, begin_norm_axis); + int left = static_cast(matrix_dim[0]); + int right = static_cast(matrix_dim[1]); + + auto input_map = ConstEigenMatrixMapRowMajor(x->data(), left, right); + + auto mean_map = EigenMatrixMapRowMajor(mean->data(), left, 1); + auto var_map = EigenMatrixMapRowMajor(var->data(), left, 1); + auto output_map = EigenMatrixMapRowMajor(output->data(), left, right); + + auto squre = [](T ele) { return ele * ele; }; + auto add_epslion = [epsilon](T ele) { return ele + epsilon; }; + + mean_map = input_map.rowwise().mean(); + var_map = (input_map - mean_map.replicate(1, right)) + .unaryExpr(squre) + .rowwise() + .mean() + .unaryExpr(add_epslion); + + auto inv_std_func = [](T ele) { return std::sqrt(1 / ele); }; + // TODO(zcd): Some thinking about output_map, is it appropriate that + // `output_map` and `input_map` point to the same memory. + auto inv_std = var_map.unaryExpr(inv_std_func); + if (scale && bias) { + auto scale_map = + ConstEigenMatrixMapRowMajor(scale->data(), 1, right); + auto bias_map = ConstEigenMatrixMapRowMajor(bias->data(), 1, right); + output_map = (input_map - mean_map.replicate(1, right)) + .cwiseProduct(inv_std.replicate(1, right)) + .cwiseProduct(scale_map.replicate(left, 1)) + + bias_map.replicate(left, 1); + } else if (scale) { + auto scale_map = + ConstEigenMatrixMapRowMajor(scale->data(), 1, right); + output_map = (input_map - mean_map.replicate(1, right)) + .cwiseProduct(inv_std.replicate(1, right)) + .cwiseProduct(scale_map.replicate(left, 1)); + } else if (bias) { + auto bias_map = ConstEigenMatrixMapRowMajor(bias->data(), 1, right); + output_map = (input_map - mean_map.replicate(1, right)) + .cwiseProduct(inv_std.replicate(1, right)) + + bias_map.replicate(left, 1); + } else { + output_map = (input_map - mean_map.replicate(1, right)) + .cwiseProduct(inv_std.replicate(1, right)); + } + } +}; + +class LayerNormGradOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext *ctx) const override { + // check input + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Scale"), + "Input(Scale) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Mean"), + "Input(Mean) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Variance"), + "Input(Variance) of LayerNormOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Y")), + "Input(Y@GRAD) of LayerNormOp should not be null."); + + // check output + if (ctx->HasOutput(framework::GradVarName("X"))) { + ctx->SetOutputDim(framework::GradVarName("X"), ctx->GetInputDim("X")); + } + if (ctx->HasOutput(framework::GradVarName("Scale"))) { + ctx->SetOutputDim(framework::GradVarName("Scale"), + ctx->GetInputDim("Scale")); + } + if (ctx->HasOutput(framework::GradVarName("Bias"))) { + ctx->SetOutputDim(framework::GradVarName("Bias"), + ctx->GetInputDim("Bias")); + } + } + + protected: + framework::OpKernelType GetExpectedKernelType( + const framework::ExecutionContext &ctx) const override { + const auto *var = ctx.InputVar(framework::GradVarName("Y")); + if (var == nullptr) { + PADDLE_THROW("can't find Y@GRAD"); + } + const Tensor *t = nullptr; + if (var->IsType()) { + t = &var->Get(); + } else if (var->IsType()) { + t = &var->Get(); + } + if (t == nullptr) { + PADDLE_THROW("can't find Y@GRAD"); + } + return framework::OpKernelType(framework::ToDataType(t->type()), + ctx.GetPlace()); + } +}; + +template +class LayerNormGradKernel + : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext &ctx) const override { + const auto *x = ctx.Input("X"); + const auto *mean = ctx.Input("Mean"); + const auto *var = ctx.Input("Variance"); + const auto *scale = ctx.Input("Scale"); + const auto *d_y = ctx.Input(framework::GradVarName("Y")); + + const auto &x_dims = x->dims(); + + const auto begin_norm_axis = ctx.Attr("begin_norm_axis"); + auto matrix_dim = framework::flatten_to_2d(x_dims, begin_norm_axis); + int left = static_cast(matrix_dim[0]); + int right = static_cast(matrix_dim[1]); + + // init output + auto *d_x = ctx.Output(framework::GradVarName("X")); + auto *d_scale = ctx.Output(framework::GradVarName("Scale")); + auto *d_bias = ctx.Output(framework::GradVarName("Bias")); + + auto x_map = ConstEigenMatrixMapRowMajor(x->data(), left, right); + auto d_y_map = ConstEigenMatrixMapRowMajor(d_y->data(), left, right); + auto mean_map = ConstEigenMatrixMapRowMajor(mean->data(), left, 1); + auto var_map = ConstEigenMatrixMapRowMajor(var->data(), left, 1); + + if (d_bias) { + d_bias->mutable_data(ctx.GetPlace()); + auto d_bias_map = EigenMatrixMapRowMajor(d_bias->data(), 1, right); + d_bias_map = d_y_map.colwise().sum(); + } + if (d_scale) { + d_scale->mutable_data(ctx.GetPlace()); + auto d_scale_map = + EigenMatrixMapRowMajor(d_scale->data(), 1, right); + auto inv_std_func = [](T ele) { return std::sqrt(1 / ele); }; + // There are two equation to compute d_scale. One uses "Y" and the other + // does not use "Y" + d_scale_map = + ((x_map - mean_map.replicate(1, right)) + .cwiseProduct( + var_map.unaryExpr(inv_std_func).replicate(1, right)) + .cwiseProduct(d_y_map)) + .colwise() + .sum(); + } + + if (d_x) { + d_x->mutable_data(ctx.GetPlace()); + auto d_x_map = EigenMatrixMapRowMajor(d_x->data(), left, right); + auto triple_product_func = [](T ele) { return ele * ele * ele; }; + auto inv_std_func = [](T ele) { return std::sqrt(1 / ele); }; + // TODO(zcd): these code can be refined + if (d_scale) { + auto scale_map = + ConstEigenMatrixMapRowMajor(scale->data(), 1, right); + // dy_dx + auto dx_end = var_map.unaryExpr(inv_std_func) + .replicate(1, right) + .cwiseProduct(d_y_map) + .cwiseProduct(scale_map.replicate(left, 1)); + // dy_dmean_dx + auto dx_mean = (T(-1.0) / right) * + var_map.unaryExpr(inv_std_func) + .replicate(1, right) + .cwiseProduct(d_y_map) + .cwiseProduct(scale_map.replicate(left, 1)) + .rowwise() + .sum() + .replicate(1, right); + // dy_var_dx + auto dvar_end_part = (x_map - mean_map.replicate(1, right)) + .cwiseProduct(scale_map.replicate(left, 1)) + .cwiseProduct(d_y_map) + .rowwise() + .sum(); + auto dvar_end = var_map.unaryExpr(inv_std_func) + .unaryExpr(triple_product_func) + .cwiseProduct(dvar_end_part) + .replicate(1, right); + auto dx_var = + (T(-1.0) / right) * + (x_map - mean_map.replicate(1, right)).cwiseProduct(dvar_end); + + d_x_map = dx_end + dx_mean + dx_var; + } else { + // dy_dx + auto dx_end = var_map.unaryExpr(inv_std_func) + .replicate(1, right) + .cwiseProduct(d_y_map); + // dy_dmean_dx + auto dx_mean = (T(-1.0) / right) * + var_map.unaryExpr(inv_std_func) + .replicate(1, right) + .cwiseProduct(d_y_map) + .rowwise() + .sum() + .replicate(1, right); + // dy_var_dx + auto dvar_end_part = (x_map - mean_map.replicate(1, right)) + .cwiseProduct(d_y_map) + .rowwise() + .sum(); + auto dvar_end = var_map.unaryExpr(inv_std_func) + .unaryExpr(triple_product_func) + .cwiseProduct(dvar_end_part) + .replicate(1, right); + auto dx_var = + (T(-1.0) / right) * + (x_map - mean_map.replicate(1, right)).cwiseProduct(dvar_end); + + d_x_map = dx_end + dx_mean + dx_var; + } + } + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP(layer_norm, ops::LayerNormOp, ops::LayerNormOpMaker, + layer_norm_grad, ops::LayerNormGradOp); +REGISTER_OP_CPU_KERNEL( + layer_norm, + ops::LayerNormKernel); +REGISTER_OP_CPU_KERNEL( + layer_norm_grad, + ops::LayerNormGradKernel); diff --git a/paddle/operators/layer_norm_op.h b/paddle/operators/layer_norm_op.h new file mode 100644 index 0000000000000000000000000000000000000000..bca35b91e6f52d35dee14aac9d080b52914942e3 --- /dev/null +++ b/paddle/operators/layer_norm_op.h @@ -0,0 +1,35 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include "paddle/framework/eigen.h" +#include "paddle/framework/op_registry.h" + +namespace paddle { +namespace operators { + +template +class LayerNormKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override; +}; + +template +class LayerNormGradKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override; +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/linear_chain_crf_op.cc b/paddle/operators/linear_chain_crf_op.cc index 975e394c78db037a125adeb2c86e3c74dc0eb6f8..e24bf622b7f11e61198ab5238f47ba7edff2f4da 100644 --- a/paddle/operators/linear_chain_crf_op.cc +++ b/paddle/operators/linear_chain_crf_op.cc @@ -187,7 +187,7 @@ class LinearChainCRFOp : public framework::OperatorWithKernel { const framework::ExecutionContext& ctx) const override { return framework::OpKernelType( framework::ToDataType(ctx.Input("Emission")->type()), - ctx.device_context()); + platform::CPUPlace()); } }; @@ -248,7 +248,7 @@ class LinearChainCRFGradOp : public framework::OperatorWithKernel { framework::ToDataType( ctx.Input(framework::GradVarName("LogLikelihood")) ->type()), - ctx.device_context()); + platform::CPUPlace()); } }; diff --git a/paddle/operators/linear_chain_crf_op.h b/paddle/operators/linear_chain_crf_op.h index f502ebefde1fbd4b366f76d2915d94a23a124e5f..afc197a1c38091df5bf7d11ef07a4193ad6417cd 100644 --- a/paddle/operators/linear_chain_crf_op.h +++ b/paddle/operators/linear_chain_crf_op.h @@ -65,57 +65,14 @@ class LinearChainCRFOpKernel : public framework::OpKernel { const size_t level = 0; const size_t seq_num = in_lod[level].size() - 1; - // These local variables hold the inputs and outputs, garanteeing them on - // CPU memory, to provide a consistent reference. - // TODO(caoying) Fix this by moving all these local variables into the - // class's data members once we can profile the whole training process. - LoDTensor* emission_weights = nullptr; - LoDTensor emission_weight_tensor; - Tensor* transition_weights = nullptr; - Tensor transition_weight_tensor; - LoDTensor* label = nullptr; - LoDTensor label_tensor; - - Tensor* emission_exps = nullptr; - Tensor emission_exps_tensor; - Tensor* transition_exps = nullptr; - Tensor transition_exps_tensor; - Tensor* alpha = nullptr; - Tensor alpha_tensor; - Tensor* ll = nullptr; - Tensor ll_tensor; - - if (platform::is_gpu_place(ctx.GetPlace())) { - emission_weights = &emission_weight_tensor; - transition_weights = &transition_weight_tensor; - label = &label_tensor; - - CopyInputsToCpuMemory( - ctx.device_context(), *ctx.Input("Emission"), - *ctx.Input("Transition"), *ctx.Input("Label"), - emission_weights, transition_weights, label); - - emission_exps = &emission_exps_tensor; - emission_exps->Resize(emission_weights->dims()); - - transition_exps = &transition_exps_tensor; - transition_exps->Resize(transition_weights->dims()); - - alpha = &alpha_tensor; - alpha->Resize(ctx.Output("Alpha")->dims()); - - ll = &ll_tensor; - } else { - emission_weights = - const_cast(ctx.Input("Emission")); - transition_weights = const_cast(ctx.Input("Transition")); - label = const_cast(ctx.Input("Label")); - - emission_exps = ctx.Output("EmissionExps"); - transition_exps = ctx.Output("TransitionExps"); - alpha = ctx.Output("Alpha"); - ll = ctx.Output("LogLikelihood"); - } + const LoDTensor* emission_weights = ctx.Input("Emission"); + const Tensor* transition_weights = ctx.Input("Transition"); + const LoDTensor* label = ctx.Input("Label"); + + Tensor* emission_exps = ctx.Output("EmissionExps"); + Tensor* transition_exps = ctx.Output("TransitionExps"); + Tensor* alpha = ctx.Output("Alpha"); + Tensor* ll = ctx.Output("LogLikelihood"); // Because the computation codes only runs on CPU, here the memory for all // the outputs is FIXED to be allocated on the CPU memory. @@ -173,61 +130,9 @@ class LinearChainCRFOpKernel : public framework::OpKernel { one_seq, one_seq_row_max, one_seq_exps, *transition_weights, *transition_exps, one_seq_label, &one_seq_alpha); } - - if (platform::is_gpu_place(ctx.GetPlace())) { - CopyOutputsToGpuMemory( - ctx.device_context(), *emission_exps, *transition_exps, *alpha, *ll, - ctx.Output("EmissionExps"), - ctx.Output("TransitionExps"), ctx.Output("Alpha"), - ctx.Output("LogLikelihood")); - } }; private: - void CopyInputsToCpuMemory(const platform::DeviceContext& ctx, - const LoDTensor& emission_weights_src, - const Tensor& transition_weights_src, - const LoDTensor& label_src, - LoDTensor* emission_weights_dst, - Tensor* transition_weights_dst, - LoDTensor* label_dst) const { - // Copy the inputs from GPU memory to CPU memory if this operators runs on - // GPU device. - auto copyLoDTensor = [](const platform::DeviceContext& ctx, - const LoDTensor& src, LoDTensor* dst) { - dst->mutable_data(src.dims(), platform::CPUPlace()); - framework::Copy(src, platform::CPUPlace(), ctx, dst); - }; - - copyLoDTensor(ctx, emission_weights_src, emission_weights_dst); - copyLoDTensor(ctx, label_src, label_dst); - - transition_weights_dst->mutable_data(transition_weights_src.dims(), - platform::CPUPlace()); - framework::Copy(transition_weights_src, platform::CPUPlace(), ctx, - transition_weights_dst); - } - - void CopyOutputsToGpuMemory(const platform::DeviceContext& ctx, - const Tensor& emission_exps_src, - const Tensor& transition_exps_src, - const Tensor& alpha_src, const Tensor& ll_src, - Tensor* emission_exps_dst, - Tensor* transition_exps_dst, Tensor* alpha_dst, - Tensor* ll_dst) const { - // Copy the forward results from CPU memory to GPU memory if this - // operators runs on GPU device. - auto copyTensor = [](const platform::DeviceContext& ctx, const Tensor& src, - Tensor* dst) { - dst->mutable_data(platform::CUDAPlace()); - framework::Copy(src, platform::CUDAPlace(), ctx, dst); - }; - copyTensor(ctx, emission_exps_src, emission_exps_dst); - copyTensor(ctx, transition_exps_src, transition_exps_dst); - copyTensor(ctx, alpha_src, alpha_dst); - copyTensor(ctx, ll_src, ll_dst); - } - T ForwardOneSequence(const Tensor& emission, const Tensor& emission_row_max, const Tensor& emission_exps, const Tensor& trans_weights, const Tensor& trans_weight_exps, const Tensor& label, @@ -296,63 +201,17 @@ class LinearChainCRFGradOpKernel : public framework::OpKernel { auto lod = ctx.Input("Label")->lod(); PADDLE_ENFORCE(lod.size(), "Input(Label) must be a sequence."); - // These local variables hold the inputs and outputs, garanteeing them on - // CPU memory, to provide a consistent reference. - // TODO(caoying) Fix this by moving all these local variables into the - // class's data members once we can profile the training process, or - // implementing a real GPU kernel for CRF. - Tensor* label = nullptr; - Tensor label_tensor; - Tensor* emission_exps = nullptr; - Tensor emission_exps_tensor; - Tensor* transition_exps = nullptr; - Tensor transition_exps_tensor; - Tensor* alpha = nullptr; - Tensor alpha_tensor; - Tensor ll_grad_tensor; - T* ll_grad = nullptr; - - Tensor* emission_grad = nullptr; - Tensor emission_grad_tensor; - Tensor* transition_grad = nullptr; - Tensor transition_grad_tensor; - - if (platform::is_gpu_place(ctx.GetPlace())) { - label = &label_tensor; - emission_exps = &emission_exps_tensor; - transition_exps = &transition_exps_tensor; - alpha = &alpha_tensor; - CopyInputsToCpuMemory( - ctx.device_context(), *ctx.Input("Label"), - *ctx.Input("EmissionExps"), - *ctx.Input("TransitionExps"), *ctx.Input("Alpha"), - *ctx.Input(framework::GradVarName("LogLikelihood")), label, - emission_exps, transition_exps, alpha, &ll_grad_tensor); - ll_grad = ll_grad_tensor.data(); - - if (ctx.Output(framework::GradVarName("Emission"))) { - emission_grad = &emission_grad_tensor; - emission_grad->Resize(emission_exps->dims()); - } + const Tensor* label = ctx.Input("Label"); + const Tensor* emission_exps = ctx.Input("EmissionExps"); + const Tensor* transition_exps = ctx.Input("TransitionExps"); + const Tensor* alpha = ctx.Input("Alpha"); + const T* ll_grad = + ctx.Input(framework::GradVarName("LogLikelihood"))->data(); - if (ctx.Output(framework::GradVarName("Transition"))) { - transition_grad = &transition_grad_tensor; - transition_grad->Resize(transition_exps->dims()); - } - } else { - label = const_cast(ctx.Input("Label")); - emission_exps = const_cast(ctx.Input("EmissionExps")); - transition_exps = - const_cast(ctx.Input("TransitionExps")); - alpha = const_cast(ctx.Input("Alpha")); - ll_grad = const_cast( - ctx.Input(framework::GradVarName("LogLikelihood"))) - ->data(); - - emission_grad = ctx.Output(framework::GradVarName("Emission")); - transition_grad = - ctx.Output(framework::GradVarName("Transition")); - } + Tensor* emission_grad = + ctx.Output(framework::GradVarName("Emission")); + Tensor* transition_grad = + ctx.Output(framework::GradVarName("Transition")); // TODO(caoying) Fix this constraint. When the Input(Emission) is from the // data reader operator, it can have no gradients. @@ -389,58 +248,9 @@ class LinearChainCRFGradOpKernel : public framework::OpKernel { one_seq_emission_exps, *transition_exps, one_seq_alpha, one_seq_label, &one_seq_beta, transition_grad, &one_seq_emission_grad); } - - if (platform::is_gpu_place(ctx.GetPlace())) { - CopyOutputsToGpuMemory( - ctx.device_context(), emission_grad, transition_grad, - ctx.Output(framework::GradVarName("Emission")), - ctx.Output(framework::GradVarName("Transition"))); - } }; private: - void CopyInputsToCpuMemory(const platform::DeviceContext& ctx, - const LoDTensor& label_src, - const Tensor& emission_exps_src, - const Tensor& transition_exps_src, - const Tensor& alpha_src, const Tensor& ll_grad_src, - Tensor* label_dst, Tensor* emission_exps_dst, - Tensor* transition_exps_dst, Tensor* alpha_dst, - Tensor* ll_grad_dst) const { - // Copy the inputs from GPU memory to CPU memory when this operators runs on - // GPU device. - label_dst->mutable_data(label_src.dims(), platform::CPUPlace()); - framework::Copy(label_src, platform::CPUPlace(), ctx, label_dst); - - auto copyTensor = [](const platform::DeviceContext& ctx, const Tensor& src, - Tensor* dst) { - dst->mutable_data(src.dims(), platform::CPUPlace()); - framework::Copy(src, platform::CPUPlace(), ctx, dst); - }; - copyTensor(ctx, emission_exps_src, emission_exps_dst); - copyTensor(ctx, transition_exps_src, transition_exps_dst); - copyTensor(ctx, alpha_src, alpha_dst); - copyTensor(ctx, ll_grad_src, ll_grad_dst); - } - - void CopyOutputsToGpuMemory(const platform::DeviceContext& ctx, - const Tensor* emission_grad_src, - const Tensor* transition_grad_src, - Tensor* emission_grad_dst, - Tensor* transition_grad_dst) const { - // Copy the backward results from CPU memory to GPU - // memory if this operators runs on GPU device. - auto copyTensor = [](const platform::DeviceContext& ctx, const Tensor* src, - Tensor* dst) { - if (src && dst) { - dst->mutable_data(platform::CUDAPlace()); - framework::Copy(*src, platform::CUDAPlace(), ctx, dst); - } - }; - copyTensor(ctx, emission_grad_src, emission_grad_dst); - copyTensor(ctx, transition_grad_src, transition_grad_dst); - } - void BackwardOneSequence(const platform::CPUDeviceContext& ctx, const T ll_grad, const Tensor& emission_exps, const Tensor& transition_exps, const Tensor& alpha, diff --git a/paddle/operators/load_combine_op.cc b/paddle/operators/load_combine_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..f4be793d7bf1f346c011842c57fb5b5179a697d6 --- /dev/null +++ b/paddle/operators/load_combine_op.cc @@ -0,0 +1,108 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ +#include + +#include "paddle/framework/op_registry.h" +#include "paddle/platform/device_context.h" + +namespace paddle { +namespace operators { + +class LoadCombineOp : public framework::OperatorBase { + public: + LoadCombineOp(const std::string &type, + const framework::VariableNameMap &inputs, + const framework::VariableNameMap &outputs, + const framework::AttributeMap &attrs) + : OperatorBase(type, inputs, outputs, attrs) {} + void Run(const framework::Scope &scope, + const platform::Place &place) const override { + auto filename = Attr("file_path"); + + std::ifstream fin(filename); + PADDLE_ENFORCE(static_cast(fin), + "Cannot open file %s for load_combine op", filename); + + auto out_var_names = Outputs("Out"); + PADDLE_ENFORCE_GT( + static_cast(out_var_names.size()), 0, + "The number of output variables should be greater than 0."); + + platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); + auto &dev_ctx = *pool.Get(place); + + for (size_t i = 0; i < out_var_names.size(); i++) { + auto *out_var = scope.FindVar(out_var_names[i]); + + PADDLE_ENFORCE(out_var != nullptr, "Output variable %s cannot be found", + out_var_names[i]); + + auto *tensor = out_var->GetMutable(); + + // Error checking + PADDLE_ENFORCE(static_cast(fin), "Cannot read more from file %s", + filename); + + // Get data from fin to tensor + DeserializeFromStream(fin, tensor, dev_ctx); + + if (platform::is_gpu_place(place)) { + // copy CPU to GPU + framework::LoDTensor cpu_tensor; + cpu_tensor.ShareDataWith(*tensor); + cpu_tensor.set_lod(tensor->lod()); + + // reset tensor + out_var->Clear(); + tensor = out_var->GetMutable(); + tensor->set_lod(cpu_tensor.lod()); + Copy(cpu_tensor, place, dev_ctx, tensor); + } + } + } +}; + +class LoadCombineOpProtoMaker : public framework::OpProtoAndCheckerMaker { + public: + LoadCombineOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddOutput( + "Out", + "(vector) The output LoDTensors that will be read from the input file.") + .AsDuplicable(); + AddAttr("file_path", + "(string) " + "LoDTensors will be loaded from \"file_path\".") + .AddCustomChecker( + [](const std::string &path) { return !path.empty(); }); + AddComment(R"DOC( +LoadCombine Operator. + +LoadCombine operator loads LoDTensor variables from a file. The file should +contain one or more LoDTensors serialized using the SaveCombine operator. The +LoadCombine operator applies a deserialization strategy to appropriately load +the LodTensors, and this strategy complements the serialization strategy used +in the SaveCombine operator. Hence, the LoadCombine operator is tightly coupled +with the SaveCombine operator, and can only deserialize one or more LoDTensors +that were saved using the SaveCombine operator. + +)DOC"); + } +}; +} // namespace operators +} // namespace paddle +namespace ops = paddle::operators; + +REGISTER_OPERATOR(load_combine, ops::LoadCombineOp, + ops::LoadCombineOpProtoMaker); diff --git a/paddle/operators/lookup_table_op.cc b/paddle/operators/lookup_table_op.cc index bb03def4391da80c6219f7863d300fd3c8d8c7ac..2405852f53d46356a474897d3a111d1c94eed081 100644 --- a/paddle/operators/lookup_table_op.cc +++ b/paddle/operators/lookup_table_op.cc @@ -66,6 +66,12 @@ class LookupTableOpMaker : public framework::OpProtoAndCheckerMaker { "(boolean, default false) " "Sparse update") .SetDefault(false); + AddAttr("padding_idx", + "(int64, default -1) " + "If the value is -1, it makes no effect to lookup. " + "Otherwise the given value indicates padding the output " + "with zeros whenever lookup encounters it in Ids.") + .SetDefault(-1); AddComment(R"DOC( Lookup Table Operator. diff --git a/paddle/operators/lookup_table_op.cu b/paddle/operators/lookup_table_op.cu index 261a28da694bf551d8d9e630139680aebc4be51a..07372808bbf078bd2e9b0bb5782b95a046253f46 100644 --- a/paddle/operators/lookup_table_op.cu +++ b/paddle/operators/lookup_table_op.cu @@ -21,9 +21,11 @@ limitations under the License. */ namespace paddle { namespace operators { -template +template __global__ void LookupTable(T* output, const T* table, const int64_t* ids, - const int64_t N, const int64_t K, const int64_t D) { + const int64_t N, const int64_t K, const int64_t D, + const int64_t padding_idx) { int idx = threadIdx.x; int idy = blockIdx.x + threadIdx.y * GridDimX; @@ -34,7 +36,14 @@ __global__ void LookupTable(T* output, const T* table, const int64_t* ids, T* out = output + idy * D; const T* tab = table + id * D; for (int i = idx; i < D; i += BlockDimX) { - out[i] = tab[i]; + if (PaddingFlag) { + if (id == padding_idx) + out[i] = static_cast(0); + else + out[i] = tab[i]; + } else { + out[i] = tab[i]; + } } idy += BlockDimY * GridDimX; } @@ -67,6 +76,7 @@ class LookupTableCUDAKernel : public framework::OpKernel { auto* table_t = context.Input("W"); auto* ids_t = context.Input("Ids"); auto* output_t = context.Output("Out"); + int64_t padding_idx = context.Attr("padding_idx"); size_t N = table_t->dims()[0]; size_t D = table_t->dims()[1]; @@ -77,10 +87,17 @@ class LookupTableCUDAKernel : public framework::OpKernel { dim3 threads(128, 8); dim3 grids(8, 1); - LookupTable< - T, 128, 8, - 8><<>>( - output, table, ids, N, K, D); + + if (padding_idx == -1) + LookupTable< + T, 128, 8, 8, + false><<>>( + output, table, ids, N, K, D, padding_idx); + else + LookupTable< + T, 128, 8, 8, + true><<>>( + output, table, ids, N, K, D, padding_idx); } }; @@ -91,6 +108,8 @@ class LookupTableGradCUDAKernel : public framework::OpKernel { auto& dev_ctx = context.template device_context(); bool is_sparse = context.Attr("is_sparse"); + // Since paddings are not trainable and fixed in forward, the gradient of + // paddings makes no sense and we don't deal with it in backward. if (is_sparse) { auto* ids = context.Input("Ids"); auto* table = context.Input("W"); @@ -106,8 +125,8 @@ class LookupTableGradCUDAKernel : public framework::OpKernel { new_rows.resize(ids_dim[0]); auto gpu_place = boost::get(context.GetPlace()); - memory::Copy(platform::CPUPlace(), new_rows.data(), gpu_place, ids_data, - ids_dim[0] * sizeof(int64_t), stream); + memory::Copy(platform::CPUPlace(), new_rows.cuda_data(), gpu_place, + ids_data, ids_dim[0] * sizeof(int64_t), stream); d_table->set_rows(new_rows); diff --git a/paddle/operators/lookup_table_op.h b/paddle/operators/lookup_table_op.h index 2fd3335868406455ec01f9ded6bacc7bda5e2a67..0842c422f7bfd3cad9b36dfdbab930f3cc4a8728 100644 --- a/paddle/operators/lookup_table_op.h +++ b/paddle/operators/lookup_table_op.h @@ -32,16 +32,30 @@ class LookupTableKernel : public framework::OpKernel { auto* table_t = context.Input("W"); // float tensor auto* ids_t = context.Input("Ids"); // int tensor auto* output_t = context.Output("Out"); // float tensor + int64_t padding_idx = context.Attr("padding_idx"); int N = table_t->dims()[0]; int D = table_t->dims()[1]; auto* ids = ids_t->data(); auto* table = table_t->data(); auto* output = output_t->mutable_data(context.GetPlace()); - for (int64_t i = 0; i < ids_t->numel(); ++i) { - PADDLE_ENFORCE_LT(ids[i], N); - PADDLE_ENFORCE_GE(ids[i], 0); - memcpy(output + i * D, table + ids[i] * D, D * sizeof(T)); + + if (padding_idx == -1) { + for (int64_t i = 0; i < ids_t->numel(); ++i) { + PADDLE_ENFORCE_LT(ids[i], N); + PADDLE_ENFORCE_GE(ids[i], 0); + memcpy(output + i * D, table + ids[i] * D, D * sizeof(T)); + } + } else { + for (int64_t i = 0; i < ids_t->numel(); ++i) { + if (ids[i] == padding_idx) { + memset(output + i * D, 0, D * sizeof(T)); + } else { + PADDLE_ENFORCE_LT(ids[i], N); + PADDLE_ENFORCE_GE(ids[i], 0); + memcpy(output + i * D, table + ids[i] * D, D * sizeof(T)); + } + } } } }; @@ -51,6 +65,8 @@ class LookupTableGradKernel : public framework::OpKernel { public: void Compute(const framework::ExecutionContext& context) const override { bool is_sparse = context.Attr("is_sparse"); + // Since paddings are not trainable and fixed in forward, the gradient of + // paddings makes no sense and we don't deal with it in backward. if (is_sparse) { auto* ids = context.Input("Ids"); auto* table = context.Input("W"); diff --git a/paddle/operators/lstm_op.cc b/paddle/operators/lstm_op.cc index 3b90b64b4effacf7240fb1bee8c0aa44251ad727..afb095a04e73c2f09b828c01630ef2347ff49613 100644 --- a/paddle/operators/lstm_op.cc +++ b/paddle/operators/lstm_op.cc @@ -117,7 +117,7 @@ class LSTMOpMaker : public framework::OpProtoAndCheckerMaker { AddInput("C0", "(Tensor, optional) the initial cell state is an optional " "input. This is a tensor with shape (N x D), where N is the " - "batch size. `H0` and `C0` can be NULL but only at the same time") + "batch size. `H0` and `C0` can be NULL but only at the same time.") .AsDispensable(); AddInput("Weight", "(Tensor) the learnable hidden-hidden weights." diff --git a/paddle/operators/lstm_op.h b/paddle/operators/lstm_op.h index c57ee414dc5b3417549c8ac3a7fd57a9c8f452df..72e95b75e29c88c5944607ceaa40435bac7a745c 100644 --- a/paddle/operators/lstm_op.h +++ b/paddle/operators/lstm_op.h @@ -27,11 +27,12 @@ using Tensor = framework::Tensor; template inline void ReorderInitState(const DeviceContext& ctx, - const framework::Tensor& src, const size_t* index, + const framework::Tensor& src, + framework::Vector index_lod, framework::Tensor* dst, bool indexed_src) { math::CopyMatrixRowsFunctor row_shuffle; dst->mutable_data(src.dims(), ctx.GetPlace()); - row_shuffle(ctx, src, index, *dst, indexed_src); + row_shuffle(ctx, src, index_lod, *dst, indexed_src); } template @@ -84,7 +85,9 @@ class LSTMKernel : public framework::OpKernel { } lstm_value.prev_state_value = nullptr; Tensor ordered_c0; - const size_t* order = batch_gate->lod()[2].data(); + + framework::Vector order(batch_gate->lod()[2]); + if (cell_t0) { // Since the batch computing for LSTM reorders the input sequence // according to their length. The initialized cell state also needs @@ -202,7 +205,8 @@ class LSTMGradKernel : public framework::OpKernel { // ordered_h0_g/c0_g is the reordered gradient of hidden/cell // initialization. Tensor ordered_h0, ordered_c0, ordered_h0_g, ordered_c0_g; - const size_t* order = batch_gate->lod()[2].data(); + framework::Vector order(batch_gate->lod()[2]); + if (c0) { ReorderInitState(device_ctx, *c0, order, &ordered_c0, true); diff --git a/paddle/operators/lstmp_op.cc b/paddle/operators/lstmp_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..c96b30ba353fabc48630258ea8f88f741b8c415e --- /dev/null +++ b/paddle/operators/lstmp_op.cc @@ -0,0 +1,331 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/lstmp_op.h" + +namespace paddle { +namespace operators { + +class LSTMPOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("Input"), + "Input(Input) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Weight"), + "Input(Weight) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("ProjWeight"), + "Input(ProjWeight) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Bias"), + "Input(Bias) of LSTMP operator should not be null."); + + PADDLE_ENFORCE(ctx->HasOutput("Projection"), + "Output(Projection) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Cell"), + "Output(Cell) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("BatchGate"), + "Output(BatchGate) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("BatchCellPreAct"), + "Output(BatchCellPreAct) of LSTMP operator should not be " + "null."); + PADDLE_ENFORCE(ctx->HasOutput("BatchHidden"), + "Output(BatchHidden) of LSTMP operator should not be null."); + + auto in_dims = ctx->GetInputDim("Input"); + PADDLE_ENFORCE_EQ(in_dims.size(), 2, + "Input(X)'s rank of LSTMP operator must be 2."); + + int frame_size = in_dims[1] / 4; + auto w_dims = ctx->GetInputDim("Weight"); + auto proj_dims = ctx->GetInputDim("ProjWeight"); + PADDLE_ENFORCE_EQ(w_dims.size(), 2, + "The rank of Input(Weight) should be 2."); + PADDLE_ENFORCE_EQ(w_dims[0], proj_dims[1], + "The first dimension of Input(Weight) " + "should be %d.", + proj_dims[1]); + PADDLE_ENFORCE_EQ(w_dims[1], 4 * frame_size, + "The second dimension of Input(Weight) " + "should be 4 * %d.", + frame_size); + + PADDLE_ENFORCE_EQ(proj_dims.size(), 2, + "The rank of Input(ProjWeight) should be 2."); + PADDLE_ENFORCE_EQ(proj_dims[0], frame_size, + "The first dimension of Input(ProjWeight) " + "should be %d.", + frame_size); + + if (ctx->HasInput("H0")) { + PADDLE_ENFORCE(ctx->HasInput("C0"), + "Input(C0) of LSTMP operator should not be null after " + "Input(H0) provided."); + auto h_dims = ctx->GetInputDim("H0"); + auto c_dims = ctx->GetInputDim("C0"); + PADDLE_ENFORCE(h_dims == c_dims, + "The dimension of Input(H0) and Input(C0) " + "should be the same."); + ctx->SetOutputDim("OrderedP0", {h_dims[0], proj_dims[1]}); + } + + auto b_dims = ctx->GetInputDim("Bias"); + PADDLE_ENFORCE_EQ(b_dims.size(), 2, "The rank of Input(Bias) should be 2."); + PADDLE_ENFORCE_EQ(b_dims[0], 1, + "The first dimension of Input(Bias) should be 1."); + + if (ctx->Attrs().Get("use_peepholes")) { + PADDLE_ENFORCE_EQ(b_dims[1], 7 * frame_size, + "The second dimension of Input(Bias) should be " + "7 * %d if enable peepholes connection", + frame_size); + } else { + PADDLE_ENFORCE_EQ(b_dims[1], 4 * frame_size, + "The second dimension of Input(Bias) should be " + "4 * %d if disable peepholes connection", + frame_size); + } + + framework::DDim out_dims({in_dims[0], frame_size}); + framework::DDim proj_out_dims({in_dims[0], proj_dims[1]}); + ctx->SetOutputDim("Projection", proj_out_dims); + ctx->SetOutputDim("Cell", out_dims); + ctx->SetOutputDim("BatchGate", in_dims); + ctx->SetOutputDim("BatchCellPreAct", out_dims); + ctx->SetOutputDim("BatchHidden", out_dims); + ctx->ShareLoD("Input", "Projection"); + ctx->ShareLoD("Input", "Cell"); + } + + protected: + framework::OpKernelType GetExpectedKernelType( + const framework::ExecutionContext& ctx) const override { + return framework::OpKernelType( + framework::ToDataType(ctx.Input("Input")->type()), + ctx.device_context()); + } +}; + +class LSTMPOpMaker : public framework::OpProtoAndCheckerMaker { + public: + LSTMPOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("Input", + "(LoDTensor) the input for sequence data, which supports " + "variable-time length input sequence. The underlying tensor in " + "this LoDTensor is a matrix with shape (T X 4D), where T is the " + "total time steps in this mini-batch, D is the hidden size."); + AddInput("H0", + "(Tensor, optional) the initial hidden state is an optional " + "input. This is a tensor with shape (N x D), where N is the " + "batch size and D is the hidden size.") + .AsDispensable(); + AddInput("C0", + "(Tensor, optional) the initial cell state is an optional " + "input. This is a tensor with shape (N x D), where N is the " + "batch size. `C0` should not be null if `H0` provided.") + .AsDispensable(); + AddInput("Weight", + "(Tensor) the learnable hidden-hidden weights." + " - The shape is (P x 4D), where P is the projection layer size " + "and D is the hidden size." + " - Weight = {W_cr, W_ir, W_fr, W_or}"); + AddInput("ProjWeight", + "(Tensor) the learnable weight of the projection layer." + " - The shape is (D x P), where P is the recurrent projection " + "layer size and D is the hidden size." + " - ProjWeight = {W_rh}"); + AddInput("Bias", + "(Tensor) the learnable biases, which contains two parts: " + "input-hidden biases and peephole connections weights if " + "setting `use_peepholes` to `True`. " + "1. `use_peepholes = False` " + " - The shape is (1 x 4D). " + " - Bias = {b_c, b_i, b_f, b_o}." + "2. `use_peepholes = True` " + " - The shape is (1 x 7D). " + " - Bias = {b_c, b_i, b_f, b_o, W_ic, W_fc, W_oc}."); + AddOutput("Projection", + "(LoDTensor) the projection of the hidden state of LSTMP " + "operator. The shape is (T x P), and LoD is the same with the " + "`Input`."); + AddOutput("Cell", + "(LoDTensor) the cell state of LSTMP operator. " + "The shape is (T x D), and lod is the same with the `Input`."); + AddOutput("BatchGate", + "(LoDTensor) This LoDTensor contains input gate, forget gate " + "and output gate after the activations. This LoDTensor has the " + "same shape as the reorganized input, which is also be called " + "batch input. The LoD size is 2. The first-level LoD is the " + "batch offsets and the second contains the indices, which " + "denotes the position of reorganized sequence in the raw input.") + .AsIntermediate(); + AddOutput("BatchCellPreAct", + "(LoDTensor) the pre-activation cell state reorganized in batch. " + "This LoDTensor is obtained in the forward and used in the " + "backward.") + .AsIntermediate(); + AddOutput("BatchHidden", + "(LoDTensor) the hidden state reorganized in batch. " + "This LoDTensor is obtained in the forward and used in the " + "backward.") + .AsIntermediate(); + AddOutput("OrderedP0", + "(Tensor) the projection of the initial hidden state " + "H0. This is a tensor with shape (N x P), where N is the " + "batch size and P is the hidden size.") + .AsIntermediate(); + AddAttr("use_peepholes", + "(bool, defalut: True) " + "whether to enable diagonal/peephole connections.") + .SetDefault(true); + AddAttr("is_reverse", + "(bool, defalut: False) " + "whether to compute reversed LSTMP.") + .SetDefault(false); + AddAttr( + "gate_activation", + "(string, default: sigmoid)" + "The activation for input gate, forget gate and output " + "gate, `sigmoid` by default.") + .SetDefault("sigmoid") + .InEnum({"sigmoid", "tanh", "relu", "identity"}); + AddAttr("cell_activation", + "(string, default: tanh)" + "The activation for cell output, `tanh` by defalut.") + .SetDefault("tanh") + .InEnum({"sigmoid", "tanh", "relu", "identity"}); + AddAttr("candidate_activation", + "(string, default: tanh)" + "The activation for candidate hidden state, " + "`tanh` by default.") + .SetDefault("tanh") + .InEnum({"sigmoid", "tanh", "relu", "identity"}); + AddAttr("proj_activation", + "(string, default: tanh)" + "The activation for projection output, " + "`tanh` by defalut.") + .SetDefault("tanh") + .InEnum({"sigmoid", "tanh", "relu", "identity"}); + AddComment(R"DOC( +Long-Short Term Memory with recurrent Projection layer (LSTMP) Operator. + +LSTMP has a separate projection layer after the LSTM layer, projecting the +original hidden state to a lower-dimensional one, which is proposed to reduce +the number of total parameters and furthermore computational complexity for +the LSTM, espeacially for the case that the size of output units is relative +large (https://research.google.com/pubs/archive/43905.pdf). + +The formula is as follows: + +$$ +i_t = \sigma(W_{ix}x_{t} + W_{ir}r_{t-1} + W_{ic}c_{t-1} + b_i) \\ + +f_t = \sigma(W_{fx}x_{t} + W_{fr}r_{t-1} + W_{fc}c_{t-1} + b_f) \\ + +\tilde{c_t} = act_g(W_{cx}x_t + W_{cr}r_{t-1} + b_c) \\ + +o_t = \sigma(W_{ox}x_{t} + W_{or}r_{t-1} + W_{oc}c_t + b_o) \\ + +c_t = f_t \odot c_{t-1} + i_t \odot \tilde{c_t} \\ + +h_t = o_t \odot act_h(c_t) \\ + +r_t = \overline{act_h}(W_{rh}h_t) +$$ + +where the W terms denote weight matrices (e.g. $W_{xi}$ is the matrix +of weights from the input gate to the input), $W_{ic}, W_{fc}, W_{oc}$ +are diagonal weight matrices for peephole connections. In our implementation, +we use vectors to reprenset these diagonal weight matrices. The b terms +denote bias vectors ($b_i$ is the input gate bias vector), $\sigma$ +is the activation, such as logistic sigmoid function, and +$i, f, o$ and $c$ are the input gate, forget gate, output gate, +and cell activation vectors, respectively, all of which have the same size as +the cell output activation vector $h$. Here $h$ is usually called the hidden +state and $r$ denotes its recurrent projection. And $\tilde{c_t}$ is also +called the candidate hidden state, whose computation is based on the current +input and previous hidden state. + +The $\odot$ is the element-wise product of the vectors. $act_g$ and $act_h$ +are the cell input and cell output activation functions and `tanh` is usually +used for them. $\overline{act_h}$ is the activation function for the +projection output, usually using `identity` or same as $act_h$. + +Note that these $W_{xi}x_{t}, W_{xf}x_{t}, W_{xc}x_{t}, W_{xo}x_{t}$ +operations on the input $x_{t}$ are NOT included in this operator. +Users can choose to use fully-connected operator before LSTMP operator. + +)DOC"); + } +}; + +class LSTMPGradOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("Input"), + "Input(Input) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Projection"), + "Input(Projection) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Cell"), + "Input(Cell) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Weight"), + "Input(Weight) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("ProjWeight"), + "Input(ProjWeight) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Bias"), + "Input(Bias) of LSTMP operator should not be null."); + + PADDLE_ENFORCE(ctx->HasInput("BatchGate"), + "Input(BatchGate) of LSTMP operator should not be null."); + PADDLE_ENFORCE(ctx->HasInput("BatchCellPreAct"), + "Input(BatchGate) of LSTMP operator should not be null."); + + auto SetOutGradDim = [&ctx](const std::string& name) { + auto g_name = framework::GradVarName(name); + if (ctx->HasOutput(g_name)) + ctx->SetOutputDim(g_name, ctx->GetInputDim(name)); + }; + + SetOutGradDim("Input"); + SetOutGradDim("Weight"); + SetOutGradDim("ProjWeight"); + SetOutGradDim("Bias"); + SetOutGradDim("H0"); + SetOutGradDim("C0"); + } + + protected: + framework::OpKernelType GetExpectedKernelType( + const framework::ExecutionContext& ctx) const override { + return framework::OpKernelType( + framework::ToDataType(ctx.Input("Input")->type()), + ctx.device_context()); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP(lstmp, ops::LSTMPOp, ops::LSTMPOpMaker, lstmp_grad, + ops::LSTMPGradOp); +REGISTER_OP_CPU_KERNEL( + lstmp, ops::LSTMPKernel, + ops::LSTMPKernel); +REGISTER_OP_CPU_KERNEL( + lstmp_grad, ops::LSTMPGradKernel, + ops::LSTMPGradKernel); diff --git a/paddle/operators/lstmp_op.cu b/paddle/operators/lstmp_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..7fcbcfecc871976fdfbfffbbb4e0243b91351a29 --- /dev/null +++ b/paddle/operators/lstmp_op.cu @@ -0,0 +1,24 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/lstmp_op.h" + +namespace ops = paddle::operators; +REGISTER_OP_CUDA_KERNEL( + lstmp, ops::LSTMPKernel, + ops::LSTMPKernel); +REGISTER_OP_CUDA_KERNEL( + lstmp_grad, + ops::LSTMPGradKernel, + ops::LSTMPGradKernel); diff --git a/paddle/operators/lstmp_op.h b/paddle/operators/lstmp_op.h new file mode 100644 index 0000000000000000000000000000000000000000..e064a155dfadd8104fa80727a962cb2e24ade29f --- /dev/null +++ b/paddle/operators/lstmp_op.h @@ -0,0 +1,496 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include "paddle/operators/activation_op.h" +#include "paddle/operators/math/detail/activation_functions.h" +#include "paddle/operators/math/lstm_compute.h" +#include "paddle/operators/math/math_function.h" +#include "paddle/operators/math/sequence2batch.h" + +#include "paddle/framework/eigen.h" +#include "paddle/framework/op_registry.h" + +namespace paddle { +namespace operators { + +using LoDTensor = framework::LoDTensor; +using Tensor = framework::Tensor; + +template +using EigenMatrix = framework::EigenMatrix; + +template +inline void ReorderInitState(const DeviceContext& ctx, + const framework::Tensor& src, + framework::Vector index, + framework::Tensor* dst, bool indexed_src) { + math::CopyMatrixRowsFunctor row_shuffle; + dst->mutable_data(src.dims(), ctx.GetPlace()); + row_shuffle(ctx, src, index, *dst, indexed_src); +} + +template +class LSTMPKernel : public framework::OpKernel { + public: + template + void ActCompute(const math::detail::ActivationType act_type, const Device& d, + X x, Y y) const { + if (act_type == math::detail::ActivationType::kIdentity) + y.device(d) = x; + else if (act_type == math::detail::ActivationType::kSigmoid) + SigmoidFunctor()(d, x, y); + else if (act_type == math::detail::ActivationType::kTanh) + TanhFunctor()(d, x, y); + else if (act_type == math::detail::ActivationType::kReLU) + ReluFunctor()(d, x, y); + else + PADDLE_THROW("unsupported activation type"); + } + + void Compute(const framework::ExecutionContext& ctx) const override { + auto* input = ctx.Input("Input"); + auto* weight = ctx.Input("Weight"); + auto* proj_weight = ctx.Input("ProjWeight"); + auto* bias = ctx.Input("Bias"); + + auto* hidden_t0 = ctx.Input("H0"); + auto* ordered_proj0 = ctx.Output("OrderedP0"); + auto* cell_t0 = ctx.Input("C0"); + + auto* batch_gate = ctx.Output("BatchGate"); + batch_gate->mutable_data(ctx.GetPlace()); + auto* proj_out = ctx.Output("Projection"); + proj_out->mutable_data(ctx.GetPlace()); + auto* cell_out = ctx.Output("Cell"); + cell_out->mutable_data(ctx.GetPlace()); + + bool is_reverse = ctx.Attr("is_reverse"); + math::LoDTensor2BatchFunctor to_batch; + auto& device_ctx = ctx.template device_context(); + to_batch(device_ctx, *input, *batch_gate, true, is_reverse); + + auto in_dims = input->dims(); + int frame_size = static_cast(in_dims[1] / 4); + framework::DDim dims({in_dims[0], frame_size}); + framework::DDim proj_dims({in_dims[0], proj_weight->dims()[1]}); + + if (bias) { + Tensor b = *bias; + b.Resize({bias->numel(), 1}); + Tensor gate_bias = b.Slice(0, 4 * frame_size); + math::RowwiseAdd add_bias; + add_bias(device_ctx, *batch_gate, gate_bias, batch_gate); + } + + math::LstmMetaValue lstmp_value; + if (bias && ctx.Attr("use_peepholes")) { + T* bias_data = const_cast(bias->data()); + // the code style in LstmpMetaValue will be updated later. + + lstmp_value.check_ig = bias_data + 4 * frame_size; + lstmp_value.check_fg = lstmp_value.check_ig + frame_size; + lstmp_value.check_og = lstmp_value.check_fg + frame_size; + } else { + lstmp_value.check_ig = nullptr; + lstmp_value.check_fg = nullptr; + lstmp_value.check_og = nullptr; + } + lstmp_value.prev_state_value = nullptr; + Tensor ordered_c0; + + framework::Vector order(batch_gate->lod()[2]); + + if (cell_t0) { + // Since the batch computing for LSTMP reorders the input sequence + // according to their length. The initialized cell state also needs + // to reorder. + ReorderInitState(device_ctx, *cell_t0, order, + &ordered_c0, true); + lstmp_value.prev_state_value = ordered_c0.data(); + } + + // Use the local variable as here. + LoDTensor batch_proj, batch_cell; + auto* batch_cell_pre_act = ctx.Output("BatchCellPreAct"); + batch_cell_pre_act->mutable_data(dims, ctx.GetPlace()); + auto* batch_hidden = ctx.Output("BatchHidden"); + batch_hidden->mutable_data(dims, ctx.GetPlace()); // T x D + batch_proj.mutable_data(proj_dims, ctx.GetPlace()); // T x P + batch_cell.mutable_data(dims, ctx.GetPlace()); // T x D + + auto batch_starts = batch_gate->lod()[0]; + size_t num_batch = batch_starts.size() - 1; + auto gate_act = math::detail::GetActivationType( + ctx.Attr("gate_activation")); + auto cell_act = math::detail::GetActivationType( + ctx.Attr("cell_activation")); + auto cand_act = math::detail::GetActivationType( + ctx.Attr("candidate_activation")); + auto proj_act = math::detail::GetActivationType( + ctx.Attr("proj_activation")); + auto& place = *ctx.template device_context().eigen_device(); + + for (size_t n = 0; n < num_batch; n++) { + int bstart = static_cast(batch_starts[n]); + int bend = static_cast(batch_starts[n + 1]); + + Tensor gate_t = batch_gate->Slice(bstart, bend); + Tensor hidden_t = batch_hidden->Slice(bstart, bend); + Tensor proj_t = batch_proj.Slice(bstart, bend); + Tensor cell_t = batch_cell.Slice(bstart, bend); + Tensor cell_pre_act_t = batch_cell_pre_act->Slice(bstart, bend); + + int cur_batch_size = bend - bstart; + + if (n > 0) { + int pre_h_start = static_cast(batch_starts[n - 1]); + int pre_h_end = pre_h_start + cur_batch_size; + auto pre_proj_t = batch_proj.Slice(pre_h_start, pre_h_end); + math::matmul(device_ctx, pre_proj_t, false, *weight, + false, static_cast(1.0), &gate_t, + static_cast(1.0)); + } else if (hidden_t0) { + // If n == 0 and there is no initialized hidden state, that is to say + // the H0 is zeros, the calculation W_h * H0 will be skiped. + // If n == 0 and there is initialized hidden state, calculate W_h * H0. + + // Since the batch computing for LSTMP reorders the input sequence + // according to their length. The initialized hidden state also needs + // to reorder. + + Tensor ordered_h0; + ordered_proj0->mutable_data(ctx.GetPlace()); + ReorderInitState(device_ctx, *hidden_t0, order, + &ordered_h0, true); + math::matmul(device_ctx, ordered_h0, false, + *proj_weight, false, static_cast(1.0), + ordered_proj0, static_cast(0.0)); + if (proj_act != math::detail::ActivationType::kIdentity) { + auto proj0_dev = EigenMatrix::From(*ordered_proj0); + ActCompute(cell_act, place, proj0_dev, proj0_dev); + } + math::matmul(device_ctx, *ordered_proj0, false, + *weight, false, static_cast(1.0), + &gate_t, static_cast(1.0)); + } + + lstmp_value.gate_value = gate_t.data(); + lstmp_value.output_value = hidden_t.data(); + lstmp_value.state_value = cell_t.data(); + lstmp_value.state_active_value = cell_pre_act_t.data(); + math::LstmUnitFunctor::compute( + device_ctx, lstmp_value, frame_size, cur_batch_size, gate_act, + cell_act, cand_act); + lstmp_value.prev_state_value = lstmp_value.state_value; + math::matmul(device_ctx, hidden_t, false, *proj_weight, + false, static_cast(1.0), &proj_t, + static_cast(0.0)); + if (proj_act != math::detail::ActivationType::kIdentity) { + auto proj_t_dev = EigenMatrix::From(proj_t); + ActCompute(cell_act, place, proj_t_dev, proj_t_dev); + } + } + + math::Batch2LoDTensorFunctor to_seq; + batch_proj.set_lod(batch_gate->lod()); + // restore the output hidden in LoDTensor from the batch hidden + to_seq(device_ctx, batch_proj, *proj_out); + + batch_cell.set_lod(batch_gate->lod()); + // restore the output cell state in LoDTensor from the batch cell + to_seq(device_ctx, batch_cell, *cell_out); + } +}; + +template +class LSTMPGradKernel : public framework::OpKernel { + public: + template + void ActGradCompute(const math::detail::ActivationType act_type, + const Device& d, X x, Y y, DX dx, DY dy) const { + // x is dummy and won't be used even in Relu(use y instead) + if (act_type == math::detail::ActivationType::kIdentity) + dx.device(d) = dy; + else if (act_type == math::detail::ActivationType::kSigmoid) + SigmoidGradFunctor()(d, x, y, dy, dx); + else if (act_type == math::detail::ActivationType::kTanh) + TanhGradFunctor()(d, x, y, dy, dx); + else if (act_type == math::detail::ActivationType::kReLU) + ReluGradFunctor()(d, x, y, dy, dx); + else + PADDLE_THROW("unsupported activation type"); + } + + void Compute(const framework::ExecutionContext& ctx) const override { + auto* input = ctx.Input("Input"); + auto* weight = ctx.Input("Weight"); + auto* proj_weight = ctx.Input("ProjWeight"); + auto* bias = ctx.Input("Bias"); + + auto* proj_out = ctx.Input("Projection"); + auto* cell_out = ctx.Input("Cell"); + + auto* batch_gate = ctx.Input("BatchGate"); + auto* batch_cell_pre_act = ctx.Input("BatchCellPreAct"); + auto* batch_hidden = ctx.Input("BatchHidden"); + + auto* projection_g = + ctx.Input(framework::GradVarName("Projection")); + + auto* in_g = ctx.Output(framework::GradVarName("Input")); + auto* weight_g = ctx.Output(framework::GradVarName("Weight")); + auto* proj_weight_g = + ctx.Output(framework::GradVarName("ProjWeight")); + auto* bias_g = ctx.Output(framework::GradVarName("Bias")); + + auto* h0 = ctx.Input("H0"); + auto* ordered_proj0 = ctx.Input("OrderedP0"); + auto* c0 = ctx.Input("C0"); + + auto* h0_g = ctx.Output(framework::GradVarName("H0")); + auto* c0_g = ctx.Output(framework::GradVarName("C0")); + + auto& device_ctx = ctx.template device_context(); + math::SetConstant zero; + if (weight_g) { + weight_g->mutable_data(ctx.GetPlace()); + zero(device_ctx, weight_g, static_cast(0.0)); + } + if (proj_weight_g) { + proj_weight_g->mutable_data(ctx.GetPlace()); + zero(device_ctx, proj_weight_g, static_cast(0.0)); + } + + // ordered_h0/c0 is the reordered hidden/cell initialization. + // ordered_h0_g/c0_g is the reordered gradient of hidden/cell + // initialization. + Tensor ordered_h0, ordered_c0, ordered_h0_g, ordered_c0_g; + + framework::Vector order(batch_gate->lod()[2]); + + if (c0) { + ReorderInitState(device_ctx, *c0, order, &ordered_c0, + true); + } + if (c0 && c0_g) { + ordered_c0_g.mutable_data(c0_g->dims(), ctx.GetPlace()); + } + + auto in_dims = input->dims(); + auto out_dims = cell_out->dims(); + framework::DDim proj_dims({in_dims[0], proj_weight->dims()[1]}); + int frame_size = static_cast(in_dims[1] / 4); + PADDLE_ENFORCE_EQ(frame_size, out_dims[1]); + + math::LstmMetaValue lstmp_value; + if (bias && ctx.Attr("use_peepholes")) { + T* bias_data = const_cast(bias->data()); + lstmp_value.check_ig = bias_data + 4 * frame_size; + lstmp_value.check_fg = lstmp_value.check_ig + frame_size; + lstmp_value.check_og = lstmp_value.check_fg + frame_size; + } else { + lstmp_value.check_ig = nullptr; + lstmp_value.check_fg = nullptr; + lstmp_value.check_og = nullptr; + } + + math::LstmMetaGrad lstmp_grad; + + if (bias && bias_g) { + bias_g->mutable_data(ctx.GetPlace()); + zero(device_ctx, bias_g, static_cast(0.0)); + } + if (bias && bias_g && ctx.Attr("use_peepholes")) { + T* bias_g_data = bias_g->data(); + lstmp_grad.check_ig_grad = bias_g_data + 4 * frame_size; + lstmp_grad.check_fg_grad = lstmp_grad.check_ig_grad + frame_size; + lstmp_grad.check_og_grad = lstmp_grad.check_fg_grad + frame_size; + } else { + lstmp_grad.check_ig_grad = nullptr; + lstmp_grad.check_fg_grad = nullptr; + lstmp_grad.check_og_grad = nullptr; + } + + math::LoDTensor2BatchFunctor to_batch; + + auto ToBatch = [&batch_gate, &to_batch]( + const DeviceContext& ctx, const framework::LoDTensor& src, + const framework::DDim& dims, framework::LoDTensor& dst) { + dst.mutable_data(dims, ctx.GetPlace()); + dst.set_lod(batch_gate->lod()); + to_batch(ctx, src, dst, false); + }; + + LoDTensor batch_hidden_g, batch_proj, batch_proj_g, batch_cell; + batch_hidden_g.mutable_data(out_dims, ctx.GetPlace()); + ToBatch(device_ctx, *proj_out, proj_dims, batch_proj); // T x P + ToBatch(device_ctx, *projection_g, proj_dims, batch_proj_g); // T x P + ToBatch(device_ctx, *cell_out, out_dims, batch_cell); // T x D + + LoDTensor batch_cell_g, batch_gate_g; + batch_cell_g.mutable_data(out_dims, ctx.GetPlace()); + // TODO(qingqing) support the case output cell has gradient. + // to_batch(device_ctx, *cell_g, batch_cell_g, false); + zero(device_ctx, &batch_cell_g, static_cast(0.0)); + batch_gate_g.mutable_data(batch_gate->dims(), ctx.GetPlace()); + batch_gate_g.set_lod(batch_gate->lod()); + + auto gate_act = math::detail::GetActivationType( + ctx.Attr("gate_activation")); + auto cell_act = math::detail::GetActivationType( + ctx.Attr("cell_activation")); + auto cand_act = math::detail::GetActivationType( + ctx.Attr("candidate_activation")); + auto proj_act = math::detail::GetActivationType( + ctx.Attr("proj_activation")); + auto& place = *ctx.template device_context().eigen_device(); + + auto batch_starts = batch_gate->lod()[0]; + size_t num_batch = batch_starts.size() - 1; + for (int n = static_cast(num_batch) - 1; n >= 0; n--) { + int bstart = static_cast(batch_starts[n]); + int bend = static_cast(batch_starts[n + 1]); + + Tensor cur_proj = batch_proj.Slice(bstart, bend); + Tensor proj_g = batch_proj_g.Slice(bstart, bend); + if (proj_act != math::detail::ActivationType::kIdentity) { + auto cur_proj_dev = EigenMatrix::From(cur_proj); + auto proj_g_dev = EigenMatrix::From(proj_g); + ActGradCompute(cell_act, place, cur_proj_dev, cur_proj_dev, proj_g_dev, + proj_g_dev); + } + /* hidden state backwarad */ + Tensor out_g = batch_hidden_g.Slice(bstart, bend); + math::matmul(device_ctx, proj_g, false, *proj_weight, + true, static_cast(1.0), &out_g, + static_cast(0.0)); + /* projection weight backward*/ + if (proj_weight_g) { + Tensor hidden_t = batch_hidden->Slice(bstart, bend); + math::matmul(device_ctx, hidden_t, true, proj_g, + false, static_cast(1.0), + proj_weight_g, static_cast(1.0)); + } + + Tensor gate = batch_gate->Slice(bstart, bend); + Tensor cell = batch_cell.Slice(bstart, bend); + Tensor cell_pre_act = batch_cell_pre_act->Slice(bstart, bend); + lstmp_value.gate_value = gate.data(); + lstmp_value.state_value = cell.data(); + lstmp_value.state_active_value = cell_pre_act.data(); + + Tensor gate_g = batch_gate_g.Slice(bstart, bend); + Tensor cell_g = batch_cell_g.Slice(bstart, bend); + lstmp_grad.state_grad = cell_g.data(); + lstmp_grad.gate_grad = gate_g.data(); + lstmp_grad.output_grad = out_g.data(); + + if (n > 0) { + int bstart_pre = static_cast(batch_starts[n - 1]); + Tensor cell_pre = batch_cell.Slice(bstart_pre, bstart); + Tensor cell_pre_g = batch_cell_g.Slice(bstart_pre, bstart); + lstmp_value.prev_state_value = cell_pre.data(); + lstmp_grad.prev_state_grad = cell_pre_g.data(); + } else { + lstmp_value.prev_state_value = c0 ? ordered_c0.data() : nullptr; + lstmp_grad.prev_state_grad = c0_g ? ordered_c0_g.data() : nullptr; + } + + int cur_batch_size = bend - bstart; + math::LstmUnitGradFunctor::compute( + device_ctx, lstmp_value, lstmp_grad, frame_size, cur_batch_size, + gate_act, cell_act, cand_act); + + if (n > 0) { + int pre_h_start = static_cast(batch_starts[n - 1]); + int pre_h_end = pre_h_start + cur_batch_size; + auto pre_proj_g = batch_proj_g.Slice(pre_h_start, pre_h_end); + math::matmul(device_ctx, gate_g, false, *weight, true, + static_cast(1.0), &pre_proj_g, + static_cast(1.0)); + if (weight_g) { + /* weight backward*/ + auto pre_proj = batch_proj.Slice(pre_h_start, pre_h_end); + math::matmul(device_ctx, pre_proj, true, gate_g, + false, static_cast(1.0), weight_g, + static_cast(1.0)); + } + } else { + if (h0 && weight_g) { + ReorderInitState(device_ctx, *h0, order, + &ordered_h0, true); + if (weight_g) { + math::matmul(device_ctx, *ordered_proj0, true, + gate_g, false, static_cast(1.0), + weight_g, static_cast(1.0)); + } + } + if (h0 && (h0_g || proj_weight_g)) { + ordered_h0_g.mutable_data(h0_g->dims(), ctx.GetPlace()); + Tensor proj0_g; + proj0_g.Resize({in_dims[0], proj_weight->dims()[1]}); + proj0_g.mutable_data(ctx.GetPlace()); + math::matmul(device_ctx, gate_g, false, *weight, + true, static_cast(1.0), &proj0_g, + static_cast(0.0)); + if (proj_act != math::detail::ActivationType::kIdentity) { + auto proj0_dev = EigenMatrix::From(*ordered_proj0); + auto proj0_g_dev = EigenMatrix::From(proj0_g); + ActGradCompute(cell_act, place, proj0_dev, proj0_dev, proj0_g_dev, + proj0_g_dev); + } + if (h0_g) { + math::matmul( + device_ctx, proj0_g, false, *proj_weight, true, + static_cast(1.0), &ordered_h0_g, static_cast(0.0)); + } + if (proj_weight_g) { + math::matmul(device_ctx, ordered_h0, true, + proj0_g, false, static_cast(1.0), + proj_weight_g, static_cast(1.0)); + } + } + } + } + + math::Batch2LoDTensorFunctor to_seq; + if (in_g) { + /* backward data */ + in_g->mutable_data(ctx.GetPlace()); + to_seq(device_ctx, batch_gate_g, *in_g); + } + if (bias && bias_g) { + /* backward bias */ + Tensor b_g = *bias_g; + b_g.Resize({bias_g->numel(), 1}); + Tensor gate_bias_g = b_g.Slice(0, 4 * frame_size); + math::ColwiseSum col_sum; + col_sum(device_ctx, batch_gate_g, &gate_bias_g); + } + + if (h0 && h0_g) { + ReorderInitState(device_ctx, ordered_h0_g, order, h0_g, + false); + } + if (c0 && c0_g) { + ReorderInitState(device_ctx, ordered_c0_g, order, c0_g, + false); + } + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/math/CMakeLists.txt b/paddle/operators/math/CMakeLists.txt index c607704efac86982c8c22e462381aaab488a9b69..28c5aec1996ad04a6cb551ac68c14b613d16858e 100644 --- a/paddle/operators/math/CMakeLists.txt +++ b/paddle/operators/math/CMakeLists.txt @@ -11,7 +11,7 @@ if(WITH_GPU) nv_library(sequence_pooling SRCS sequence_pooling.cc sequence_pooling.cu DEPS device_context math_function) nv_library(vol2col SRCS vol2col.cc vol2col.cu DEPS device_context tensor) nv_library(context_project SRCS context_project.cc context_project.cu DEPS device_context math_function) - nv_library(sequence2batch SRCS sequence2batch.cc sequence2batch.cu DEPS device_context tensor) + nv_library(sequence2batch SRCS sequence2batch.cc sequence2batch.cu DEPS device_context tensor math_function) nv_library(sequence_padding SRCS sequence_padding.cc sequence_padding.cu DEPS lod_tensor device_context) nv_library(sequence_scale SRCS sequence_scale.cc sequence_scale.cu DEPS lod_tensor device_context) nv_library(lstm_compute SRCS lstm_compute.cc lstm_compute.cu DEPS device_context activation_functions) @@ -28,7 +28,7 @@ else() cc_library(sequence_pooling SRCS sequence_pooling.cc DEPS device_context math_function) cc_library(vol2col SRCS vol2col.cc DEPS device_context tensor) cc_library(context_project SRCS context_project.cc DEPS device_context math_function) - cc_library(sequence2batch SRCS sequence2batch.cc DEPS device_context tensor) + cc_library(sequence2batch SRCS sequence2batch.cc DEPS device_context tensor math_function) cc_library(sequence_padding SRCS sequence_padding.cc DEPS lod_tensor device_context) cc_library(sequence_scale SRCS sequence_scale.cc DEPS lod_tensor device_context) cc_library(lstm_compute SRCS lstm_compute.cc DEPS device_context activation_functions) diff --git a/paddle/operators/math/matmul.h b/paddle/operators/math/matmul.h index 7048e11e6f27a075892c28681a3c4913a27b3f3e..ae7f1fe9be5066a0ac0ac522849d481fc66a19be 100644 --- a/paddle/operators/math/matmul.h +++ b/paddle/operators/math/matmul.h @@ -41,10 +41,24 @@ class MatMulFunctor { "Input tensor a must be at least 1-dimensional."); PADDLE_ENFORCE_GE(dim_b.size(), 1, "Input tensor b must be at least 1-dimensional."); - PADDLE_ENFORCE_LE(dim_a.size(), 3, - "Input tensor a must be at most 3-dimensional."); - PADDLE_ENFORCE_LE(dim_b.size(), 3, - "Input tensor b must be at most 3-dimensional."); + + std::vector out_dim; + int64_t batch_count = 1; + if (dim_a.size() > 3) { + PADDLE_ENFORCE(dim_b.size() == dim_a.size(), + "The dimensions of X and Y must be the same, and both of " + "them should be %d-dimensional.", + dim_b.size()); + // The first rank-2 dimensions are accumulated on the batch_count, and the + // last two dimensions are used for matrix multiplication. + for (int j = 0; j < dim_a.size() - 2; ++j) { + PADDLE_ENFORCE_EQ(dim_b[j], dim_a[j], + "The %d-th dimension of X and Y must be the same.", + j); + out_dim.push_back(dim_a[j]); + batch_count *= dim_a[j]; + } + } int M = 0, N = 0, kA = 0, kB = 0, batchCountA = 0, batchCountB = 0, strideA = 0, strideB = 0; @@ -67,7 +81,11 @@ class MatMulFunctor { strideA = M * kA; break; default: - assert(false); + batchCountA = batch_count; + size_t mat_s = dim_a.size() - 2; + M = trans_a ? dim_a[mat_s + 1] : dim_a[mat_s]; + kA = trans_a ? dim_a[mat_s] : dim_a[mat_s + 1]; + strideA = M * kA; } switch (dim_b.size()) { @@ -88,7 +106,11 @@ class MatMulFunctor { strideB = kB * N; break; default: - assert(false); + batchCountB = batch_count; + size_t mat_s = dim_b.size() - 2; + kB = trans_b ? dim_b[mat_s + 1] : dim_b[mat_s]; + N = trans_b ? dim_b[mat_s] : dim_b[mat_s + 1]; + strideB = kB * N; } PADDLE_ENFORCE_EQ( diff --git a/paddle/operators/math/sampler.cc b/paddle/operators/math/sampler.cc new file mode 100644 index 0000000000000000000000000000000000000000..4f1cbfe31ac68499a51eda600b38b879f7ca055f --- /dev/null +++ b/paddle/operators/math/sampler.cc @@ -0,0 +1,70 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "sampler.h" + +namespace paddle { +namespace random { + +Sampler::~Sampler() {} + +UniformSampler::UniformSampler(int64 range) + : Sampler(range), inv_range_(1.0 / range) { + random_engine_ = std::make_shared(seed_); + dist_ = std::make_shared>(0, range); +} + +UniformSampler::UniformSampler(int64 range, unsigned int seed) + : Sampler(range, seed), inv_range_(1.0 / range) { + random_engine_ = std::make_shared(seed_); + dist_ = std::make_shared>(0, range); +} + +int64 UniformSampler::Sample() const { return (*dist_)(*random_engine_); } + +float UniformSampler::Probability(int64 value) const { return inv_range_; } + +LogUniformSampler::LogUniformSampler(int64 range) + : Sampler(range), log_range_(log(range + 1)) { + random_engine_ = std::make_shared(seed_); + dist_ = std::make_shared>(0, 1); +} + +LogUniformSampler::LogUniformSampler(int64 range, unsigned int seed) + : Sampler(range, seed), log_range_(log(range + 1)) { + random_engine_ = std::make_shared(seed_); + dist_ = std::make_shared>(0, 1); +} +int64 LogUniformSampler::Sample() const { + // Got Log Uniform distribution from uniform distribution by + // inverse_transform_sampling method + // More details: + // https://wanghaoshuang.github.io/2017/11/Log-uniform-distribution-sampler/ + const int64 value = + static_cast(exp((*dist_)(*random_engine_) * log_range_)) - 1; + // Mathematically, value should be <= range_, but might not be due to some + // floating point roundoff, so we mod by range_. + return value % range_; +} + +float LogUniformSampler::Probability(int64 value) const { + // Given f(x) = 1/[(x+1) * log_range_] + // The value's probability is integral of f(x) from value to (value + 1) + // More details: + // https://wanghaoshuang.github.io/2017/11/Log-uniform-distribution-sampler + return (log((value + 2.0) / (value + 1.0))) / log_range_; +} + +} // namespace random +} // namespace paddle diff --git a/paddle/operators/math/sampler.h b/paddle/operators/math/sampler.h new file mode 100644 index 0000000000000000000000000000000000000000..8f82089e7bd9e0ae6282459b650c225d6765faee --- /dev/null +++ b/paddle/operators/math/sampler.h @@ -0,0 +1,100 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include +#include +typedef long int64; +namespace paddle { +namespace operators { +namespace math { + +// TODO(wanghaoshuang): Support for GPU + +/** +* Sample integers from [0, range). +*/ +class Sampler { + public: + explicit Sampler(int64 range) : range_(range) { + PADDLE_ENFORCE_GT(range, 0); + std::random_device r; + seed_ = r(); + } + explicit Sampler(int64 range, unsigned int seed) + : range_(range), seed_(seed) { + PADDLE_ENFORCE_GT(range, 0); + } + virtual ~Sampler(); + // Sample a single value + virtual int64 Sample() const = 0; + // The probability that a single call to Sample() returns the given value. + virtual float Probability(int64 value) const = 0; + + int64 range() { return range_; }; + + protected: + const int64 range_; + unsigned int seed_; +}; + +/** + * Sample integers from [0, range). + * And the distribution function is: + * P(x) = 1 / range + */ +class UniformSampler : public Sampler { + public: + explicit UniformSampler(int64 range); + + explicit UniformSampler(int64 range, unsigned int seed); + + ~UniformSampler() override {} + + int64 Sample() const override; + + float Probability(int64 value) const override; + + private: + const float inv_range_; + std::shared_ptr random_engine_; + std::shared_ptr> dist_; +}; + +/** + * Sample integers from [0, range). + * And the distribution function is: + * P(x) = (1/ln(range+1)) * ln(1 + 1/(x + 1)) + */ +class LogUniformSampler : public Sampler { + public: + explicit LogUniformSampler(int64 range); + + explicit LogUniformSampler(int64 range, unsigned int seed); + + ~LogUniformSampler() override {} + + int64 Sample() const override; + + float Probability(int64 value) const override; + + private: + const float log_range_; + std::shared_ptr random_engine_; + std::shared_ptr> dist_; +}; + +} // math +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/math/selected_rows_functor.cu b/paddle/operators/math/selected_rows_functor.cu index 0ee456f9bc61436bd0f2f8ef20dd1654e7e56d56..acdd87cb3550bc5f3891aed6fefd4301a3395f9f 100644 --- a/paddle/operators/math/selected_rows_functor.cu +++ b/paddle/operators/math/selected_rows_functor.cu @@ -31,7 +31,7 @@ struct SelectedRowsAdd { PADDLE_ENFORCE_EQ(in1_height, input2.height()); output->set_height(in1_height); - auto& in1_rows = input1.rows(); + framework::Vector in1_rows(input1.rows()); auto& in2_rows = input2.rows(); std::vector out_rows; out_rows.reserve(in1_rows.size() + in2_rows.size()); @@ -108,7 +108,7 @@ struct SelectedRowsAddTensor { PADDLE_ENFORCE_EQ(in1_height, out_dims[0]); auto& in1_value = input1.value(); - auto& in1_rows = input1.rows(); + framework::Vector in1_rows(input1.rows()); int64_t in1_row_numel = in1_value.numel() / in1_rows.size(); PADDLE_ENFORCE_EQ(in1_row_numel, input2.numel() / in1_height); @@ -126,7 +126,7 @@ struct SelectedRowsAddTensor { dim3 grid(1, in1_rows.size()); SelectedRowsAddTensorKernel< T, block_size><<>>( - in1_data, in1_rows.data(), out_data, in1_row_numel); + in1_data, in1_rows.cuda_data(), out_data, in1_row_numel); auto out_eigen = framework::EigenVector::Flatten(*output); auto in2_eigen = framework::EigenVector::Flatten(input2); @@ -146,7 +146,7 @@ struct SelectedRowsAddTo { auto in1_height = input1.height(); PADDLE_ENFORCE_EQ(in1_height, input2->height()); - auto& in1_rows = input1.rows(); + framework::Vector in1_rows(input1.rows()); auto& in2_rows = *(input2->mutable_rows()); auto& in1_value = input1.value(); @@ -204,7 +204,7 @@ struct SelectedRowsAddToTensor { PADDLE_ENFORCE_EQ(in1_height, in2_dims[0]); auto& in1_value = input1.value(); - auto& in1_rows = input1.rows(); + framework::Vector in1_rows(input1.rows()); int64_t in1_row_numel = in1_value.numel() / in1_rows.size(); PADDLE_ENFORCE_EQ(in1_row_numel, input2->numel() / in1_height); @@ -216,7 +216,7 @@ struct SelectedRowsAddToTensor { dim3 grid(1, in1_rows.size()); SelectedRowsAddToTensorKernel< T, block_size><<>>( - in1_data, in1_rows.data(), in2_data, in1_row_numel); + in1_data, in1_rows.cuda_data(), in2_data, in1_row_numel); } }; @@ -257,7 +257,7 @@ struct MergeAdd { framework::SelectedRows operator()(const platform::CUDADeviceContext& context, const framework::SelectedRows& input) { framework::SelectedRows out; - auto input_rows = input.rows(); + framework::Vector input_rows(input.rows()); std::set row_set(input_rows.begin(), input_rows.end()); std::vector merge_rows(row_set.begin(), row_set.end()); @@ -283,9 +283,9 @@ struct MergeAdd { MergeAddKernel< T, 256><<(context) - .stream()>>>(input_data, input.rows().data(), out_data, - out.rows().data(), out.rows().size(), - input_width); + .stream()>>>(input_data, input_rows.cuda_data(), out_data, + out.mutable_rows()->cuda_data(), + out.rows().size(), input_width); return out; } }; @@ -370,8 +370,8 @@ struct UpdateToTensor { dim3 threads(platform::PADDLE_CUDA_NUM_THREADS, 1); dim3 grid(1, in1_rows.size()); UpdateToTensorKernel<<< - grid, threads, 0, context.stream()>>>(in1_data, in1_rows.data(), op, - in2_data, in1_row_numel); + grid, threads, 0, context.stream()>>>(in1_data, in1_rows.cuda_data(), + op, in2_data, in1_row_numel); } }; } // namespace scatter diff --git a/paddle/operators/math/sequence2batch.cc b/paddle/operators/math/sequence2batch.cc index e459a42ca251a9fc79f745f48a118ce898a0f77e..17abce1c2f809f75edb2c5dc46709094c2ce10c3 100644 --- a/paddle/operators/math/sequence2batch.cc +++ b/paddle/operators/math/sequence2batch.cc @@ -23,8 +23,10 @@ template class CopyMatrixRowsFunctor { public: void operator()(const platform::CPUDeviceContext& context, - const framework::Tensor& src, const size_t* index, - framework::Tensor& dst, bool is_src_index) { + const framework::Tensor& src, + framework::Vector index_lod, framework::Tensor& dst, + bool is_src_index) { + size_t* index = index_lod.data(); auto src_dims = src.dims(); auto dst_dims = dst.dims(); PADDLE_ENFORCE_EQ(src_dims.size(), 2UL, diff --git a/paddle/operators/math/sequence2batch.cu b/paddle/operators/math/sequence2batch.cu index 452ae8951000872b706f7e4227a62dbf98109e7e..f27631271a42b4d64abef00d7f119b85e32edda4 100644 --- a/paddle/operators/math/sequence2batch.cu +++ b/paddle/operators/math/sequence2batch.cu @@ -42,8 +42,10 @@ template class CopyMatrixRowsFunctor { public: void operator()(const platform::CUDADeviceContext& context, - const framework::Tensor& src, const size_t* index, - framework::Tensor& dst, bool is_src_index) { + const framework::Tensor& src, + framework::Vector index_lod, framework::Tensor& dst, + bool is_src_index) { + size_t* index = index_lod.cuda_data(); auto src_dims = src.dims(); auto dst_dims = dst.dims(); PADDLE_ENFORCE_EQ(src_dims.size(), 2, diff --git a/paddle/operators/math/sequence2batch.h b/paddle/operators/math/sequence2batch.h index a5c43a2c7d4d729c35a20a27de2a23141e6019bc..6db0427b4174a09dd254d771e8d3d215cc6571a9 100644 --- a/paddle/operators/math/sequence2batch.h +++ b/paddle/operators/math/sequence2batch.h @@ -35,7 +35,7 @@ class CopyMatrixRowsFunctor { // copy the input src to the indexed rows of output dst. // The indexed rows are based on the input index. void operator()(const DeviceContext& context, const framework::Tensor& src, - const size_t* index, framework::Tensor& dst, + framework::Vector index_lod, framework::Tensor& dst, bool is_src_index); }; @@ -66,7 +66,7 @@ class LoDTensor2BatchFunctor { PADDLE_ENFORCE_EQ(lods[1].size(), static_cast(lod_tensor.dims()[0])); CopyMatrixRowsFunctor to_batch; - to_batch(context, lod_tensor, lods[1].data(), batch, true); + to_batch(context, lod_tensor, lods[1], batch, true); return; } @@ -144,7 +144,7 @@ class LoDTensor2BatchFunctor { batch.set_lod(batch_lods); CopyMatrixRowsFunctor to_batch; - to_batch(context, lod_tensor, seq2batch_idx, batch, true); + to_batch(context, lod_tensor, batch_lods[1], batch, true); } }; @@ -159,8 +159,7 @@ class Batch2LoDTensorFunctor { PADDLE_ENFORCE_EQ(in_lod[1].size(), static_cast(lod_tensor.dims()[0])); CopyMatrixRowsFunctor to_seq; - size_t* index = in_lod[1].data(); - to_seq(context, batch, index, lod_tensor, false); + to_seq(context, batch, in_lod[1], lod_tensor, false); } }; diff --git a/paddle/operators/math/sequence_padding.cc b/paddle/operators/math/sequence_padding.cc index fd66455eaef60209b9ca334480951a9f7687729b..2e69aa47eb8a060a6cac6588b2f37960898aba92 100644 --- a/paddle/operators/math/sequence_padding.cc +++ b/paddle/operators/math/sequence_padding.cc @@ -32,7 +32,8 @@ class PaddingLoDTensorFunctor { framework::LoD abs_offset_lod = framework::ToAbsOffset(lod); auto seq_dims = seq.dims(); - PADDLE_ENFORCE_EQ(seq_dims[0], abs_offset_lod[level].back(), + PADDLE_ENFORCE_EQ(seq_dims[0], + static_cast(abs_offset_lod[level].back()), "The first dimension of LoDTensor seq should be " "equal to the sum of all sequences's length."); @@ -41,32 +42,32 @@ class PaddingLoDTensorFunctor { "The input padding should be a 3-D Tensor of shape " "[max_sequence_length, num_sequences, sequence_width]."); - const size_t max_sequence_length = MaximumSequenceLength(lod, level); + const int64_t max_sequence_length = MaximumSequenceLength(lod, level); PADDLE_ENFORCE_EQ(padding_dims[0], max_sequence_length, "The first dimension of Tensor padding should be the " "maximum length of all sequences in LoDTensor seq."); - const size_t num_sequences = abs_offset_lod[level].size() - 1; + const int64_t num_sequences = abs_offset_lod[level].size() - 1; PADDLE_ENFORCE_EQ(padding_dims[1], num_sequences, "The second dimension of Tensor padding should be the " "number of sequences in LoDTensor seq."); - const size_t sequence_width = seq.numel() / seq_dims[0]; + const int64_t sequence_width = seq.numel() / seq_dims[0]; PADDLE_ENFORCE_EQ(padding_dims[2], sequence_width, "The third dimension of Tensor padding should be the " "width of sequence in LoDTensor seq."); const T* seq_data = seq.data(); T* padding_data = padding.data(); - for (size_t i = 0; i < max_sequence_length; ++i) { - for (size_t j = 0; j < num_sequences; ++j) { - size_t start_pos = abs_offset_lod[level][j]; - size_t sequence_length = abs_offset_lod[level][j + 1] - start_pos; + for (int64_t i = 0; i < max_sequence_length; ++i) { + for (int64_t j = 0; j < num_sequences; ++j) { + int64_t start_pos = abs_offset_lod[level][j]; + int64_t sequence_length = abs_offset_lod[level][j + 1] - start_pos; if (i < sequence_length) { // i > 0 => sequence_length > 0 T scale = norm_by_times ? (1.0f / static_cast(sequence_length)) : 1.0f; - for (size_t k = 0; k < sequence_width; ++k) { + for (int64_t k = 0; k < sequence_width; ++k) { padding_data[(i * num_sequences + j) * sequence_width + k] = seq_data[(start_pos + i) * sequence_width + k] * scale; } @@ -93,7 +94,8 @@ class UnpaddingLoDTensorFunctor { framework::LoD abs_offset_lod = framework::ToAbsOffset(lod); auto seq_dims = seq.dims(); - PADDLE_ENFORCE_EQ(seq_dims[0], abs_offset_lod[level].back(), + PADDLE_ENFORCE_EQ(seq_dims[0], + static_cast(abs_offset_lod[level].back()), "The first dimension of LoDTensor seq should be " "equal to the sum of all sequences's length."); @@ -102,31 +104,31 @@ class UnpaddingLoDTensorFunctor { "The input padding should be a 3-D Tensor of shape " "[max_sequnece_length, num_sequences, sequence_width]."); - const size_t max_sequence_length = MaximumSequenceLength(lod, level); + const int64_t max_sequence_length = MaximumSequenceLength(lod, level); PADDLE_ENFORCE_EQ(padding_dims[0], max_sequence_length, "The first dimension of Tensor padding should be " "the maximum length of all sequences in LoDTensor seq."); - const size_t num_sequences = abs_offset_lod[level].size() - 1; + const int64_t num_sequences = abs_offset_lod[level].size() - 1; PADDLE_ENFORCE_EQ(padding_dims[1], num_sequences, "The second dimension of Tensor padding should be " "the number of sequences in LoDTensor seq."); - const size_t sequence_width = seq.numel() / seq_dims[0]; + const int64_t sequence_width = seq.numel() / seq_dims[0]; PADDLE_ENFORCE_EQ(padding_dims[2], sequence_width, "The third dimension of Tensor padding should be the " "width of sequence in LoDTensor seq."); const T* padding_data = padding.data(); T* seq_data = seq.data(); - for (size_t i = 0; i < num_sequences; ++i) { - size_t start_pos = abs_offset_lod[level][i]; - size_t sequence_length = abs_offset_lod[level][i + 1] - start_pos; - for (size_t j = 0; j < sequence_length; ++j) { + for (int64_t i = 0; i < num_sequences; ++i) { + int64_t start_pos = abs_offset_lod[level][i]; + int64_t sequence_length = abs_offset_lod[level][i + 1] - start_pos; + for (int64_t j = 0; j < sequence_length; ++j) { // sequence_width > j > 0 T scale = norm_by_times ? (1.0f / static_cast(sequence_length)) : 1.0f; - for (size_t k = 0; k < sequence_width; ++k) { + for (int64_t k = 0; k < sequence_width; ++k) { seq_data[(start_pos + j) * sequence_width + k] = padding_data[(j * num_sequences + i) * sequence_width + k] * scale; diff --git a/paddle/operators/math/sequence_padding.cu b/paddle/operators/math/sequence_padding.cu index e4be178f81581dea2e84cf488b01d5f7f4cc0030..65c9cfe4a0ec14d220ad237baa71703a783ed0fa 100644 --- a/paddle/operators/math/sequence_padding.cu +++ b/paddle/operators/math/sequence_padding.cu @@ -71,7 +71,8 @@ class PaddingLoDTensorFunctor { framework::LoD abs_offset_lod = framework::ToAbsOffset(lod); auto seq_dims = seq.dims(); - PADDLE_ENFORCE_EQ(seq_dims[0], abs_offset_lod[level].back(), + PADDLE_ENFORCE_EQ(seq_dims[0], + static_cast(abs_offset_lod[level].back()), "The first dimension of LoDTensor seq should be " "equal to the sum of all sequences's length."); @@ -80,17 +81,17 @@ class PaddingLoDTensorFunctor { "The input padding should be a 3-D Tensor of shape " "[max_sequence_length, num_sequences, sequence_width]."); - size_t max_sequence_length = MaximumSequenceLength(lod, level); + int64_t max_sequence_length = MaximumSequenceLength(lod, level); PADDLE_ENFORCE_EQ(padding_dims[0], max_sequence_length, "The first dimension of Tensor padding should be the " "maximum length of all sequences in LoDTensor seq."); - const size_t num_sequences = abs_offset_lod[level].size() - 1; + const int64_t num_sequences = abs_offset_lod[level].size() - 1; PADDLE_ENFORCE_EQ(padding_dims[1], num_sequences, "The second dimension of Tensor padding should be the " "number of sequences in LoDTensor seq."); - const size_t sequence_width = seq.numel() / seq_dims[0]; + const int64_t sequence_width = seq.numel() / seq_dims[0]; PADDLE_ENFORCE_EQ(padding_dims[2], sequence_width, "The third dimension of Tensor padding should be the " "width of sequence in LoDTensor seq."); @@ -101,7 +102,7 @@ class PaddingLoDTensorFunctor { return; } - const size_t kBlockSize = 512; + const int64_t kBlockSize = 512; /* At least use 32 threads to copy sequence_width elements, * and at least 8 elements for each thread. @@ -119,12 +120,14 @@ class PaddingLoDTensorFunctor { T* padding_data = padding.data(); if (norm_by_times) { SequencePaddingKernel<<>>( - padding_data, const_cast(seq_data), abs_offset_lod[level].data(), - sequence_width, max_sequence_length, num_sequences); + padding_data, const_cast(seq_data), + abs_offset_lod[level].cuda_data(), sequence_width, + max_sequence_length, num_sequences); } else { SequencePaddingKernel<<>>( - padding_data, const_cast(seq_data), abs_offset_lod[level].data(), - sequence_width, max_sequence_length, num_sequences); + padding_data, const_cast(seq_data), + abs_offset_lod[level].cuda_data(), sequence_width, + max_sequence_length, num_sequences); } } }; @@ -143,7 +146,8 @@ class UnpaddingLoDTensorFunctor { framework::LoD abs_offset_lod = framework::ToAbsOffset(lod); auto seq_dims = seq.dims(); - PADDLE_ENFORCE_EQ(seq_dims[0], abs_offset_lod[level].back(), + PADDLE_ENFORCE_EQ(seq_dims[0], + static_cast(abs_offset_lod[level].back()), "The first dimension of LoDTensor seq should be " "equal to the sum of all sequences's length."); @@ -152,17 +156,17 @@ class UnpaddingLoDTensorFunctor { "The input padding should be a 3-D Tensor of shape " "[max_sequnece_length, num_sequences, sequence_width]."); - size_t max_sequence_length = MaximumSequenceLength(lod, level); + int64_t max_sequence_length = MaximumSequenceLength(lod, level); PADDLE_ENFORCE_EQ(padding_dims[0], max_sequence_length, "The first dimension of Tensor padding should be " "the maximum length of all sequences in LoDTensor seq."); - const size_t num_sequences = abs_offset_lod[level].size() - 1; + const int64_t num_sequences = abs_offset_lod[level].size() - 1; PADDLE_ENFORCE_EQ(padding_dims[1], num_sequences, "The second dimension of Tensor padding should be " "the number of sequences in LoDTensor seq."); - const size_t sequence_width = seq.numel() / seq_dims[0]; + const int64_t sequence_width = seq.numel() / seq_dims[0]; PADDLE_ENFORCE_EQ(padding_dims[2], sequence_width, "The third dimension of Tensor padding should be the " "width of sequence in LoDTensor seq."); @@ -173,7 +177,7 @@ class UnpaddingLoDTensorFunctor { return; } - const size_t kBlockSize = 512; + const int64_t kBlockSize = 512; /* At least use 32 threads to copy sequence_width elements, * and at least 8 elements for each thread. @@ -191,12 +195,14 @@ class UnpaddingLoDTensorFunctor { T* seq_data = seq.data(); if (norm_by_times) { SequencePaddingKernel<<>>( - const_cast(padding_data), seq_data, abs_offset_lod[level].data(), - sequence_width, max_sequence_length, num_sequences); + const_cast(padding_data), seq_data, + abs_offset_lod[level].cuda_data(), sequence_width, + max_sequence_length, num_sequences); } else { SequencePaddingKernel<<>>( - const_cast(padding_data), seq_data, abs_offset_lod[level].data(), - sequence_width, max_sequence_length, num_sequences); + const_cast(padding_data), seq_data, + abs_offset_lod[level].cuda_data(), sequence_width, + max_sequence_length, num_sequences); } } }; diff --git a/paddle/operators/math/sequence_padding_test.cc b/paddle/operators/math/sequence_padding_test.cc index 9799bcd65dc65d5741813374c68a2640eaf4556c..3e504f4a15c2cb4e2380f5ff8a39d83626dae062 100644 --- a/paddle/operators/math/sequence_padding_test.cc +++ b/paddle/operators/math/sequence_padding_test.cc @@ -31,7 +31,7 @@ void TestSequencePadding(const paddle::framework::LoD& lod, cpu_seq.set_lod(lod); cpu_seq.mutable_data(seq_dims, paddle::platform::CPUPlace()); - for (size_t i = 0; i < cpu_seq.numel(); ++i) { + for (int64_t i = 0; i < cpu_seq.numel(); ++i) { cpu_seq.data()[i] = static_cast(i); } @@ -69,7 +69,7 @@ void TestSequencePadding(const paddle::framework::LoD& lod, EXPECT_EQ(cpu_seq.numel(), cpu_seq_back.numel()); EXPECT_EQ(cpu_seq.dims(), cpu_seq_back.dims()); - for (size_t i = 0; i < cpu_seq.numel(); ++i) { + for (int64_t i = 0; i < cpu_seq.numel(); ++i) { EXPECT_EQ(cpu_seq.data()[i], cpu_seq_back.data()[i]); } diff --git a/paddle/operators/math/sequence_pooling.cu b/paddle/operators/math/sequence_pooling.cu index 4c9e6b375ce7251747b9cd443d86cca0858c84ef..f66534a6812a66c737445ea96914a393077d7d65 100644 --- a/paddle/operators/math/sequence_pooling.cu +++ b/paddle/operators/math/sequence_pooling.cu @@ -73,7 +73,7 @@ class MaxSeqPoolFunctor { dim3 grid(num_seq, 1); auto stream = context.stream(); KeMaxSequencePool<<>>( - in_data, starts.data(), out_data, max_index, num_seq, dim); + in_data, starts.cuda_data(), out_data, max_index, num_seq, dim); } }; diff --git a/paddle/operators/math/sequence_scale.cu b/paddle/operators/math/sequence_scale.cu index ceaabd8e0fd81c927fbd4333c0aa7954b8da8513..fd4e28f6113729cd1fa9dc179bd9b601d29b8a7f 100644 --- a/paddle/operators/math/sequence_scale.cu +++ b/paddle/operators/math/sequence_scale.cu @@ -46,7 +46,7 @@ class ScaleLoDTensorFunctor { SequenceScaleKernel<<< num_seq, PADDLE_CUDA_NUM_THREADS, 0, context.stream()>>>( - seq_data, abs_offset_lod[level].data(), scales, seq_width); + seq_data, abs_offset_lod[level].cuda_data(), scales, seq_width); } }; diff --git a/paddle/operators/matmul_op.cc b/paddle/operators/matmul_op.cc index fd65d894d5749c97f860d614de354e89f6d9441d..3336978c8d8d94c69b986970274661a01ad2161d 100644 --- a/paddle/operators/matmul_op.cc +++ b/paddle/operators/matmul_op.cc @@ -41,10 +41,26 @@ class MatMulOp : public framework::OperatorWithKernel { "Input tensor X must be at least 1-dimensional."); PADDLE_ENFORCE_GE(dim_y.size(), 1, "Input tensor Y must be at least 1-dimensional."); - PADDLE_ENFORCE_LE(dim_x.size(), 3, - "Input tensor X must be at most 3-dimensional."); - PADDLE_ENFORCE_LE(dim_y.size(), 3, - "Input tensor Y must be at most 3-dimensional."); + + std::vector out_dim; + int64_t batch_count = 1; + if (dim_x.size() > 3) { + PADDLE_ENFORCE_EQ( + dim_y.size(), dim_x.size(), + "The dimensions of X and Y must be the same, and both of " + "them should be %d-dimensional.", + dim_x.size()); + + // The first rank-2 dimensions are accumulated on the batch_count, and the + // last two dimensions are used for matrix multiplication. + for (int j = 0; j < dim_x.size() - 2; ++j) { + PADDLE_ENFORCE_EQ(dim_y[j], dim_x[j], + "The %d-th dimension of X and Y must be the same.", + j); + out_dim.push_back(dim_x[j]); + batch_count *= dim_x[j]; + } + } int M = 0, N = 0, KX = 0, KY = 0, batchCountX = 0, batchCountY = 0; bool remove_initial_dim = false, remove_final_dim = false; @@ -70,7 +86,11 @@ class MatMulOp : public framework::OperatorWithKernel { KX = transpose_x ? dim_x[1] : dim_x[2]; break; default: - assert(false); + batchCountX = batch_count; + size_t mat_s = dim_x.size() - 2; + M = transpose_x ? dim_x[mat_s + 1] : dim_x[mat_s]; + KX = transpose_x ? dim_x[mat_s] : dim_x[mat_s + 1]; + break; } switch (dim_y.size()) { @@ -94,7 +114,10 @@ class MatMulOp : public framework::OperatorWithKernel { N = transpose_y ? dim_y[1] : dim_y[2]; break; default: - assert(false); + batchCountY = batch_count; + size_t mat_s = dim_y.size() - 2; + KY = transpose_y ? dim_y[mat_s + 1] : dim_y[mat_s]; + N = transpose_y ? dim_y[mat_s] : dim_y[mat_s + 1]; } PADDLE_ENFORCE_EQ( @@ -110,7 +133,11 @@ class MatMulOp : public framework::OperatorWithKernel { std::vector dim_out; if (batchCount) { - dim_out.push_back(batchCount); + if (dim_x.size() > 3) { + dim_out.insert(dim_out.begin(), out_dim.begin(), out_dim.end()); + } else { + dim_out.push_back(batchCount); + } } if (!remove_initial_dim) { dim_out.push_back(M); @@ -162,10 +189,14 @@ Examples without transpose: - X: [B, M, K], Y: [K] => Out: [B, M] - X: [M, K], Y: [B, K, N] => Out: [B, M, N] - X: [B, M, K], Y: [B, K, N] => Out: [B, M, N] +- X: [B, ..., M, K], Y: [B, ..., K, N] => Out: [B, ..., M, N] The behavior is designed to be similar to the `numpy.matmul` function. The differences are: -- Currently only rank 1 to rank 3 input tensors are supported. +- When the rank of the input data is less than or equal to 3, it + is similar to the `numpy.matmul` function. +- When the rank of the input is greater than 3, the rank of X and + Y must be equal, and the first `rank - 2` dimensions must be equal. - We add `transpose_X` and `transpose_Y` flags. Both the input `X` and `Y` can carry the LoD (Level of Details) information, diff --git a/paddle/operators/matmul_op.h b/paddle/operators/matmul_op.h index 78adc64f76f45afce64c49bcf734647e0db2d6b3..fe6a97465f8992281c909d4600bcfd8121d6a64a 100644 --- a/paddle/operators/matmul_op.h +++ b/paddle/operators/matmul_op.h @@ -137,6 +137,13 @@ class MatMulGradKernel : public framework::OpKernel { y_dims.push_back(1); } + int batch_count = 0; + // The first rank-2 dimensions are accumulated on the batch_count, and the + // last two dimensions are used for matrix multiplication. + if (x_dims.size() > 3) { + batch_count = accumulate(x_dims.begin(), x_dims.end() - 2, 1, + std::multiplies()); + } // Fix the dOut dimensions. int M = 0, N = 0, batchCountX = 0, batchCountY = 0; @@ -149,7 +156,9 @@ class MatMulGradKernel : public framework::OpKernel { M = transpose_x ? x_dims[2] : x_dims[1]; break; default: - assert(false); + batchCountX = batch_count; + size_t mat_s = x_dims.size() - 2; + M = transpose_x ? x_dims[mat_s + 1] : x_dims[mat_s]; } switch (y_dims.size()) { @@ -161,7 +170,9 @@ class MatMulGradKernel : public framework::OpKernel { N = transpose_y ? y_dims[1] : y_dims[2]; break; default: - assert(false); + batchCountY = batch_count; + size_t mat_s = y_dims.size() - 2; + N = transpose_y ? y_dims[mat_s] : y_dims[mat_s + 1]; } if (batchCountX && batchCountY) { PADDLE_ENFORCE_EQ( @@ -172,7 +183,11 @@ class MatMulGradKernel : public framework::OpKernel { int batchCount = std::max(batchCountX, batchCountY); std::vector dout_dims = {M, N}; if (batchCount) { - dout_dims.insert(dout_dims.begin(), batchCount); + if (x_dims.size() > 3) { + dout_dims.insert(dout_dims.begin(), x_dims.begin(), x_dims.end() - 2); + } else { + dout_dims.insert(dout_dims.begin(), batchCount); + } } Tensor X = Reshape(x, make_ddim(x_dims)); Tensor Y = Reshape(y, make_ddim(y_dims)); diff --git a/paddle/operators/multiplex_op.cc b/paddle/operators/multiplex_op.cc index 78263da2fbf843f6a5af2ba95aa0b219a7523b52..d275fa5cbbfbf4a949d7bb16c3acc598543ba000 100644 --- a/paddle/operators/multiplex_op.cc +++ b/paddle/operators/multiplex_op.cc @@ -119,7 +119,13 @@ REGISTER_OPERATOR(multiplex, ops::MultiplexOp, ops::MultiplexOpMaker, REGISTER_OPERATOR(multiplex_grad, ops::MultiplexGradOp); REGISTER_OP_CPU_KERNEL( multiplex, - ops::MultiplexCPUKernel); + ops::MultiplexCPUKernel, + ops::MultiplexCPUKernel, + ops::MultiplexCPUKernel, + ops::MultiplexCPUKernel); REGISTER_OP_CPU_KERNEL( multiplex_grad, - ops::MultiplexGradCPUKernel); + ops::MultiplexGradCPUKernel, + ops::MultiplexGradCPUKernel, + ops::MultiplexGradCPUKernel, + ops::MultiplexGradCPUKernel); diff --git a/paddle/operators/multiplex_op.cu b/paddle/operators/multiplex_op.cu index 4372dc2c65ec7c0f28e46cd070ea471701ce8304..546e6e7a24d3653e9904706eac51c1b833f51463 100644 --- a/paddle/operators/multiplex_op.cu +++ b/paddle/operators/multiplex_op.cu @@ -90,7 +90,13 @@ namespace ops = paddle::operators; REGISTER_OP_CUDA_KERNEL( multiplex, - ops::MultiplexGPUKernel); + ops::MultiplexGPUKernel, + ops::MultiplexGPUKernel, + ops::MultiplexGPUKernel, + ops::MultiplexGPUKernel); REGISTER_OP_CUDA_KERNEL( multiplex_grad, - ops::MultiplexGradGPUKernel); + ops::MultiplexGradGPUKernel, + ops::MultiplexGradGPUKernel, + ops::MultiplexGradGPUKernel, + ops::MultiplexGradGPUKernel); diff --git a/paddle/operators/nccl_op_test.cu.cc b/paddle/operators/nccl_op_test.cu.cc index 6546096069d4c3fbc4908a16c2dba2ac6d7e6421..072e4eb2eff1f6f3d8745ac8e16709b8e1a69725 100644 --- a/paddle/operators/nccl_op_test.cu.cc +++ b/paddle/operators/nccl_op_test.cu.cc @@ -241,7 +241,7 @@ TEST_F(NCCLTester, ncclReduceOp) { // ncclBcastOp with desc TEST_F(NCCLTester, ncclBcastOp) { std::unique_ptr op2(new f::OpDesc); - const int kRoot = 5; + const int kRoot = 0; op2->SetType("ncclBcast"); op2->SetInput("X", {"st"}); op2->SetInput("Communicator", {"comm"}); diff --git a/paddle/operators/nce_op.cc b/paddle/operators/nce_op.cc index 84ba3ead2b52547b989a4541f31ea31ffcce6c63..994ddf717e7a5b883d8071c6a47da0b4b4074f2e 100644 --- a/paddle/operators/nce_op.cc +++ b/paddle/operators/nce_op.cc @@ -124,7 +124,8 @@ class NCEOpMaker : public framework::OpProtoAndCheckerMaker { "This attribute only be used in unitest. Classes " "in this list wiil be used as negative classes " "for every samples. Under normal conditions, " - "user should avoid setting this attribute."); + "user should avoid setting this attribute.") + .SetDefault({}); AddComment(R"DOC( Compute and return the noise-contrastive estimation training loss. See [Noise-contrastive estimation: A new estimation principle for unnormalized statistical models](http://www.jmlr.org/proceedings/papers/v9/gutmann10a/gutmann10a.pdf). diff --git a/paddle/operators/nce_op.h b/paddle/operators/nce_op.h index e6b496f7896dcb412be8ff096fdccb2f0b682369..86fa13a649ce7fdcaad64e2609ceea2fb4d7e072 100644 --- a/paddle/operators/nce_op.h +++ b/paddle/operators/nce_op.h @@ -197,7 +197,8 @@ class NCEGradKernel : public framework::OpKernel { // get d_x auto d_x = context.Output(framework::GradVarName("Input")); if (d_x != nullptr) { - d_x->mutable_data(context.GetPlace()); + auto* d_x_data = d_x->mutable_data(context.GetPlace()); + std::fill(d_x_data, d_x_data + d_x->numel(), 0.0); auto d_x_matrix = EigenMatrix::From(*d_x); auto w_matrix = EigenMatrix::From(*(context.Input("Weight"))); for (int64_t i = 0; i < sample_labels->numel(); ++i) { diff --git a/paddle/operators/net_op.cc b/paddle/operators/net_op.cc index f12074a5f2d08f14b58619fc86b6aa37e4cad132..000e029840ceb842941f9cbad5758209b6fd4dd5 100644 --- a/paddle/operators/net_op.cc +++ b/paddle/operators/net_op.cc @@ -11,21 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "paddle/operators/net_op.h" #include diff --git a/paddle/operators/one_hot_op.cc b/paddle/operators/one_hot_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..e78b7468de4ea5f29378c2dc5905fdd36fb0ae2f --- /dev/null +++ b/paddle/operators/one_hot_op.cc @@ -0,0 +1,95 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "paddle/operators/one_hot_op.h" +#include "paddle/framework/framework.pb.h" + +namespace paddle { +namespace operators { + +class OneHotOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of OneHotOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Out"), + "Output(Out) of OneHotOp should not be null."); + + auto x_dims = ctx->GetInputDim("X"); + PADDLE_ENFORCE_GE(x_dims.size(), 2, + "Rank of Input(X) should be at least 2."); + PADDLE_ENFORCE_GE(x_dims[x_dims.size() - 1], 1U, + "Last dimension of Input(X) should be 1."); + + int depth = ctx->Attrs().Get("depth"); + + PADDLE_ENFORCE_GT(depth, 0, "Should provide a positive depth (%d).", depth); + + framework::DDim out_dims(x_dims); + out_dims[out_dims.size() - 1] = depth; + ctx->SetOutputDim("Out", out_dims); + ctx->ShareLoD("X", /* --> */ "Out"); + } +}; + +class OneHotOpMaker : public framework::OpProtoAndCheckerMaker { + public: + OneHotOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", + "(LoDTensor, LoDTensor) Input variable with rank at least 2. " + "The last dimension of X should be 1. Each value of X is an index " + "to indicate the position."); + AddOutput("Out", + "(Tensor, Tensor) Output tensor with same rank as X. " + "The tensor consists of one-hot representations of values in X."); + AddAttr("depth", + "A positive integer to specify the length of one-hot vector."); + AddAttr("dtype", + "An integer to specify the data type of one-hot " + "vector. The default value is FP32.") + .SetDefault(paddle::framework::proto::DataType::FP32); + AddComment(R"DOC( +One Hot Operator. This operator creates the one-hot representations for input +index values. The following example will help to explain the function of this +operator: + +X is a LoDTensor: + X.lod = [[0, 1, 4]] + X.shape = [4, 1] + X.data = [[1], [1], [3], [0]] + +set depth = 4 + +Out is a LoDTensor: + Out.lod = [[0, 1, 4]] + Out.shape = [4, 4] + Out.data = [[0., 1., 0., 0.], + [0., 1., 0., 0.], + [0., 0., 0., 1.], + [1., 0., 0., 0.]] +)DOC"); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OPERATOR(one_hot, ops::OneHotOp, ops::OneHotOpMaker, + paddle::framework::EmptyGradOpMaker); +REGISTER_OP_CPU_KERNEL( + one_hot, ops::OneHotKernel, + ops::OneHotKernel); diff --git a/paddle/operators/one_hot_op.cu b/paddle/operators/one_hot_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..16f6d9433eabd7be157ed57362a0d55d86c6ee92 --- /dev/null +++ b/paddle/operators/one_hot_op.cu @@ -0,0 +1,80 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "paddle/operators/one_hot_op.h" +#include "paddle/platform/cuda_helper.h" +#include "paddle/platform/gpu_info.h" + +namespace paddle { +namespace operators { +using platform::PADDLE_CUDA_NUM_THREADS; + +template +__global__ void FillOutputKernel(const InT* p_in_data, OutT* p_out_data, + const int64_t numel, const int depth) { + int idx = blockIdx.x * blockDim.x + threadIdx.x; + if (idx < numel) { + *(p_out_data + (idx * depth) + p_in_data[idx]) = 1.0; + } +} + +template +struct OneHotOpCUDAFunctor { + const framework::LoDTensor* in_; + framework::LoDTensor* out_; + const DeviceContext& ctx_; + int depth_; + + OneHotOpCUDAFunctor(const framework::LoDTensor* in, framework::LoDTensor* out, + int depth, const DeviceContext& ctx) + : in_(in), out_(out), depth_(depth), ctx_(ctx) {} + + template + void operator()() const { + auto* p_in_data = in_->data(); + auto numel = in_->numel(); + auto* p_out_data = out_->mutable_data(ctx_.GetPlace()); + auto stream = ctx_.stream(); + math::set_constant(ctx_, out_, 0.0); + + FillOutputKernel<<<(numel + PADDLE_CUDA_NUM_THREADS - 1) / + PADDLE_CUDA_NUM_THREADS, + PADDLE_CUDA_NUM_THREADS, 0, stream>>>( + p_in_data, p_out_data, numel, depth_); + } +}; + +using LoDTensor = framework::LoDTensor; +template +class OneHotCUDAKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& context) const override { + auto* in = context.Input("X"); + auto* out = context.Output("Out"); + int depth = context.Attr("depth"); + + framework::VisitDataType( + static_cast(context.Attr("dtype")), + OneHotOpCUDAFunctor( + in, out, depth, context.template device_context())); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP_CUDA_KERNEL( + one_hot, ops::OneHotCUDAKernel, + ops::OneHotCUDAKernel); diff --git a/paddle/operators/one_hot_op.h b/paddle/operators/one_hot_op.h new file mode 100644 index 0000000000000000000000000000000000000000..12031ede2c3cd042a3d25003b714652b4d0d4453 --- /dev/null +++ b/paddle/operators/one_hot_op.h @@ -0,0 +1,68 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#pragma once +#include "paddle/framework/op_registry.h" +#include "paddle/operators/math/math_function.h" + +namespace paddle { +namespace operators { + +template +struct OneHotOpFunctor { + const framework::LoDTensor* in_; + framework::LoDTensor* out_; + int depth_; + const DeviceContext& ctx_; + + OneHotOpFunctor(const framework::LoDTensor* in, framework::LoDTensor* out, + int depth, const DeviceContext& ctx) + : in_(in), out_(out), depth_(depth), ctx_(ctx) {} + + template + void operator()() const { + auto* p_in_data = in_->data(); + auto numel = in_->numel(); + auto* p_out_data = out_->mutable_data(ctx_.GetPlace()); + math::set_constant(ctx_, out_, 0.0); + + for (int i = 0; i < numel; ++i) { + PADDLE_ENFORCE_GE(p_in_data[i], 0, + "Illegal index value, should be at least 0."); + PADDLE_ENFORCE_LT(p_in_data[i], depth_, + "Illegal index value, should be less than depth (%d).", + depth_); + *(p_out_data + i * depth_ + p_in_data[i]) = 1.0; + } + } +}; + +using LoDTensor = framework::LoDTensor; +template +class OneHotKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& context) const override { + auto* in = context.Input("X"); + auto* out = context.Output("Out"); + int depth = context.Attr("depth"); + + framework::VisitDataType( + static_cast(context.Attr("dtype")), + OneHotOpFunctor( + in, out, depth, context.template device_context())); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/parallel_do_op.cc b/paddle/operators/parallel_do_op.cc index e1bec0421e76143bef669a4f6fa373cdf01226b2..67f9854c02fa92d0141463088915e720733306fb 100644 --- a/paddle/operators/parallel_do_op.cc +++ b/paddle/operators/parallel_do_op.cc @@ -17,6 +17,7 @@ limitations under the License. */ #include "paddle/framework/executor.h" #include "paddle/framework/op_registry.h" #include "paddle/framework/threadpool.h" +#include "paddle/operators/detail/safe_ref.h" namespace paddle { namespace operators { @@ -30,31 +31,75 @@ static constexpr char kParallelScopes[] = "parallel_scopes"; static constexpr char kParallelBlock[] = "sub_block"; -// using ParallelScopeVar = std::vector; using LoDTensor = framework::LoDTensor; -using OperatorBase = framework::OperatorBase; +using SelectedRows = framework::SelectedRows; -void SplitTensorAndMoveTensorToScopes( - const framework::Scope &scope, - const std::vector &sub_scopes, +static void SplitTensorAndMoveTensorToScopes( + const framework::Scope &scope, std::vector *sub_scopes, const std::vector &places, const std::vector &names) { - PADDLE_ENFORCE_EQ(sub_scopes.size(), places.size()); + size_t num_sub_scopes = 0; for (auto &argu : names) { - auto *var = scope.FindVar(argu); - const auto &tensor = var->Get(); + const auto &tensor = + detail::Ref(scope.FindVar(argu), + "Cannot find variable %s in the parent scope", argu) + .Get(); auto lod_tensors = tensor.SplitLoDTensor(places); for (auto &lod : lod_tensors) { VLOG(3) << lod.dims(); } + if (num_sub_scopes == 0) { + num_sub_scopes = lod_tensors.size(); + } else { + PADDLE_ENFORCE_EQ(num_sub_scopes, lod_tensors.size()); + } + PADDLE_ENFORCE_NE(num_sub_scopes, 0); + if (sub_scopes->size() == 0) { + sub_scopes->reserve(num_sub_scopes); + for (size_t i = 0; i < num_sub_scopes; ++i) { + sub_scopes->emplace_back(&scope.NewScope()); + } + } - for (size_t i = 0; i < sub_scopes.size(); ++i) { - *sub_scopes[i]->Var(argu)->GetMutable() = lod_tensors[i]; + for (size_t i = 0; i < lod_tensors.size(); ++i) { + *detail::Ref(sub_scopes->at(i)->Var(argu), + "Cannot find variable in the sub-scope", argu) + .GetMutable() = lod_tensors[i]; + } + } +} + +inline void CopyOrShare(const framework::Variable &src, + const platform::Place &dst_place, + framework::Variable *dst) { + if (src.IsType()) { + if (src.Get().place() == dst_place) { + dst->GetMutable()->ShareDataWith(src.Get()); + } else { + Copy(src.Get(), dst_place, dst->GetMutable()); + } + } else if (src.IsType()) { + auto &src_sr = src.Get(); + auto *dst_sr = dst->GetMutable(); + dst_sr->set_rows(src_sr.rows()); + dst_sr->set_height(src_sr.height()); + if (src_sr.value().place() == dst_place) { + dst_sr->mutable_value()->ShareDataWith(src_sr.value()); + } else { + Copy(src_sr.value(), dst_place, dst_sr->mutable_value()); } + } else { + PADDLE_THROW("Expect LoDTensor/SelectedRows, get %s", src.Type().name()); } } +void WaitOnPlace(const platform::Place place) { + platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); + auto &dev_ctx = *pool.Get(place); + dev_ctx.Wait(); +} + void WaitOnPlaces(const std::vector places) { platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); @@ -70,7 +115,7 @@ class ParallelDoOp : public framework::OperatorBase { const framework::VariableNameMap &inputs, const framework::VariableNameMap &outputs, const framework::AttributeMap &attrs) - : OperatorBase(type, inputs, outputs, attrs) {} + : framework::OperatorBase(type, inputs, outputs, attrs) {} void Run(const framework::Scope &scope, const platform::Place &place) const override { @@ -85,19 +130,17 @@ class ParallelDoOp : public framework::OperatorBase { auto &sub_scopes = *scope.FindVar(Output(kParallelScopes)) ->GetMutable>(); - for (size_t place_idx = 0; place_idx < places.size(); ++place_idx) { - sub_scopes.push_back(&scope.NewScope()); - } // split input - SplitTensorAndMoveTensorToScopes(scope, sub_scopes, places, + SplitTensorAndMoveTensorToScopes(scope, &sub_scopes, places, Inputs(kInputs)); + // copy parameter for (auto ¶m : Inputs(kParameters)) { PADDLE_ENFORCE(scope.FindVar(param)->IsType(), "Only support parameter type as LoDTensor"); auto &src = scope.FindVar(param)->Get(); - for (size_t i = 0; i < places.size(); ++i) { + for (size_t i = 0; i < sub_scopes.size(); ++i) { auto &place = places[i]; auto *sub_scope = sub_scopes[i]; auto *dst = sub_scope->Var(param)->GetMutable(); @@ -108,9 +151,7 @@ class ParallelDoOp : public framework::OperatorBase { std::vector> workers; workers.reserve(places.size()); - for (size_t place_idx = 0; place_idx < places.size(); ++place_idx) { - VLOG(3) << "Run " << place_idx; - + for (size_t place_idx = 0; place_idx < sub_scopes.size(); ++place_idx) { auto &place = places[place_idx]; auto *cur_scope = sub_scopes[place_idx]; @@ -157,21 +198,16 @@ ParallelDo Operator. } }; -class ParallelDoGradOp : public OperatorBase { +class ParallelDoGradOp : public framework::OperatorBase { public: ParallelDoGradOp(const std::string &type, const framework::VariableNameMap &inputs, const framework::VariableNameMap &outputs, const framework::AttributeMap &attrs) - : OperatorBase(type, inputs, outputs, attrs) {} + : framework::OperatorBase(type, inputs, outputs, attrs) {} void Run(const framework::Scope &scope, const platform::Place &place) const override { - // // get device context from pool - // platform::DeviceContextPool &pool = - // platform::DeviceContextPool::Instance(); - // auto &dev_ctx = *pool.Get(place); - auto *block = Attr(kParallelBlock); auto *program = block->Program(); @@ -181,26 +217,16 @@ class ParallelDoGradOp : public OperatorBase { auto &places = scope.FindVar(Input(kPlaces))->Get(); // feed output@grad - SplitTensorAndMoveTensorToScopes(scope, sub_scopes, places, - Inputs(framework::GradVarName(kOutputs))); + SplitTensorAndMoveTensorToScopes( + scope, const_cast *>(&sub_scopes), + places, Inputs(framework::GradVarName(kOutputs))); WaitOnPlaces(places); - // for debugging - for (auto &s : Inputs(framework::GradVarName(kOutputs))) { - VLOG(3) << s; - VLOG(3) << scope.FindVar(s)->Get(); - for (auto *sub_scope : sub_scopes) { - VLOG(3) << sub_scope->FindVar(s)->Get(); - } - } - // exe run std::vector> workers; - for (size_t place_idx = 0; place_idx < places.size(); ++place_idx) { - VLOG(3) << "Run " << place_idx; - - auto &place = places[place_idx]; - auto *cur_scope = sub_scopes[place_idx]; + for (size_t i = 0; i < sub_scopes.size(); ++i) { + auto &place = places[i]; + auto *cur_scope = sub_scopes[i]; // execute workers.emplace_back(framework::Async([program, cur_scope, place, block] { @@ -214,35 +240,42 @@ class ParallelDoGradOp : public OperatorBase { } WaitOnPlaces(places); - // merge grad - for (auto &s : Outputs(framework::GradVarName(kParameters))) { - VLOG(3) << "merge grad " << s; - - auto &t = sub_scopes[0]->FindVar(s)->Get(); - VLOG(3) << t; + AccumulateGrad(scope, place, sub_scopes, places); + } - std::string s_buf = s + "@BUF"; - auto *t_buf = sub_scopes[0]->Var(s_buf)->GetMutable(); + void AccumulateGrad(const framework::Scope &scope, + const platform::Place &place, + const std::vector &sub_scopes, + const platform::PlaceList &places) const { + for (auto &s : Outputs(framework::GradVarName(kParameters))) { + std::string tmp_name; + auto *tmp = sub_scopes[0]->Var(&tmp_name); - for (size_t place_idx = 1; place_idx < places.size(); ++place_idx) { - auto &tt = sub_scopes[place_idx]->FindVar(s)->Get(); - VLOG(3) << place_idx; - VLOG(3) << tt; - framework::Copy(tt, places[0], t_buf); + for (size_t i = 1; i < sub_scopes.size(); ++i) { + CopyOrShare(*sub_scopes[i]->FindVar(s), places[0], tmp); + WaitOnPlace(places[0]); auto sum_op = framework::OpRegistry::CreateOp( - "sum", {{"X", {s, s_buf}}}, {{"Out", {s}}}, + "sum", {{"X", {s, tmp_name}}}, {{"Out", {s}}}, framework::AttributeMap{}); + VLOG(3) << sum_op->DebugStringEx(sub_scopes[0]); sum_op->Run(*sub_scopes[0], places[0]); - WaitOnPlaces(places); + WaitOnPlace(places[0]); } - VLOG(3) << t; - framework::Copy(t, place, scope.FindVar(s)->GetMutable()); + CopyOrShare(*sub_scopes[0]->FindVar(s), place, scope.FindVar(s)); } + WaitOnPlaces(places); } }; +std::ostream &operator<<(std::ostream &sout, + const std::vector &strs) { + std::copy(strs.begin(), strs.end(), + std::ostream_iterator(sout, ",")); + return sout; +} + class ParallelDoGradOpDescMaker : public framework::SingleGradOpDescMaker { public: using framework::SingleGradOpDescMaker::SingleGradOpDescMaker; @@ -259,6 +292,17 @@ class ParallelDoGradOpDescMaker : public framework::SingleGradOpDescMaker { this->InputGrad(input_param, false)); } } + auto *g_block = this->grad_block_[0]; + + // All variable name that needed by gradient operators + std::unordered_set all_inputs_in_grad_blocks; + + for (size_t i = 0; i < g_block->OpSize(); ++i) { + auto *op = g_block->Op(i); + for (auto &var_name : op->InputArgumentNames()) { + all_inputs_in_grad_blocks.insert(var_name); + } + } for (auto &output_param : this->OutputNames()) { if (output_param == kParallelScopes) { @@ -267,8 +311,17 @@ class ParallelDoGradOpDescMaker : public framework::SingleGradOpDescMaker { this->Output(output_param)); } else { grad->SetInput(output_param, this->Output(output_param)); - grad->SetInput(framework::GradVarName(output_param), - this->OutputGrad(output_param)); + std::vector og_names; + for (auto &og_name : this->OutputGrad(output_param)) { + if (all_inputs_in_grad_blocks.count(og_name) != 0) { + // there are some gradient operators who need the OG. So make this + // OG as an input of parallel.do + og_names.push_back(og_name); + } + // else, there is no operator who need the OG. Do not use this OG as + // an input + } + grad->SetInput(framework::GradVarName(output_param), og_names); } } grad->SetAttrMap(this->Attrs()); @@ -283,18 +336,30 @@ class ParallelDoGradOpShapeInference : public framework::InferShapeBase { void operator()(framework::InferShapeContext *ctx) const override { std::vector input{kParameters, kInputs}; std::vector output{kOutputs}; - for (auto &s : input) { - PADDLE_ENFORCE(ctx->HasInputs(s)); - PADDLE_ENFORCE(ctx->HasOutputs(framework::GradVarName(s)), - "Cannot find the gradient variable %s", - framework::GradVarName(s)); - } + + PADDLE_ENFORCE(ctx->HasInputs(kParameters)); + PADDLE_ENFORCE(ctx->HasOutputs(framework::GradVarName(kParameters))); + PADDLE_ENFORCE(ctx->HasInputs(kInputs)); + for (auto &s : output) { PADDLE_ENFORCE(ctx->HasInputs(s)); } - for (auto &s : input) { - ctx->SetOutputsDim(framework::GradVarName(s), ctx->GetInputsDim(s)); + + ctx->SetOutputsDim(framework::GradVarName(kParameters), + ctx->GetInputsDim(kParameters)); + + auto i_dims = ctx->GetInputsDim(kInputs); + auto ig_names = ctx->Outputs(framework::GradVarName(kInputs)); + + for (size_t i = 0; i < ig_names.size(); ++i) { + auto &ig_name = ig_names[i]; + if (ig_name == framework::kEmptyVarName) { + continue; + } + + ctx->SetDims({ig_name}, {i_dims[i]}); } + if (ctx->HasInputs(kParameters)) { PADDLE_ENFORCE(ctx->HasOutputs(framework::GradVarName(kParameters))); ctx->SetOutputsDim(framework::GradVarName(kParameters), diff --git a/paddle/operators/pool_op.cc b/paddle/operators/pool_op.cc index 3e567efd082ed913ce3c19f87c93a2868ebe8864..b97333bb1a13a0170c325520b86ac73e68282f91 100644 --- a/paddle/operators/pool_op.cc +++ b/paddle/operators/pool_op.cc @@ -64,6 +64,13 @@ void PoolOp::InferShape(framework::InferShapeContext *ctx) const { framework::OpKernelType PoolOp::GetExpectedKernelType( const framework::ExecutionContext &ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); + use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); +#ifdef PADDLE_WITH_CUDA + if (platform::is_gpu_place(ctx.GetPlace())) { + auto &dev_ctx = ctx.template device_context(); + use_cudnn &= dev_ctx.cudnn_handle() != nullptr; + } +#endif framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; @@ -88,6 +95,13 @@ void PoolOpGrad::InferShape(framework::InferShapeContext *ctx) const { framework::OpKernelType PoolOpGrad::GetExpectedKernelType( const framework::ExecutionContext &ctx) const { bool use_cudnn = ctx.Attr("use_cudnn"); + use_cudnn &= platform::is_gpu_place(ctx.GetPlace()); +#ifdef PADDLE_WITH_CUDA + if (platform::is_gpu_place(ctx.GetPlace())) { + auto &dev_ctx = ctx.template device_context(); + use_cudnn &= dev_ctx.cudnn_handle() != nullptr; + } +#endif framework::LibraryType library_; if (use_cudnn) { library_ = framework::LibraryType::kCUDNN; diff --git a/paddle/operators/pool_op.h b/paddle/operators/pool_op.h index c3d82ecbdeb412f0234fcddc27361d79b58c7122..d6ba5e298a4939e31fde71bf5bf8484640a7ceaf 100644 --- a/paddle/operators/pool_op.h +++ b/paddle/operators/pool_op.h @@ -139,10 +139,8 @@ class PoolGradKernel : public framework::OpKernel { auto& dev_ctx = context.template device_context(); if (in_x_grad) { in_x_grad->mutable_data(context.GetPlace()); - auto temp = framework::EigenVector::Flatten(*in_x_grad); - temp.device( - *context.template device_context().eigen_device()) = - temp.constant(static_cast(0)); + paddle::operators::math::SetConstant set_constant; + set_constant(dev_ctx, in_x_grad, 0.0); switch (ksize.size()) { case 2: { diff --git a/paddle/operators/prior_box_op.cc b/paddle/operators/prior_box_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..105ff4ac3e3ba889aad880f4204af15829c6da47 --- /dev/null +++ b/paddle/operators/prior_box_op.cc @@ -0,0 +1,154 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/prior_box_op.h" + +namespace paddle { +namespace operators { + +class PriorBoxOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("Input"), + "Input(Input) of PriorBoxOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput("Image"), + "Input(Image) of PriorBoxOp should not be null."); + + auto image_dims = ctx->GetInputDim("Image"); + auto input_dims = ctx->GetInputDim("Input"); + PADDLE_ENFORCE(image_dims.size() == 4, "The layout of image is NCHW."); + PADDLE_ENFORCE(input_dims.size() == 4, "The layout of input is NCHW."); + + PADDLE_ENFORCE_LT(input_dims[2], image_dims[2], + "The height of input must smaller than image."); + + PADDLE_ENFORCE_LT(input_dims[3], image_dims[3], + "The width of input must smaller than image."); + + auto min_sizes = ctx->Attrs().Get>("min_sizes"); + auto max_sizes = ctx->Attrs().Get>("max_sizes"); + auto variances = ctx->Attrs().Get>("variances"); + auto aspect_ratios = ctx->Attrs().Get>("aspect_ratios"); + bool flip = ctx->Attrs().Get("flip"); + + PADDLE_ENFORCE_GT(min_sizes.size(), 0, + "Size of min_sizes must be at least 1."); + for (size_t i = 0; i < min_sizes.size(); ++i) { + PADDLE_ENFORCE_GT(min_sizes[i], 0, "min_sizes[%d] must be positive.", i); + } + + std::vector aspect_ratios_vec; + ExpandAspectRatios(aspect_ratios, flip, aspect_ratios_vec); + + int num_priors = aspect_ratios_vec.size() * min_sizes.size(); + if (max_sizes.size() > 0) { + PADDLE_ENFORCE_EQ(max_sizes.size(), min_sizes.size(), + "The number of min_size and max_size must be equal."); + for (size_t i = 0; i < min_sizes.size(); ++i) { + PADDLE_ENFORCE_GT(max_sizes[i], min_sizes[i], + "max_size[%d] must be greater than min_size[%d].", i, + i); + num_priors += 1; + } + } + + PADDLE_ENFORCE_EQ(variances.size(), 4, "Must and only provide 4 variance."); + for (size_t i = 0; i < variances.size(); ++i) { + PADDLE_ENFORCE_GT(variances[i], 0.0, + "variance[%d] must be greater than 0.", i); + } + + const float step_h = ctx->Attrs().Get("step_h"); + PADDLE_ENFORCE_GT(step_h, 0.0, "step_h should be larger than 0."); + const float step_w = ctx->Attrs().Get("step_w"); + PADDLE_ENFORCE_GT(step_w, 0.0, "step_w should be larger than 0."); + + std::vector dim_vec(4); + dim_vec[0] = input_dims[2]; + dim_vec[1] = input_dims[3]; + dim_vec[2] = num_priors; + dim_vec[3] = 4; + ctx->SetOutputDim("Boxes", framework::make_ddim(dim_vec)); + ctx->SetOutputDim("Variances", framework::make_ddim(dim_vec)); + } +}; + +class PriorBoxOpMaker : public framework::OpProtoAndCheckerMaker { + public: + PriorBoxOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("Input", + "(Tensor, default Tensor), " + "the input feature data of PriorBoxOp, The layout is NCHW."); + AddInput("Image", + "(Tensor, default Tensor), " + "the input image data of PriorBoxOp, The layout is NCHW."); + AddOutput("Boxes", + "(Tensor, default Tensor), the output prior boxes of " + "PriorBoxOp. The layout is [H, W, num_priors, 4]. " + "H is the height of input, W is the width of input, num_priors " + "is the box count of each position."); + AddOutput("Variances", + "(Tensor, default Tensor), the expanded variances of " + "PriorBoxOp. The layout is [H, W, num_priors, 4]. " + "H is the height of input, W is the width of input, num_priors " + "is the box count of each position."); + AddAttr>("min_sizes", "(vector) ", + "List of min sizes of generated prior boxes."); + AddAttr>("max_sizes", "(vector) ", + "List of max sizes of generated prior boxes."); + AddAttr>( + "aspect_ratios", "(vector) ", + "List of aspect ratios of generated prior boxes."); + AddAttr>( + "variances", "(vector) ", + "List of variances to be encoded in prior boxes."); + AddAttr("flip", "(bool) ", "Whether to flip aspect ratios.") + .SetDefault(true); + AddAttr("clip", "(bool) ", "Whether to clip out-of-boundary boxes.") + .SetDefault(true); + AddAttr("step_w", + "Prior boxes step across width, 0 for auto calculation.") + .SetDefault(0.0); + AddAttr("step_h", + "Prior boxes step across height, 0 for auto calculation.") + .SetDefault(0.0); + AddAttr("offset", + "(float) " + "Prior boxes center offset.") + .SetDefault(0.5); + AddComment(R"DOC( +Prior box operator +Generate prior boxes for SSD(Single Shot MultiBox Detector) algorithm. +Each position of the input produce N prior boxes, N is determined by + the count of min_sizes, max_sizes and aspect_ratios, The size of the + box is in range(min_size, max_size) interval, which is generated in + sequence according to the aspect_ratios. + +Please get more information from the following papers: +https://arxiv.org/abs/1512.02325. +)DOC"); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OP_WITHOUT_GRADIENT(prior_box, ops::PriorBoxOp, ops::PriorBoxOpMaker); +REGISTER_OP_CPU_KERNEL( + prior_box, ops::PriorBoxOpKernel, + ops::PriorBoxOpKernel); diff --git a/paddle/operators/prior_box_op.h b/paddle/operators/prior_box_op.h new file mode 100644 index 0000000000000000000000000000000000000000..e0a663ace8f38c2d08fd4714c1247d3313ffae3e --- /dev/null +++ b/paddle/operators/prior_box_op.h @@ -0,0 +1,188 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once +#include "paddle/framework/op_registry.h" +#include "paddle/operators/math/math_function.h" +#include "paddle/platform/transform.h" + +namespace paddle { +namespace operators { + +inline void ExpandAspectRatios(const std::vector& input_aspect_ratior, + bool flip, + std::vector& output_aspect_ratior) { + constexpr float epsilon = 1e-6; + output_aspect_ratior.clear(); + output_aspect_ratior.push_back(1.); + for (size_t i = 0; i < input_aspect_ratior.size(); ++i) { + float ar = input_aspect_ratior[i]; + bool already_exist = false; + for (size_t j = 0; j < output_aspect_ratior.size(); ++j) { + if (fabs(ar - output_aspect_ratior[j]) < epsilon) { + already_exist = true; + break; + } + } + if (!already_exist) { + output_aspect_ratior.push_back(ar); + if (flip) { + output_aspect_ratior.push_back(1. / ar); + } + } + } +} + +template +struct ClipFunctor { + HOSTDEVICE T operator()(T in) const { + return std::min(std::max(in, 0.), 1.); + } +}; + +template +class PriorBoxOpKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + auto* input = ctx.Input("Input"); + auto* image = ctx.Input("Image"); + auto* boxes = ctx.Output("Boxes"); + auto* vars = ctx.Output("Variances"); + + auto min_sizes = ctx.Attr>("min_sizes"); + auto max_sizes = ctx.Attr>("max_sizes"); + auto input_aspect_ratio = ctx.Attr>("aspect_ratios"); + auto variances = ctx.Attr>("variances"); + auto flip = ctx.Attr("flip"); + auto clip = ctx.Attr("clip"); + + std::vector aspect_ratios; + ExpandAspectRatios(input_aspect_ratio, flip, aspect_ratios); + + T step_w = static_cast(ctx.Attr("step_w")); + T step_h = static_cast(ctx.Attr("step_h")); + T offset = static_cast(ctx.Attr("offset")); + + auto img_width = image->dims()[3]; + auto img_height = image->dims()[2]; + + auto feature_width = input->dims()[3]; + auto feature_height = input->dims()[2]; + + T step_width, step_height; + if (step_w == 0 || step_h == 0) { + step_width = static_cast(img_width) / feature_width; + step_height = static_cast(img_height) / feature_height; + } else { + step_width = step_w; + step_height = step_h; + } + + int num_priors = aspect_ratios.size() * min_sizes.size(); + if (max_sizes.size() > 0) { + num_priors += max_sizes.size(); + } + + boxes->mutable_data(ctx.GetPlace()); + vars->mutable_data(ctx.GetPlace()); + + auto e_boxes = framework::EigenTensor::From(*boxes); + for (int h = 0; h < feature_height; ++h) { + for (int w = 0; w < feature_width; ++w) { + T center_x = (w + offset) * step_width; + T center_y = (h + offset) * step_height; + T box_width, box_height; + int idx = 0; + for (size_t s = 0; s < min_sizes.size(); ++s) { + int min_size = min_sizes[s]; + // first prior: aspect_ratio = 1, size = min_size + box_width = box_height = min_size; + // xmin + e_boxes(h, w, idx, 0) = (center_x - box_width / 2.) / img_width; + // ymin + e_boxes(h, w, idx, 1) = (center_y - box_height / 2.) / img_height; + // xmax + e_boxes(h, w, idx, 2) = (center_x + box_width / 2.) / img_width; + // ymax + e_boxes(h, w, idx, 3) = (center_y + box_height / 2.) / img_height; + + idx++; + if (max_sizes.size() > 0) { + int max_size = max_sizes[s]; + // second prior: aspect_ratio = 1, + // size = sqrt(min_size * max_size) + box_width = box_height = sqrt(min_size * max_size); + // xmin + e_boxes(h, w, idx, 0) = (center_x - box_width / 2.) / img_width; + // ymin + e_boxes(h, w, idx, 1) = (center_y - box_height / 2.) / img_height; + // xmax + e_boxes(h, w, idx, 2) = (center_x + box_width / 2.) / img_width; + // ymax + e_boxes(h, w, idx, 3) = (center_y + box_height / 2.) / img_height; + idx++; + } + + // rest of priors + for (size_t r = 0; r < aspect_ratios.size(); ++r) { + float ar = aspect_ratios[r]; + if (fabs(ar - 1.) < 1e-6) { + continue; + } + box_width = min_size * sqrt(ar); + box_height = min_size / sqrt(ar); + // xmin + e_boxes(h, w, idx, 0) = (center_x - box_width / 2.) / img_width; + // ymin + e_boxes(h, w, idx, 1) = (center_y - box_height / 2.) / img_height; + // xmax + e_boxes(h, w, idx, 2) = (center_x + box_width / 2.) / img_width; + // ymax + e_boxes(h, w, idx, 3) = (center_y + box_height / 2.) / img_height; + idx++; + } + } + } + } + + if (clip) { + platform::Transform trans; + ClipFunctor clip_func; + trans(ctx.template device_context(), + boxes->data(), boxes->data() + boxes->numel(), + boxes->data(), clip_func); + } + + framework::Tensor var_t; + var_t.mutable_data( + framework::make_ddim({1, static_cast(variances.size())}), + ctx.GetPlace()); + auto var_et = framework::EigenTensor::From(var_t); + for (size_t i = 0; i < variances.size(); ++i) { + var_et(0, i) = variances[i]; + } + + int box_num = feature_height * feature_width * num_priors; + auto var_dim = vars->dims(); + vars->Resize({box_num, static_cast(variances.size())}); + + auto e_vars = framework::EigenMatrix::From(*vars); + e_vars = var_et.broadcast(Eigen::DSizes(box_num, 1)); + + vars->Resize(var_dim); + } +}; // namespace operators + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/recv_op.cc b/paddle/operators/recv_op.cc index f9ed7516826319da422fbb0af4e5c277afa7ae40..49e1eb3402482e7ff12d9b2b640f7271a80cf6d9 100644 --- a/paddle/operators/recv_op.cc +++ b/paddle/operators/recv_op.cc @@ -27,12 +27,13 @@ limitations under the License. */ #include "paddle/operators/detail/grpc_server.h" #include "paddle/operators/detail/sendrecvop_utils.h" #include "paddle/operators/detail/simple_block_queue.h" - -#define LISTEN_TERMINATE_MESSAGE "TERMINATE@RECV" +#include "paddle/string/printf.h" namespace paddle { namespace operators { +constexpr char kOptimizeBlock[] = "OptimizeBlock"; + void RunServer(std::shared_ptr service) { service->RunSyncUpdate(); VLOG(4) << "RunServer thread end"; @@ -46,7 +47,7 @@ static void CreateTensorFromMessageType(framework::Variable *var, var->GetMutable(); } else { PADDLE_THROW( - "VraibleMessage type %d is not in " + "VariableMessage type %d is not in " "[LoDTensor, SelectedRows]", var_type); } @@ -77,93 +78,84 @@ class RecvOp : public framework::OperatorBase { if (grads_counter_.find(varname) == grads_counter_.end()) { grads_counter_[varname] = 0; } - char ret[256]; - snprintf(ret, sizeof(ret), "%s.trainer_%d", varname.c_str(), - grads_counter_[varname]++); - return std::string(ret); + return string::Sprintf("%s.trainer_%d", varname, grads_counter_[varname]++); } void Run(const framework::Scope &scope, const platform::Place &dev_place) const override { - // FIXME(typhoonzero): no new scopes for every run. - framework::Scope &recv_scope = scope.NewScope(); platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); auto &dev_ctx = *pool.Get(dev_place); + framework::Scope &recv_scope = scope.NewScope(); // FIXME(Yancey1989): initialize rpc server with laze mode. rpc_service_->SetScope(&recv_scope); rpc_service_->SetDevCtx(&dev_ctx); auto param_list = Attr>("ParamList"); auto grad_list = Attr>("GradList"); - auto trainer_count = Attr("Trainers"); - size_t param_count = param_list.size(); + auto fan_in = Attr("Fanin"); + + auto *block = Attr(kOptimizeBlock); + auto *program = block->Program(); + framework::Executor executor(dev_place); - rpc_service_->Reset(); // TODO(typhoonzero): change this to a while_op for every cluster-batch. bool exit_flag = false; - VLOG(4) << "param_count:" << param_count - << " trainer_count:" << trainer_count; while (!exit_flag) { - // TODO(gognwb): simply this loop. - // Get from multiple trainers, we don't care about order in which - // the gradient arrives, just add suffix 0~n then average the gradient. - for (size_t i = 0; i < param_count * trainer_count; ++i) { - // blocking get one var from client. + // Get from multiple trainers, we don't care about the order in which + // the gradients arrives, just add suffix 0~n and merge the gradient. + rpc_service_->SetCond(0); + size_t recv_var_cnt = 0; + int batch_barrier = 0; + while (batch_barrier != fan_in) { const detail::MessageWithName &v = rpc_service_->Get(); auto grad_var_name = v.first; if (grad_var_name == LISTEN_TERMINATE_MESSAGE) { - VLOG(4) << "received LISTEN_TERMINATE_MESSAGE and RunOp.Run() exit"; + LOG(INFO) << "received terminate message and exit"; exit_flag = true; break; - } - auto it = std::find(grad_list.begin(), grad_list.end(), grad_var_name); - std::string param_var_name; - if (it != grad_list.end()) { - param_var_name = param_list[it - grad_list.begin()]; + } else if (grad_var_name == BATCH_BARRIER_MESSAGE) { + VLOG(3) << "recv batch barrier message"; + batch_barrier++; + continue; } else { - LOG(ERROR) << "grad have no paired param found!\"" << grad_var_name - << "\""; - } - VLOG(3) << "recved grad: " << grad_var_name - << " updating param: " << param_var_name; - - auto *merged_grad = recv_scope.FindVar(grad_var_name); - if (merged_grad == nullptr) { - auto *ptr = recv_scope.Var(grad_var_name); - CreateTensorFromMessageType(ptr, v.second.type()); - VLOG(3) << "Create Variable " << grad_var_name - << " on recv scope, which pointer is " << ptr << " type is " - << v.second.type(); + // receive a variable + recv_var_cnt++; + auto it = + std::find(grad_list.begin(), grad_list.end(), grad_var_name); + std::string param_var_name; + if (it != grad_list.end()) { + param_var_name = param_list[it - grad_list.begin()]; + } else { + LOG(ERROR) << "grad has no paired param:" << grad_var_name; + } + VLOG(3) << "received grad: " << grad_var_name + << " updating param: " << param_var_name; + + if (fan_in > 1) { + grad_var_name = this->GetGradVarNameForTrainer(grad_var_name); + } + auto *var = recv_scope.FindVar(grad_var_name); + if (var == nullptr) { + LOG(ERROR) << "Can not find server side var: " << grad_var_name; + PADDLE_THROW("Can not find server side var"); + } + detail::DeserializeFromMessage(v.second, dev_ctx, var); } - - if (trainer_count > 1) { - grad_var_name = this->GetGradVarNameForTrainer(grad_var_name); - } - - auto *var = recv_scope.Var(grad_var_name); - detail::DeserializeFromMessage(v.second, dev_ctx, var); } - + VLOG(3) << "recv " << recv_var_cnt << " parmeters for one barrier."; + // TODO(Yancey1989): merge SelectedRows variables here if (exit_flag) { break; } - rpc_service_->Reset(); - - std::string program_str = Attr("OptimizeProgram"); - framework::proto::ProgramDesc program_desc; - program_desc.ParseFromString(program_str); - framework::ProgramDesc program(program_desc); - framework::Executor executor(dev_place); - // Run sub graph to get optimized tensor try { - executor.Run(program, &recv_scope, 0, /*global_block*/ + executor.Run(*program, &recv_scope, block->ID(), /*global_block*/ false /*create_local_scope*/, false /*create_vars*/); } catch (std::exception &e) { LOG(ERROR) << "run sub program error " << e.what(); } - - rpc_service_->Done(); + rpc_service_->SetCond(1); + rpc_service_->WaitClientGet(recv_var_cnt); grads_counter_.clear(); } // while(true) } @@ -178,28 +170,27 @@ class RecvOpMaker : public framework::OpProtoAndCheckerMaker { public: RecvOpMaker(OpProto *proto, OpAttrChecker *op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { - AddInput("RX", "(Tensor) Input tensor to be optimized").AsDuplicable(); AddComment(R"DOC( Recv operator -This operator will recv tensor from send_op +This operator will recieve tensor from send_op )DOC"); AddAttr("endpoint", "(string, default 127.0.0.1:6164)" "IP address to listen on.") .SetDefault("127.0.0.1:6164") .AddCustomChecker([](const std::string &ip) { return !ip.empty(); }); - AddAttr("OptimizeProgram", "type string", - "Serialized ProgramDesc string for recv to run."); + AddAttr( + kOptimizeBlock, "Serialized ProgramDesc string for recv to run."); AddAttr>( "ParamList", "type list of string", - "grad->param name mapping to find which param to optimize.") + "grad->param name mapping to find which parameters to optimize.") .SetDefault({}); AddAttr>( "GradList", "type list of string", - "grad->param name mapping to find which param to optimize.") + "grad->param name mapping to find which parameters to optimize.") .SetDefault({}); - AddAttr("Trainers", "type int", + AddAttr("Fanin", "type int", "Number of trainers in the current cluster job") .SetDefault(1); } diff --git a/paddle/operators/reduce_op.cc b/paddle/operators/reduce_op.cc index 172d28bb3b647901d4de7bc03c9de21e3468a364..84f24a909597915f0eebb6c9cad37510cbe93e7b 100644 --- a/paddle/operators/reduce_op.cc +++ b/paddle/operators/reduce_op.cc @@ -13,7 +13,6 @@ See the License for the specific language governing permissions and limitations under the License. */ #include "paddle/operators/reduce_op.h" -#include "paddle/operators/net_op.h" namespace paddle { namespace operators { @@ -38,10 +37,14 @@ class ReduceOp : public framework::OperatorWithKernel { dim, x_rank, "The dim should be in the range [-rank(input), rank(input))."); bool reduce_all = ctx->Attrs().Get("reduce_all"); + bool keep_dim = ctx->Attrs().Get("keep_dim"); if (reduce_all) { - ctx->SetOutputDim("Out", {1}); + if (keep_dim) + ctx->SetOutputDim( + "Out", framework::make_ddim(std::vector(x_rank, 1))); + else + ctx->SetOutputDim("Out", {1}); } else { - bool keep_dim = ctx->Attrs().Get("keep_dim"); auto dims_vector = vectorize(x_dims); if (keep_dim || x_rank == 1) { dims_vector[dim] = 1; @@ -129,7 +132,7 @@ If reduce_all is true, just reduce along all dimensions and output a scalar. } void SetComment(std::string name, std::string op) { - Replace(comment_, "{ReduceOP}", name); + Replace(comment_, "{ReduceOp}", name); Replace(comment_, "{reduce}", op); } }; @@ -190,10 +193,22 @@ REGISTER_OP(reduce_min, ops::ReduceOp, ops::ReduceMinOpMaker, reduce_min_grad, #define REGISTER_REDUCE_CPU_KERNEL(reduce_type, functor, grad_functor) \ REGISTER_OP_CPU_KERNEL(reduce_type, \ ops::ReduceKernel); \ + float, ops::functor>, \ + ops::ReduceKernel, \ + ops::ReduceKernel, \ + ops::ReduceKernel); \ REGISTER_OP_CPU_KERNEL( \ reduce_type##_grad, \ ops::ReduceGradKernel, \ + ops::ReduceGradKernel, \ + ops::ReduceGradKernel, \ + ops::ReduceGradKernel); FOR_EACH_KERNEL_FUNCTOR(REGISTER_REDUCE_CPU_KERNEL); diff --git a/paddle/operators/reduce_op.cu b/paddle/operators/reduce_op.cu index 1dd948ed8a79cce8468f2fe210b5636e7dd1f99e..4ed1e051db4df579afe1c1ca24a06fa1baf3e13a 100644 --- a/paddle/operators/reduce_op.cu +++ b/paddle/operators/reduce_op.cu @@ -20,10 +20,22 @@ namespace ops = paddle::operators; #define REGISTER_REDUCE_GPU_KERNEL(reduce_type, functor, grad_functor) \ REGISTER_OP_CUDA_KERNEL( \ reduce_type, ops::ReduceKernel); \ + float, ops::functor>, \ + ops::ReduceKernel, \ + ops::ReduceKernel, \ + ops::ReduceKernel); \ REGISTER_OP_CUDA_KERNEL( \ reduce_type##_grad, \ ops::ReduceGradKernel, \ + ops::ReduceGradKernel, \ + ops::ReduceGradKernel, \ + ops::ReduceGradKernel); FOR_EACH_KERNEL_FUNCTOR(REGISTER_REDUCE_GPU_KERNEL); diff --git a/paddle/operators/reshape_op.cc b/paddle/operators/reshape_op.cc index 58e8fd6124d8c076337ae9bb2f5103e7a3cb7ff0..b9743a5df1092917d13a50aa20ea7e7c52b8d151 100644 --- a/paddle/operators/reshape_op.cc +++ b/paddle/operators/reshape_op.cc @@ -90,14 +90,10 @@ Reshape Operator. Reshape Input(X) into the shape specified by Attr(shape). An example: -Given a 2-D tensor X with 2 rows and 2 columns - - [[1, 2], [3, 4]] +Given a 2-D tensor X with 2 rows and 2 columns : [[1, 2], [3, 4]] and target shape = [1, 4], the reshape operator will transform -the tensor X into a 2-D tensor: - - [[1, 2, 3, 4]] +the tensor X into a 2-D tensor: [[1, 2, 3, 4]] One dimension in the target shape can be set -1, representing that its size is unknown. In this case, the real dimension will be infered from diff --git a/paddle/operators/row_conv_op.cu b/paddle/operators/row_conv_op.cu index 41f2c5b9de91ade15b4010f56377675cfd1b611c..b3825212e1ac41b13a2f4cad2c128da39c5f6e71 100644 --- a/paddle/operators/row_conv_op.cu +++ b/paddle/operators/row_conv_op.cu @@ -307,7 +307,7 @@ class RowConvKernel int input_dim = X->dims()[1]; int num_sequence = batch_indices.size() - 1; int future_context = Filter->dims()[0]; - size_t *idx = batch_indices.data(); + size_t *idx = batch_indices.cuda_data(); auto stream = context.cuda_device_context().stream(); if (future_context <= 32) { @@ -345,7 +345,7 @@ class RowConvGradKernel int input_dim = X->dims()[1]; int num_sequence = batch_indices.size() - 1; int future_context = Filter->dims()[0]; - size_t *idx = batch_indices.data(); + size_t *idx = batch_indices.cuda_data(); auto &device_ctx = context.cuda_device_context(); math::SetConstant zero; diff --git a/paddle/operators/save_combine_op.cc b/paddle/operators/save_combine_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..bffa2908bc42d73332f22fa3706d24ab49cd4b38 --- /dev/null +++ b/paddle/operators/save_combine_op.cc @@ -0,0 +1,141 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include +#include +#include +#include +#include +#include "paddle/framework/data_type.h" +#include "paddle/framework/framework.pb.h" +#include "paddle/framework/lod_tensor.h" +#include "paddle/framework/op_registry.h" +#include "paddle/platform/device_context.h" + +namespace paddle { +namespace operators { + +// TODO(sidgoyal78): These function are needed by other files (save_op), move +// them to paddle::filesystem namespace. (as noted by yuyang18 in save_op). +constexpr char kSEP = '/'; +static bool FileExists(const std::string &filepath) { + struct stat buffer; + return (stat(filepath.c_str(), &buffer) == 0); +} + +static std::string DirName(const std::string &filepath) { + auto pos = filepath.rfind(kSEP); + if (pos == std::string::npos) { + return ""; + } + return filepath.substr(0, pos); +} + +static void MkDir(const char *path) { + if (mkdir(path, 0755)) { + PADDLE_ENFORCE_EQ(errno, EEXIST, "%s mkdir failed!", path); + } +} + +static void MkDirRecursively(const char *fullpath) { + if (*fullpath == '\0') return; // empty string + if (FileExists(fullpath)) return; + + MkDirRecursively(DirName(fullpath).c_str()); + MkDir(fullpath); +} + +class SaveCombineOp : public framework::OperatorBase { + public: + SaveCombineOp(const std::string &type, + const framework::VariableNameMap &inputs, + const framework::VariableNameMap &outputs, + const framework::AttributeMap &attrs) + : OperatorBase(type, inputs, outputs, attrs) {} + void Run(const framework::Scope &scope, + const platform::Place &place) const override { + auto filename = Attr("file_path"); + auto overwrite = Attr("overwrite"); + + bool is_present = FileExists(filename); + if (is_present && !overwrite) { + PADDLE_THROW("%s exists!, cannot save_combine to it when overwrite=false", + filename, overwrite); + } + + MkDirRecursively(DirName(filename).c_str()); + std::ofstream fout(filename); + PADDLE_ENFORCE(static_cast(fout), "Cannot open %s to write", + filename); + + auto inp_var_names = Inputs("X"); + PADDLE_ENFORCE_GT(static_cast(inp_var_names.size()), 0, + "The number of input variables should be greater than 0"); + + // get device context from pool + platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance(); + auto &dev_ctx = *pool.Get(place); + + for (size_t i = 0; i < inp_var_names.size(); i++) { + auto *var = scope.FindVar(inp_var_names[i]); + + PADDLE_ENFORCE(var != nullptr, + "Cannot find variable %s for save_combine_op", + inp_var_names[i]); + PADDLE_ENFORCE(var->IsType(), + "SaveCombineOp only supports LoDTensor, %s has wrong type", + inp_var_names[i]); + + auto &tensor = var->Get(); + // Serialize tensor + framework::SerializeToStream(fout, tensor, dev_ctx); + } + fout.close(); + } +}; + +class SaveCombineOpProtoMaker : public framework::OpProtoAndCheckerMaker { + public: + SaveCombineOpProtoMaker(OpProto *proto, OpAttrChecker *op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput( + "X", + "(vector) Input LoDTensors that need to be saved together in a file.") + .AsDuplicable(); + AddComment(R"DOC( +SaveCombine operator + +This operator will serialize and write a list of input LoDTensor variables +to a file on disk. +)DOC"); + AddAttr("overwrite", + "(boolean, default true)" + "Overwrite the output file if it exists.") + .SetDefault(true); + AddAttr( + "file_path", + "(string)" + "The \"file_path\" where the LoDTensor variables will be saved.") + .AddCustomChecker( + [](const std::string &path) { return !path.empty(); }); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; + +REGISTER_OPERATOR(save_combine, ops::SaveCombineOp, + ops::SaveCombineOpProtoMaker); diff --git a/paddle/operators/save_load_combine_op_test.cc b/paddle/operators/save_load_combine_op_test.cc new file mode 100644 index 0000000000000000000000000000000000000000..f3ddc4a6c55d72e4e444869a1ebcd7662c892317 --- /dev/null +++ b/paddle/operators/save_load_combine_op_test.cc @@ -0,0 +1,180 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include +#include +#include +#include "gtest/gtest.h" +#include "paddle/framework/op_registry.h" + +USE_NO_KERNEL_OP(save_combine); +USE_NO_KERNEL_OP(load_combine); + +int* CreateForSaveCombineOp(int x, int y, const std::vector& lod_info, + std::string var_name, + paddle::platform::CPUPlace& place, + paddle::framework::Scope& scope, + paddle::framework::LoD& expect_lod) { + auto var = scope.Var(var_name); + auto tensor = var->GetMutable(); + tensor->Resize({x, y}); + expect_lod.resize(1); + for (size_t i = 0; i < lod_info.size(); i++) { + expect_lod[0].push_back(lod_info[i]); + } + tensor->set_lod(expect_lod); + int* expect = tensor->mutable_data(place); + for (int64_t i = 0; i < tensor->numel(); ++i) { + expect[i] = static_cast(i); + } + return expect; +} + +paddle::framework::LoDTensor* GeneratePlaceholderBeforeLoad( + const std::string out_var_name, paddle::framework::Scope& scope) { + auto load_var = scope.Var(out_var_name); + auto target = load_var->GetMutable(); + return target; +} + +int* GetValuesAfterLoadCombineOp(paddle::framework::LoDTensor* target, + paddle::framework::Scope& scope, + paddle::framework::LoD& actual_lod) { + int* actual = target->data(); + actual_lod = target->lod(); + return actual; +} + +void CheckValues(int* expect, int* actual, paddle::framework::LoD expect_lod, + paddle::framework::LoD actual_lod, const int& numel) { + for (int64_t i = 0; i < numel; ++i) { + EXPECT_EQ(expect[i], actual[i]); + } + EXPECT_EQ(expect_lod.size(), actual_lod.size()); + for (size_t i = 0; i < expect_lod.size(); ++i) { + for (size_t j = 0; j < expect_lod[i].size(); ++j) { + EXPECT_EQ(expect_lod[i][j], actual_lod[i][j]); + } + } +} + +// Here, we create 4 LoDTensors and use save_combine_op to first save these +// in a single file. Then, we use load_combine_op to load these sequentially +TEST(SaveLoadCombineOp, CPU) { + paddle::framework::Scope scope; + paddle::platform::CPUPlace place; + + std::vector lod1 = {0, 1, 2, 3, 10}; + int numel1 = 100; + paddle::framework::LoD expect_lod1; + int* expect1 = CreateForSaveCombineOp(10, 10, lod1, "test_var1", place, scope, + expect_lod1); + + std::vector lod2 = {0, 2, 5, 10}; + int numel2 = 200; + paddle::framework::LoD expect_lod2; + int* expect2 = CreateForSaveCombineOp(10, 20, lod2, "test_var2", place, scope, + expect_lod2); + + std::vector lod3 = {0, 2, 3, 20}; + int numel3 = 4000; + paddle::framework::LoD expect_lod3; + int* expect3 = CreateForSaveCombineOp(20, 200, lod3, "test_var3", place, + scope, expect_lod3); + + std::vector lod4 = {0, 1, 20}; + int numel4 = 1000; + paddle::framework::LoD expect_lod4; + int* expect4 = CreateForSaveCombineOp(20, 50, lod4, "test_var4", place, scope, + expect_lod4); + + // Set attributes + std::string filename = "check_tensor.ls"; + paddle::framework::AttributeMap attrs; + attrs.insert({"file_path", std::string(filename)}); + + // Run the save_combine_op + auto save_combine_op = paddle::framework::OpRegistry::CreateOp( + "save_combine", + {{"X", {"test_var1", "test_var2", "test_var3", "test_var4"}}}, {}, attrs); + save_combine_op->Run(scope, place); + + // Set up output vars + auto target1 = GeneratePlaceholderBeforeLoad("out_var1", scope); + auto target2 = GeneratePlaceholderBeforeLoad("out_var2", scope); + auto target3 = GeneratePlaceholderBeforeLoad("out_var3", scope); + auto target4 = GeneratePlaceholderBeforeLoad("out_var4", scope); + + // Run the load_combine_op + auto load_combine_op = paddle::framework::OpRegistry::CreateOp( + "load_combine", {}, + {{"Out", {"out_var1", "out_var2", "out_var3", "out_var4"}}}, attrs); + load_combine_op->Run(scope, place); + + paddle::framework::LoD actual_lod1, actual_lod2, actual_lod3, actual_lod4; + int* actual1 = GetValuesAfterLoadCombineOp(target1, scope, actual_lod1); + int* actual2 = GetValuesAfterLoadCombineOp(target2, scope, actual_lod2); + int* actual3 = GetValuesAfterLoadCombineOp(target3, scope, actual_lod3); + int* actual4 = GetValuesAfterLoadCombineOp(target4, scope, actual_lod4); + + CheckValues(expect1, actual1, expect_lod1, actual_lod1, numel1); + CheckValues(expect2, actual2, expect_lod2, actual_lod2, numel2); + CheckValues(expect3, actual3, expect_lod3, actual_lod3, numel3); + CheckValues(expect4, actual4, expect_lod4, actual_lod4, numel4); +} + +// Test with original SaveLoadTest +TEST(SaveLoadTestWithCombineOp, CPU) { + paddle::framework::Scope scope; + paddle::platform::CPUPlace place; + + auto var = scope.Var("test_var"); + auto tensor = var->GetMutable(); + tensor->Resize({3, 10}); + paddle::framework::LoD expect_lod; + expect_lod.resize(1); + expect_lod[0].push_back(0); + expect_lod[0].push_back(1); + expect_lod[0].push_back(2); + expect_lod[0].push_back(3); + + tensor->set_lod(expect_lod); + int* expect = tensor->mutable_data(place); + for (int64_t i = 0; i < tensor->numel(); ++i) { + expect[i] = static_cast(i); + } + paddle::framework::AttributeMap attrs; + attrs.insert({"file_path", std::string("check_t.save")}); + + auto save_op = paddle::framework::OpRegistry::CreateOp( + "save_combine", {{"X", {"test_var"}}}, {}, attrs); + save_op->Run(scope, place); + + auto load_var = scope.Var("out_var"); + auto target = load_var->GetMutable(); + auto load_op = paddle::framework::OpRegistry::CreateOp( + "load_combine", {}, {{"Out", {"out_var"}}}, attrs); + load_op->Run(scope, place); + int* actual = target->data(); + for (int64_t i = 0; i < tensor->numel(); ++i) { + EXPECT_EQ(expect[i], actual[i]); + } + auto& actual_lod = target->lod(); + EXPECT_EQ(expect_lod.size(), actual_lod.size()); + for (size_t i = 0; i < expect_lod.size(); ++i) { + for (size_t j = 0; j < expect_lod[i].size(); ++j) { + EXPECT_EQ(expect_lod[i][j], actual_lod[i][j]); + } + } +} diff --git a/paddle/operators/save_load_op_test.cc b/paddle/operators/save_load_op_test.cc index 40103d864fb58804b39ca5f3c63e802a430ce886..d829d5da174b73613da9dcfcd308a5b05e12bce9 100644 --- a/paddle/operators/save_load_op_test.cc +++ b/paddle/operators/save_load_op_test.cc @@ -24,7 +24,7 @@ TEST(SaveLoadOp, CPU) { auto var = scope.Var("test_var"); auto tensor = var->GetMutable(); - tensor->Resize({10, 10}); + tensor->Resize({3, 10}); paddle::framework::LoD expect_lod; expect_lod.resize(1); expect_lod[0].push_back(0); diff --git a/paddle/operators/scale_op.cc b/paddle/operators/scale_op.cc index f634ebe9a2a4648bd08f00af635ef22e8d86a8de..c0e614743a894dece2cdc395d0b28df7e86e921d 100644 --- a/paddle/operators/scale_op.cc +++ b/paddle/operators/scale_op.cc @@ -48,7 +48,7 @@ Scale operator $$Out = scale*X$$ )DOC"); AddAttr("scale", - "(float, default 0)" + "(float, default 1.0)" "The scaling factor of the scale operator.") .SetDefault(1.0); } diff --git a/paddle/operators/send_op.cc b/paddle/operators/send_op.cc index 7c81a9524d6609a65b3167d95053bf4e85eef0db..be41b527f2289d5d657a58f3cb6d7be725323cd0 100644 --- a/paddle/operators/send_op.cc +++ b/paddle/operators/send_op.cc @@ -37,35 +37,53 @@ class SendOp : public framework::OperatorBase { auto ins = Inputs("X"); auto outs = Outputs("Out"); std::vector epmap = Attr>("epmap"); + std::vector endpoints = + Attr>("endpoints"); platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance(); auto& ctx = *pool.Get(place); + + auto client_var_name = Output("RPCClient"); + PADDLE_ENFORCE_NOT_NULL(scope.FindVar(client_var_name), + "Can not find variable '%s' in the scope.", + client_var_name); + auto* client_var = scope.FindVar(client_var_name); + detail::RPCClient* rpc_client = client_var->GetMutable(); + for (size_t i = 0; i < ins.size(); i++) { - client_.AsyncSendVariable(epmap[i], ctx, scope, ins[i]); + VLOG(3) << "sending " << ins[i] << " to " << epmap[i]; + rpc_client->AsyncSendVariable(epmap[i], ctx, scope, ins[i]); } + PADDLE_ENFORCE(rpc_client->Wait()); - for (size_t i = 0; i < outs.size(); i++) { - client_.AsyncGetVariable(epmap[i], ctx, scope, outs[i]); + for (auto& ep : endpoints) { + VLOG(3) << "batch barrier, ep: " << ep; + rpc_client->AsyncSendBatchBarrier(ep); } + PADDLE_ENFORCE(rpc_client->Wait()); - PADDLE_ENFORCE(client_.Wait()); + for (size_t i = 0; i < outs.size(); i++) { + VLOG(3) << "getting " << outs[i] << " from " << epmap[i]; + rpc_client->AsyncGetVariable(epmap[i], ctx, scope, outs[i]); + } + PADDLE_ENFORCE(rpc_client->Wait()); } - - private: - mutable detail::RPCClient client_; }; class SendOpMaker : public framework::OpProtoAndCheckerMaker { public: SendOpMaker(OpProto* proto, OpAttrChecker* op_checker) : OpProtoAndCheckerMaker(proto, op_checker) { - AddInput("X", "(Tensor) Input tensor to be send").AsDuplicable(); - AddOutput("Out", "(Tensor) Output tensor to get from server") + AddInput("X", "(Tensor) Input tensor to be sent").AsDuplicable(); + AddOutput("Out", "(Tensor) Output tensor to be received from server") .AsDuplicable(); + AddOutput("RPCClient", + "(RPCClient) The RPC client object which is" + "initialized at most once."); AddComment(R"DOC( -Recv operator +Send operator -This operator will send tensor to recv_op. +This operator will send tensor to recv_op at the parameter server. )DOC"); AddAttr>("endpoints", "(string vector, default 127.0.0.1:6164)" diff --git a/paddle/operators/send_recv_op_test.cc b/paddle/operators/send_recv_op_test.cc index ea091694798475dfd9631910a750405be950c20c..045a0f5434f339bab345d14881ed05450ce6588d 100644 --- a/paddle/operators/send_recv_op_test.cc +++ b/paddle/operators/send_recv_op_test.cc @@ -130,10 +130,7 @@ void StartServerNet(bool is_sparse) { attrs.insert({"endpoint", std::string("127.0.0.1:6174")}); attrs.insert({"ParamList", std::vector({"Out"})}); attrs.insert({"GradList", std::vector({"x1"})}); - std::string program_proto; - PADDLE_ENFORCE(program.Proto()->SerializeToString(&program_proto)); - - attrs.insert({"OptimizeProgram", program_proto}); + attrs.insert({"OptimizeBlock", block}); recv_op = f::OpRegistry::CreateOp("recv", {{"RX", {"x1"}}}, {}, attrs); recv_op->Run(scope, place); } diff --git a/paddle/operators/sequence_erase_op.cc b/paddle/operators/sequence_erase_op.cc index d17b2686238b2d2f872331edfdbb095fb8693b87..aa0c00aa6f7854ee5e34aef78970971b78df6514 100644 --- a/paddle/operators/sequence_erase_op.cc +++ b/paddle/operators/sequence_erase_op.cc @@ -86,4 +86,5 @@ REGISTER_OP_WITHOUT_GRADIENT(sequence_erase, ops::SequenceEraseOp, ops::SequenceEraseOpMaker); REGISTER_OP_CPU_KERNEL( sequence_erase, - ops::SequenceEraseKernel); + ops::SequenceEraseKernel, + ops::SequenceEraseKernel); diff --git a/paddle/operators/sequence_erase_op.cu b/paddle/operators/sequence_erase_op.cu index 5da8eba3e1ac1fb85dfc65c2fd801574599e02d9..a5311f15f0c607c880a6f12c0bef10b2dd8c8a79 100644 --- a/paddle/operators/sequence_erase_op.cu +++ b/paddle/operators/sequence_erase_op.cu @@ -23,27 +23,22 @@ using platform::PADDLE_CUDA_NUM_THREADS; using LoDTensor = framework::LoDTensor; template -__global__ void LabelErasedIdx(const T* in_dat, const int in_len, - const T* tokens, const int tokens_len, - int* num_erased) { +__global__ void LabelErasedIdx(const T* in_dat, const int64_t in_len, + const int* tokens, const size_t tokens_len, + size_t* num_erased) { int index = blockIdx.x * blockDim.x + threadIdx.x; if (index < in_len) { - int erased = 0; - for (int i = 0; i < tokens_len; ++i) { + for (size_t i = 0; i < tokens_len; ++i) { if (in_dat[index] == tokens[i]) { - erased = 1; + num_erased[index + 1] = 1; + break; } } - num_erased[index + 1] = erased; - if (index == 0) { - num_erased[0] = 0; - } } } -template -__global__ void GetOutLod(const T* num_erased, const int* in_lod, - const int lod_len, int* out_lod0) { +__global__ void GetOutLod(const size_t* num_erased, const size_t* in_lod, + const size_t lod_len, size_t* out_lod0) { int index = blockIdx.x * blockDim.x + threadIdx.x; if (index < lod_len) { out_lod0[index] = in_lod[index] - num_erased[in_lod[index]]; @@ -51,11 +46,11 @@ __global__ void GetOutLod(const T* num_erased, const int* in_lod, } template -__global__ void SetOutput(const T* in_dat, const int in_len, - const int* num_erased, T* out_dat) { +__global__ void SetOutput(const T* in_dat, const int64_t in_len, + const size_t* num_erased, T* out_dat) { int index = blockIdx.x * blockDim.x + threadIdx.x; if (index < in_len) { - if (in_dat[index] != in_dat[index + 1]) { + if (num_erased[index] == num_erased[index + 1]) { out_dat[index - num_erased[index]] = in_dat[index]; } } @@ -72,53 +67,43 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel { PADDLE_ENFORCE_EQ(lod.size(), 1UL, "Only support one level sequence now."); PADDLE_ENFORCE_EQ(lod[0].back(), (size_t)in->numel(), "The actual size mismatches with the LoD information."); - auto tokens = ctx.Attr>("tokens"); - auto tokens_len = tokens.size(); + auto tokens = ctx.Attr>("tokens"); auto in_len = in->numel(); auto in_dat = in->data(); - auto lod0 = lod[0]; - - thrust::host_vector host_tokens(tokens_len); - for (size_t i = 0; i < tokens.size(); ++i) { - host_tokens[i] = tokens[i]; - } - thrust::device_vector dev_tokens = host_tokens; - thrust::device_vector num_erased(in_len + 1); - - T* dev_tokens_ptr = thrust::raw_pointer_cast(dev_tokens.data()); - int* num_erased_ptr = thrust::raw_pointer_cast(num_erased.data()); + // Copy tokens to GPU + thrust::device_vector dev_tokens(tokens.begin(), tokens.end()); + int* dev_tokens_ptr = thrust::raw_pointer_cast(dev_tokens.data()); + // Count number of elements to be erased + thrust::device_vector num_erased(in_len + 1, 0); + size_t* num_erased_ptr = thrust::raw_pointer_cast(num_erased.data()); auto stream = ctx.cuda_device_context().stream(); LabelErasedIdx<<<(in_len - 1) / PADDLE_CUDA_NUM_THREADS + 1, PADDLE_CUDA_NUM_THREADS, 0, stream>>>( - in_dat, in_len, dev_tokens_ptr, tokens_len, num_erased_ptr); + in_dat, in_len, dev_tokens_ptr, tokens.size(), num_erased_ptr); thrust::inclusive_scan(num_erased.begin() + 1, num_erased.end(), num_erased.begin() + 1); - // Calc LoD + // Copy LoD to GPU + auto lod0 = lod[0]; auto lod_len = lod0.size(); - thrust::host_vector host_lod(lod_len); - for (size_t i = 0; i < lod_len; ++i) { - host_lod[i] = lod0[i]; - } - thrust::device_vector dev_in_lod = host_lod; - thrust::device_vector dev_out_lod(lod_len); - int* dev_in_lod_ptr = thrust::raw_pointer_cast(dev_in_lod.data()); - int* dev_out_lod_ptr = thrust::raw_pointer_cast(dev_out_lod.data()); + thrust::device_vector dev_in_lod = lod0; + size_t* dev_in_lod_ptr = thrust::raw_pointer_cast(dev_in_lod.data()); + + // Calc output LoD + thrust::device_vector dev_out_lod(lod_len); + size_t* dev_out_lod_ptr = thrust::raw_pointer_cast(dev_out_lod.data()); GetOutLod<<<(lod_len - 1) / PADDLE_CUDA_NUM_THREADS + 1, PADDLE_CUDA_NUM_THREADS, 0, stream>>>( num_erased_ptr, dev_in_lod_ptr, lod_len, dev_out_lod_ptr); - thrust::host_vector host_out_lod = dev_out_lod; - std::vector out_lod0(lod_len, 0); - for (size_t i = 0; i < lod_len; i++) { - out_lod0[i] = host_out_lod[i]; - } + // Set LoD for output + std::vector out_lod0(dev_out_lod.begin(), dev_out_lod.end()); framework::LoD out_lod; out_lod.push_back(out_lod0); out->set_lod(out_lod); // Set output - out->Resize({out_lod0.back(), 1}); + out->Resize({static_cast(out_lod0.back()), 1}); auto out_dat = out->mutable_data(ctx.GetPlace()); SetOutput<<<(in_len - 1) / PADDLE_CUDA_NUM_THREADS + 1, PADDLE_CUDA_NUM_THREADS, 0, stream>>>(in_dat, in_len, @@ -130,4 +115,5 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel { } // namespace paddle REGISTER_OP_CUDA_KERNEL(sequence_erase, - paddle::operators::SequenceEraseOpCUDAKernel); + paddle::operators::SequenceEraseOpCUDAKernel, + paddle::operators::SequenceEraseOpCUDAKernel); diff --git a/paddle/operators/sequence_expand_op.cc b/paddle/operators/sequence_expand_op.cc index b40ec617e42110e0ab5168a8ac675adaf760fb3c..d34dbd35b6df2dac275fbe2c41f99b8549217d5b 100644 --- a/paddle/operators/sequence_expand_op.cc +++ b/paddle/operators/sequence_expand_op.cc @@ -58,7 +58,7 @@ This operator expands input(X) according to LOD of input(Y). Following are cases to better explain how this works: Case 1: -Given 2-level a LoDTensor input(X) +Given a 2-level LoDTensor input(X) X.lod = [[0, 2, 3], [0, 1, 3, 4]] X.data = [a, b, c, d] @@ -75,9 +75,8 @@ then we get 2-level LoDTensor Case 2: -Given a 0-level LoDTensor input(X) +Given a common Tensor input(X) X.data = [a, b, c] - X.lod = NULL X.dims = [3, 1] and input(Y) Y.lod = [[0, 2, 3, 6]] @@ -89,9 +88,8 @@ then we get 1-level LoDTensor Case 3: -Given a 0-level LoDTensor input(X) +Given a common Tensor input(X) X.data = [[a, b], [c, d], [e, f]] - X.lod = NULL X.dims = [3, 2] and input(Y) Y.lod = [[0, 2, 3, 6]] diff --git a/paddle/operators/sequence_expand_op.h b/paddle/operators/sequence_expand_op.h index 2ba628e9c37278025e31779ab0468db46f2ff40a..6021526eee8e0a1f58885f6de38b14048787a828 100644 --- a/paddle/operators/sequence_expand_op.h +++ b/paddle/operators/sequence_expand_op.h @@ -32,6 +32,7 @@ class SequenceExpandKernel : public framework::OpKernel { const T* x_data = x->data(); auto x_dims = x->dims(); auto* y = context.Input("Y"); + PADDLE_ENFORCE(!y->lod().empty(), "y should have lod"); PADDLE_ENFORCE_EQ(static_cast(x_dims[0]), y->lod().back().size() - 1, "The size of last lod level in Input(Y)" diff --git a/paddle/operators/sequence_reshape_op.cc b/paddle/operators/sequence_reshape_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..d89a46a712c9c84a142e1e347219ed171556d761 --- /dev/null +++ b/paddle/operators/sequence_reshape_op.cc @@ -0,0 +1,135 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "paddle/operators/sequence_reshape_op.h" +#include "paddle/framework/ddim.h" + +namespace paddle { +namespace operators { + +class SequenceReshapeOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of SequenceReshapeOp should not be null."); + PADDLE_ENFORCE(ctx->HasOutput("Out"), + "Output(Out) of SequenceReshapeOp should not be null."); + auto x_dims = ctx->GetInputDim("X"); + auto x_numel = product(x_dims); + PADDLE_ENFORCE_EQ(x_dims.size(), 2U, "Rank of Input(X) should be 2."); + int new_dim = ctx->Attrs().Get("new_dim"); + if (ctx->IsRuntime()) { + ctx->SetOutputDim("Out", + {x_numel / new_dim, static_cast(new_dim)}); + } else { + // when compiling, the batch size is undetermined, just set to -1 + ctx->SetOutputDim("Out", {-1, static_cast(new_dim)}); + } + } +}; + +class SequenceReshapeOpMaker : public framework::OpProtoAndCheckerMaker { + public: + SequenceReshapeOpMaker(OpProto* proto, OpAttrChecker* op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", + "(LoDTensor, default LoDTensor) A 2-D LoDTensor with shape " + "being [N, M]."); + AddOutput("Out", + "(LoDTensor, default LoDTensor) A 2-D LoDTensor with " + "shape [T, new_dim] where T is calculated based on X.lod, M and " + "new_dim."); + AddAttr("new_dim", "Sequence dimension of the output LoDTensor."); + AddComment(R"DOC( +Sequence Reshape Operator. + +This operator will rearrange the input sequences. The new dimension is set by +attribute and length of each sequence may change longer or shorter which is +decided by original length, original dimension and new dimension. The following +example will help to illustrate the function of this operator: + +x is a LoDTensor: + x.lod = [[0, 2, 6]] + x.data = [[1, 2], [3, 4], + [5, 6], [7, 8], [9, 10], [11, 12]] + x.dims = [6, 2] + +set new_dim = 4 + +then out is a LoDTensor: + out.lod = [[0, 1, 3]] + out.data = [[1, 2, 3, 4], + [5, 6, 7, 8], [9, 10, 11, 12]] + out.dims = [3, 4] + +Currently, only 1-level LoDTensor is supported and please make sure (original +length * original dimension) can be divided by new_dim with no remainder for +each sequence. + +)DOC"); + } +}; + +class SequenceReshapeGradOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext* ctx) const override { + PADDLE_ENFORCE( + ctx->HasInput(framework::GradVarName("Out")), + "Input(Out@GRAD) of SequenceReshapeGradOp should not be null."); + PADDLE_ENFORCE(ctx->HasInput("X"), + "Input(X) of SequenceReshapeGradOp should not be null."); + + ctx->SetOutputDim(framework::GradVarName("X"), ctx->GetInputDim("X")); + ctx->ShareLoD("X", /*->*/ framework::GradVarName("X")); + } +}; + +class SequenceReshapeGradOpMaker : public framework::SingleGradOpDescMaker { + public: + using framework::SingleGradOpDescMaker::SingleGradOpDescMaker; + + protected: + std::unique_ptr Apply() const override { + auto* op_desc_ptr = new framework::OpDesc(); + op_desc_ptr->SetType("sequence_reshape_grad"); + op_desc_ptr->SetInput("X", Input("X")); + op_desc_ptr->SetInput(framework::GradVarName("Out"), OutputGrad("Out")); + op_desc_ptr->SetOutput(framework::GradVarName("X"), InputGrad("X")); + op_desc_ptr->SetAttrMap(Attrs()); + return std::unique_ptr(op_desc_ptr); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OPERATOR(sequence_reshape, ops::SequenceReshapeOp, + ops::SequenceReshapeOpMaker, ops::SequenceReshapeGradOpMaker); +REGISTER_OPERATOR(sequence_reshape_grad, ops::SequenceReshapeGradOp); +REGISTER_OP_CPU_KERNEL( + sequence_reshape, + ops::SequenceReshapeKernel, + ops::SequenceReshapeKernel, + ops::SequenceReshapeKernel, + ops::SequenceReshapeKernel); +REGISTER_OP_CPU_KERNEL( + sequence_reshape_grad, + ops::SequenceReshapeGradKernel, + ops::SequenceReshapeGradKernel, + ops::SequenceReshapeGradKernel, + ops::SequenceReshapeGradKernel); diff --git a/paddle/operators/sequence_reshape_op.cu b/paddle/operators/sequence_reshape_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..d9c2f7e9a4149371867cf2a8b81d58566999bfba --- /dev/null +++ b/paddle/operators/sequence_reshape_op.cu @@ -0,0 +1,30 @@ +/* Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/sequence_reshape_op.h" + +namespace ops = paddle::operators; +REGISTER_OP_CUDA_KERNEL( + sequence_reshape, + ops::SequenceReshapeKernel, + ops::SequenceReshapeKernel, + ops::SequenceReshapeKernel, + ops::SequenceReshapeKernel); +REGISTER_OP_CUDA_KERNEL( + sequence_reshape_grad, + ops::SequenceReshapeGradKernel, + ops::SequenceReshapeGradKernel, + ops::SequenceReshapeGradKernel, + ops::SequenceReshapeGradKernel); diff --git a/paddle/operators/sequence_reshape_op.h b/paddle/operators/sequence_reshape_op.h new file mode 100644 index 0000000000000000000000000000000000000000..aaae7ab29281b72848515b80cc60931c13a294c9 --- /dev/null +++ b/paddle/operators/sequence_reshape_op.h @@ -0,0 +1,86 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#pragma once +#include "paddle/framework/op_registry.h" +#include "paddle/operators/math/math_function.h" + +namespace paddle { +namespace operators { + +using LoDTensor = framework::LoDTensor; +template +class SequenceReshapeKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& context) const override { + auto* in = context.Input("X"); + auto* out = context.Output("Out"); + int out_width = context.Attr("new_dim"); + + auto in_dims = in->dims(); + int64_t in_width = in_dims[1]; + auto& in_lod = in->lod(); + + PADDLE_ENFORCE_EQ(in_lod.size(), 1UL, + "Only support one level sequence now."); + PADDLE_ENFORCE_EQ( + (uint64_t)in_dims[0], in_lod[0].back(), + "Inconsistent size between X.shape[0] and X.lod()[0].back()."); + + auto in_lod_l0 = in_lod[0]; + int seq_num = in_lod_l0.size() - 1; + + if (in_width == out_width) { + out->set_lod(in->lod()); + } else { + auto& out_lod = *out->mutable_lod(); + out_lod.resize(1); + out_lod[0].resize(seq_num + 1); + out_lod[0][0] = 0; + for (int i = 0; i < seq_num; ++i) { + size_t seq_len = in_lod_l0[i + 1] - in_lod_l0[i]; + size_t offset = 0; + offset = (seq_len * in_width) / out_width; + PADDLE_ENFORCE_EQ(offset * out_width, seq_len * in_width, + "Please make sure (sequence_length * dimension) can " + "be divided by new_dim with no remainder for each " + "sequence. The %dth sequence is invalid.", + i + 1); + out_lod[0][i + 1] = out_lod[0][i] + offset; + } + } + + framework::Copy(*in, context.GetPlace(), out); + out->Resize({static_cast(out->lod()[0].back()), out_width}); + } +}; + +template +class SequenceReshapeGradKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& context) const override { + auto* x_tensor_ptr = context.Input("X"); + auto* outg_tensor_ptr = + context.Input(framework::GradVarName("Out")); + auto* xg_tensor_ptr = + context.Output(framework::GradVarName("X")); + + xg_tensor_ptr->mutable_data(context.GetPlace()); + framework::Copy(*outg_tensor_ptr, context.GetPlace(), xg_tensor_ptr); + xg_tensor_ptr->Resize(x_tensor_ptr->dims()); + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/sgd_op.cu b/paddle/operators/sgd_op.cu index 42f8f8b2f072f9d204dfadcd732926b5c98dc617..29f5aa3542c26c76a1b80da61ec6752019216131 100644 --- a/paddle/operators/sgd_op.cu +++ b/paddle/operators/sgd_op.cu @@ -89,7 +89,7 @@ class SGDOpCUDAKernel : public framework::OpKernel { PADDLE_ENFORCE_EQ(in_height, out_dims[0]); auto& in_value = grad->value(); - auto& in_rows = grad->rows(); + framework::Vector in_rows(grad->rows()); int64_t in_row_numel = in_value.numel() / in_rows.size(); PADDLE_ENFORCE_EQ(in_row_numel, param_out->numel() / in_height); @@ -102,7 +102,7 @@ class SGDOpCUDAKernel : public framework::OpKernel { dim3 grid(1, in_rows.size()); SparseSGDFunctorKernel< T, 256><<>>( - in_data, in_rows.data(), learning_rate->data(), out_data, + in_data, in_rows.cuda_data(), learning_rate->data(), out_data, in_row_numel); } else { diff --git a/paddle/operators/split_selected_rows_op.cc b/paddle/operators/split_selected_rows_op.cc new file mode 100644 index 0000000000000000000000000000000000000000..0515ea13aadbd6e12e2586c937f3d1ed0a298d69 --- /dev/null +++ b/paddle/operators/split_selected_rows_op.cc @@ -0,0 +1,107 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/split_selected_rows_op.h" + +namespace paddle { +namespace operators { + +class SplitSelectedRowsOpMaker : public framework::OpProtoAndCheckerMaker { + public: + SplitSelectedRowsOpMaker(OpProto *proto, OpAttrChecker *op_checker) + : OpProtoAndCheckerMaker(proto, op_checker) { + AddInput("X", "The input SelectedRows."); + AddOutput("Out", "The outputs of input SelectedRows.").AsDuplicable(); + AddAttr>("height_sections", + "Height for each output SelectedRows.") + .SetDefault(std::vector({})); + + AddComment(R"DOC( +Split a SelectedRows with a specified rows section. +height_sections is only needed when need to split the dims of the original tensor. + +Example: + Input: + X.rows = {7, 5} + X.height = 12 + Attr: + height_sections = {4, 8} + Out: + out0.rows = {} + out0.height = 4 + + out1.rows = {5, 7} + out2.height = 8 + +)DOC"); + } +}; + +class SplitSelectedRowsOp : public framework::OperatorWithKernel { + public: + using framework::OperatorWithKernel::OperatorWithKernel; + + void InferShape(framework::InferShapeContext *ctx) const override { + PADDLE_ENFORCE(ctx->HasInput("X"), "SplitSelectedRowsOp must has input X."); + PADDLE_ENFORCE(ctx->HasOutputs("Out"), + "SplitSelectedRowsOp must has output Out."); + + std::vector height_sections = + ctx->Attrs().Get>("height_sections"); + int64_t n = ctx->Outputs("Out").size(); + + std::vector outs_dims; + outs_dims.reserve(n); + + // make output dims + for (int64_t i = 0; i < n; ++i) { + auto dims = ctx->GetInputDim("X"); + if (height_sections.size()) { + PADDLE_ENFORCE_EQ( + height_sections.size(), static_cast(n), + "The size of height section should be the same with height" + " section size."); + dims[0] = height_sections[i]; + } + outs_dims.push_back(dims); + } + ctx->SetOutputsDim("Out", outs_dims); + } +}; + +class SplitSelectedRowsGradMaker : public framework::SingleGradOpDescMaker { + public: + using framework::SingleGradOpDescMaker::SingleGradOpDescMaker; + + protected: + std::unique_ptr Apply() const override { + auto *grad_op = new framework::OpDesc(); + grad_op->SetType("sum"); + grad_op->SetInput("X", OutputGrad("Out")); + grad_op->SetOutput("Out", InputGrad("X")); + grad_op->SetAttrMap(Attrs()); + return std::unique_ptr(grad_op); + } +}; + +} // namespace operators +} // namespace paddle + +namespace ops = paddle::operators; +REGISTER_OPERATOR(split_selected_rows, ops::SplitSelectedRowsOp, + ops::SplitSelectedRowsOpMaker, + ops::SplitSelectedRowsGradMaker); +REGISTER_OP_CPU_KERNEL( + split_selected_rows, + ops::SplitSelectedRowsOpKernel); diff --git a/paddle/operators/split_selected_rows_op.cu b/paddle/operators/split_selected_rows_op.cu new file mode 100644 index 0000000000000000000000000000000000000000..983285480fd9de7a2a4d2787a9bba72c160b7fae --- /dev/null +++ b/paddle/operators/split_selected_rows_op.cu @@ -0,0 +1,19 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#include "paddle/operators/split_selected_rows_op.h" +namespace ops = paddle::operators; +REGISTER_OP_CUDA_KERNEL( + split_selected_rows, + ops::SplitSelectedRowsOpKernel); diff --git a/paddle/operators/split_selected_rows_op.h b/paddle/operators/split_selected_rows_op.h new file mode 100644 index 0000000000000000000000000000000000000000..12e64e2901e5902d187d65a12c94b8d7ef45a481 --- /dev/null +++ b/paddle/operators/split_selected_rows_op.h @@ -0,0 +1,88 @@ +/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + +http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. */ + +#pragma once + +#include +#include "paddle/framework/op_registry.h" +#include "paddle/operators/math/selected_rows_functor.h" + +namespace paddle { +namespace operators { + +static int FindOutIdx(int row, const std::vector& height_sections) { + int offset = 0; + for (size_t i = 0; i < height_sections.size(); ++i) { + if (row >= offset && row < (offset + height_sections[i])) { + return i; + } + offset += height_sections[i]; + } + return -1; +} + +template +class SplitSelectedRowsOpKernel : public framework::OpKernel { + public: + void Compute(const framework::ExecutionContext& ctx) const override { + auto* x = ctx.Input("X"); + auto outs = ctx.MultiOutput("Out"); + auto height_sections = ctx.Attr>("height_sections"); + + auto x_rows = x->rows(); + std::vector> outs_rows_idx; + outs_rows_idx.resize(outs.size()); + + auto row_numel = x->value().numel() / x->value().dims()[0]; + auto src = x->value().data(); + + for (size_t i = 0; i < x_rows.size(); ++i) { + int out_idx = FindOutIdx(x_rows[i], height_sections); + outs_rows_idx[out_idx].push_back(i); + } + auto place = ctx.GetPlace(); + + for (size_t i = 0; i < outs_rows_idx.size(); ++i) { + auto rows_idx = outs_rows_idx[i]; + if (rows_idx.size() > 0) { + auto dims = x->GetCompleteDims(); + dims[0] = rows_idx.size(); + outs[i]->mutable_value()->mutable_data(dims, x->place()); + for (auto idx : rows_idx) { + outs[i]->mutable_rows()->push_back(x_rows[idx]); + } + auto dst = outs[i]->mutable_value()->mutable_data(ctx.GetPlace()); + for (size_t j = 0; j < rows_idx.size(); j++) { + if (platform::is_cpu_place(place)) { + memory::Copy(platform::CPUPlace(), dst + j * row_numel, + platform::CPUPlace(), src + rows_idx[j] * row_numel, + sizeof(T) * row_numel); + } else { +#ifdef PADDLE_WITH_CUDA + auto stream = ctx.cuda_device_context().stream(); + memory::Copy(platform::CUDAPlace(), dst + j * row_numel, + platform::CUDAPlace(), src + rows_idx[j] * row_numel, + sizeof(T) * row_numel, stream); +#else + PADDLE_THROW("Paddle is not compiled with GPU"); +#endif + } + } + } + } + } +}; + +} // namespace operators +} // namespace paddle diff --git a/paddle/operators/sum_op.h b/paddle/operators/sum_op.h index 48201b344de0d3bd2b121a12389876dad095f10d..3d8102c3ae20c8b714cd48b4fc78dc18a0cf89a7 100644 --- a/paddle/operators/sum_op.h +++ b/paddle/operators/sum_op.h @@ -68,7 +68,32 @@ class SumKernel : public framework::OpKernel { } } } else if (out_var->IsType()) { - PADDLE_ENFORCE(!in_place, "SelectedRows not support inplace sum now"); + std::unique_ptr in0; + if (in_place) { + // If is in_place, we store the input[0] to in0 + auto &in_sel0 = in_vars[0]->Get(); + auto &rows = in_sel0.rows(); +#ifdef PADDLE_WITH_CUDA + std::vector rows_in_cpu; + rows_in_cpu.reserve(rows.size()); + for (auto item : rows) { + rows_in_cpu.push_back(item); + } + in0.reset(new framework::SelectedRows(rows_in_cpu, in_sel0.height())); +#else + in0.reset(new framework::SelectedRows(rows, in_sel0.height())); +#endif + in0->mutable_value()->ShareDataWith(in_sel0.value()); + } + + auto get_selected_row = [&](size_t i) -> const SelectedRows & { + if (i == 0 && in0) { + return *in0.get(); + } else { + return in_vars[i]->Get(); + } + }; + auto *out = context.Output("Out"); out->mutable_rows()->clear(); auto *out_value = out->mutable_value(); @@ -76,24 +101,26 @@ class SumKernel : public framework::OpKernel { // Runtime InferShape size_t first_dim = 0; for (int i = 0; i < N; i++) { - first_dim += in_vars[i]->Get().rows().size(); + auto &sel_row = get_selected_row(i); + first_dim += sel_row.rows().size(); } - auto in_dim = in_vars[0]->Get().value().dims(); - auto in_dim_vec = framework::vectorize(in_dim); - in_dim_vec[0] = static_cast(first_dim); + auto in_dim = + framework::vectorize(get_selected_row(N - 1).value().dims()); + in_dim[0] = static_cast(first_dim); - out_value->Resize(framework::make_ddim(in_dim_vec)); + out_value->Resize(framework::make_ddim(in_dim)); out_value->mutable_data(context.GetPlace()); math::SelectedRowsAddTo functor; int64_t offset = 0; for (int i = 0; i < N; i++) { - PADDLE_ENFORCE_EQ(out->height(), - in_vars[i]->Get().height()); - functor(context.template device_context(), - in_vars[i]->Get(), offset, out); - offset += in_vars[i]->Get().value().numel(); + auto &sel_row = get_selected_row(i); + + PADDLE_ENFORCE_EQ(out->height(), sel_row.height()); + functor(context.template device_context(), sel_row, + offset, out); + offset += sel_row.value().numel(); } } else if (out_var->IsType()) { auto &out_array = *out_var->GetMutable(); diff --git a/paddle/operators/top_k_op.h b/paddle/operators/top_k_op.h index e9cd9bbd4d964c28f305fb4ab4c4733ed27ebfff..bf42e15e6b234125d9ec24e8500367b9915213ab 100644 --- a/paddle/operators/top_k_op.h +++ b/paddle/operators/top_k_op.h @@ -22,6 +22,7 @@ namespace paddle { namespace operators { using Tensor = framework::Tensor; +using LoDTensor = framework::LoDTensor; template @@ -33,9 +34,9 @@ class TopkKernel : public framework::OpKernel { void Compute(const framework::ExecutionContext& ctx) const override { // Get the top k elements of each row of input tensor // FIXME: only deal with matrix(2d tensor). - auto* input = ctx.Input("X"); - auto* output = ctx.Output("Out"); - auto* indices = ctx.Output("Indices"); + auto* input = ctx.Input("X"); + auto* output = ctx.Output("Out"); + auto* indices = ctx.Output("Indices"); // k is determined by Attr const size_t k = static_cast(ctx.Attr("k")); diff --git a/paddle/operators/transpose_op.cc b/paddle/operators/transpose_op.cc index 11615d806a61b3525d2ed50f5ea5940e8d61c8f8..c7ae162638ca5e929cca14c841cc3eceeea5f64e 100644 --- a/paddle/operators/transpose_op.cc +++ b/paddle/operators/transpose_op.cc @@ -59,44 +59,39 @@ class TransposeOpMaker : public framework::OpProtoAndCheckerMaker { : OpProtoAndCheckerMaker(proto, op_checker) { AddInput( "X", - "(Tensor)The input tensor, tensors with rank at most 6 are supported"); - AddOutput("Out", "(Tensor)The output tensor"); + "(Tensor) The input tensor, tensors with rank up to 6 are supported."); + AddOutput("Out", "(Tensor)The output tensor."); AddAttr>( "axis", - "(vector)A list of values, and the size of the list should be " - "the same with the input tensor rank, the tensor will " - "permute the axes according the the values given"); + "(vector) A list of values, and the size of the list should be " + "the same with the input tensor rank. This operator permutes the input " + "tensor's axes according to the values given."); AddComment(R"DOC( Transpose Operator. -The input tensor will be permuted according to the axis values given. -The op functions is similar to how numpy.transpose works in python. +The input tensor will be permuted according to the axes given. +The behavior of this operator is similar to how `numpy.transpose` works. -For example: +- suppose the input `X` is a 2-D tensor: + $$ + X = \begin{pmatrix} + 0 &1 &2 \\ + 3 &4 &5 + \end{pmatrix}$$ - .. code-block:: text + the given `axes` is: $[1, 0]$, and $Y$ = transpose($X$, axis) - input = numpy.arange(6).reshape((2,3)) + then the output $Y$ is: - the input is: + $$ + Y = \begin{pmatrix} + 0 &3 \\ + 1 &4 \\ + 2 &5 + \end{pmatrix}$$ - array([[0, 1, 2], - [3, 4, 5]]) - - given axis is: - - [1, 0] - - output = input.transpose(axis) - - then the output is: - - array([[0, 3], - [1, 4], - [2, 5]]) - -So, given a input tensor of shape(N, C, H, W) and the axis is {0, 2, 3, 1}, -the output tensor shape will be (N, H, W, C) +- Given a input tensor with shape $(N, C, H, W)$ and the `axes` is +$[0, 2, 3, 1]$, then shape of the output tensor will be: $(N, H, W, C)$. )DOC"); } diff --git a/paddle/optimizer/parameter_optimizer_test.cc b/paddle/optimizer/parameter_optimizer_test.cc index 795d2de1d65b3a25c312d0c4e31e0105922838df..2bcfca55cc5c9720f16e715dd85bf15c6f2efbbe 100644 --- a/paddle/optimizer/parameter_optimizer_test.cc +++ b/paddle/optimizer/parameter_optimizer_test.cc @@ -11,21 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "parameter_optimizer.h" #include diff --git a/paddle/optimizer/serialization.h b/paddle/optimizer/serialization.h index 92fbf65cc6b98d7f92841bafe4ab77001ca03b7c..98548ddb7aa22558c83aff7454834caa89017388 100644 --- a/paddle/optimizer/serialization.h +++ b/paddle/optimizer/serialization.h @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #pragma once #include diff --git a/paddle/optimizer/serialization_test.cc b/paddle/optimizer/serialization_test.cc index 0f1b14eec135ab37a599965965e2e9d8bb65b90c..25a8f5d351e3e85ab5d8dee8b639d962f8fc9990 100644 --- a/paddle/optimizer/serialization_test.cc +++ b/paddle/optimizer/serialization_test.cc @@ -11,21 +11,6 @@ // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. // See the License for the specific language governing permissions and // limitations under the License. -/* - Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve. - - Licensed under the Apache License, Version 2.0 (the "License"); - you may not use this file except in compliance with the License. - You may obtain a copy of the License at - - http://www.apache.org/licenses/LICENSE-2.0 - - Unless required by applicable law or agreed to in writing, software - distributed under the License is distributed on an "AS IS" BASIS, - WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - See the License for the specific language governing permissions and - limitations under the License. -*/ #include "serialization.h" #include "gtest/gtest.h" diff --git a/paddle/platform/CMakeLists.txt b/paddle/platform/CMakeLists.txt index 44f6d85cd1510f309595ca711de2e0f767219580..d68caea99719b37816391f9bddcc5cac051025b2 100644 --- a/paddle/platform/CMakeLists.txt +++ b/paddle/platform/CMakeLists.txt @@ -10,7 +10,7 @@ cc_test(cpu_info_test SRCS cpu_info_test.cc DEPS cpu_info) nv_library(gpu_info SRCS gpu_info.cc DEPS gflags glog enforce) -cc_library(place SRCS place.cc DEPS enforce) +cc_library(place SRCS place.cc DEPS enforce boost) cc_test(place_test SRCS place_test.cc DEPS place glog gflags) add_subdirectory(dynload) @@ -39,3 +39,11 @@ nv_test(nccl_test SRCS nccl_test.cu DEPS dynload_cuda gpu_info device_context) cc_library(profiler SRCS profiler.cc DEPS device_context) cc_test(profiler_test SRCS profiler_test.cc DEPS profiler) + +if(NOT WITH_C_API AND WITH_FLUID) + file(GLOB PLATFORM_HEADERS *.h) + file(GLOB PLATFORM_dynload_HEADERS dynload/*.h) + install(FILES ${PLATFORM_HEADERS} DESTINATION include/paddle/platform) + install(FILES ${PLATFORM_HEADERS} DESTINATION include/paddle/platform/dynload) + install(FILES details/device_ptr_cast.h DESTINATION include/paddle/platform/details) +endif() diff --git a/paddle/platform/assert.h b/paddle/platform/assert.h index 70d3bf75062c5471ab54ee2c4c7637c679d9a8a3..d813b9529ba2c8d5a3f39eadeb82d7569acd5fdd 100644 --- a/paddle/platform/assert.h +++ b/paddle/platform/assert.h @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + #pragma once #define STRINGIFY(x) #x diff --git a/paddle/platform/call_once.h b/paddle/platform/call_once.h index 00337a7f051758559a0f8012d8c78dbe8e3457a6..44a4d38f679ddf6c317e52132b6cf3eb2f0a0649 100644 --- a/paddle/platform/call_once.h +++ b/paddle/platform/call_once.h @@ -29,20 +29,25 @@ namespace platform { */ template inline void call_once(std::once_flag& flag, Callable&& f, Args&&... args) { - bool good = false; + bool good = true; std::exception ex; - std::call_once(flag, - [&](Args&&... args) { - try { - f(args...); - good = true; - } catch (const std::exception& e) { - ex = e; - } catch (...) { - ex = std::runtime_error("excption caught in call_once"); - } - }, - args...); + try { + std::call_once(flag, + [&](Args&&... args) { + try { + f(args...); + } catch (const std::exception& e) { + ex = e; + good = false; + } catch (...) { + ex = std::runtime_error("excption caught in call_once"); + good = false; + } + }, + args...); + } catch (std::system_error& x) { + throw std::runtime_error("call once failed"); + } if (!good) { throw std::exception(ex); } diff --git a/paddle/platform/profiler.cc b/paddle/platform/profiler.cc index 7e2e2d968ef877f6aa8b87ab8f044e89574dffa9..2a8afc940393baaaa939471f50f2d5c63edd6a84 100644 --- a/paddle/platform/profiler.cc +++ b/paddle/platform/profiler.cc @@ -47,16 +47,16 @@ inline uint64_t GetTimeInNsec() { } Event::Event(EventKind kind, std::string name, uint32_t thread_id, - DeviceContext* dev_ctx) + const DeviceContext* dev_ctx) : kind_(kind), name_(name), thread_id_(thread_id), has_cuda_(false) { #ifdef PADDLE_WITH_CUDA - auto* cuda_dev_ctx = static_cast(dev_ctx); - if (cuda_dev_ctx) { + has_cuda_ = dev_ctx ? platform::is_gpu_place(dev_ctx->GetPlace()) : false; + if (has_cuda_) { + auto* cuda_dev_ctx = static_cast(dev_ctx); PADDLE_ENFORCE(cudaGetDevice(&device_)); PADDLE_ENFORCE(cudaEventCreate(&event_)); auto stream = cuda_dev_ctx->stream(); PADDLE_ENFORCE(cudaEventRecord(event_, stream)); - has_cuda_ = true; } #endif cpu_ns_ = GetTimeInNsec(); @@ -114,19 +114,20 @@ inline EventList& GetEventList() { return *g_event_list; } -void Mark(const std::string& name, DeviceContext* dev_ctx) { +void Mark(const std::string& name, const DeviceContext* dev_ctx) { GetEventList().Record(EventKind::kMark, name, g_thread_id, dev_ctx); } -void PushEvent(const std::string& name, DeviceContext* dev_ctx) { +void PushEvent(const std::string& name, const DeviceContext* dev_ctx) { GetEventList().Record(EventKind::kPushRange, name, g_thread_id, dev_ctx); } -void PopEvent(const std::string& name, DeviceContext* dev_ctx) { +void PopEvent(const std::string& name, const DeviceContext* dev_ctx) { GetEventList().Record(EventKind::kPopRange, name, g_thread_id, dev_ctx); } -RecordEvent::RecordEvent(const std::string& name, DeviceContext* dev_ctx) { +RecordEvent::RecordEvent(const std::string& name, + const DeviceContext* dev_ctx) { if (g_state == ProfilerState::kDisabled) return; dev_ctx_ = dev_ctx; name_ = name; @@ -155,6 +156,7 @@ void EnableProfiler(ProfilerState state) { DeviceContext* dev_ctx = new CUDADeviceContext(CUDAPlace(d)); Mark("_cuda_startup_", dev_ctx); dev_ctx->Wait(); + delete dev_ctx; }); } } @@ -163,14 +165,17 @@ void EnableProfiler(ProfilerState state) { Mark("_start_profiler_", nullptr); } -std::vector> DisableProfiler() { - PADDLE_ENFORCE(g_state != ProfilerState::kDisabled, - "Can't disable profiling, since it's not starting."); - // Mark the profiling stop. - Mark("_stop_profiler_", nullptr); - g_state = ProfilerState::kDisabled; - std::vector> result; +void ResetProfiler() { std::lock_guard guard(g_all_event_lists_mutex); + for (auto it = g_all_event_lists.begin(); it != g_all_event_lists.end(); + ++it) { + (*it)->Clear(); + } +} + +std::vector> GetAllEvents() { + std::lock_guard guard(g_all_event_lists_mutex); + std::vector> result; for (auto it = g_all_event_lists.begin(); it != g_all_event_lists.end(); ++it) { result.emplace_back((*it)->Reduce()); @@ -178,6 +183,18 @@ std::vector> DisableProfiler() { return result; } +void DisableProfiler(EventSortingKey sorted_key) { + PADDLE_ENFORCE(g_state != ProfilerState::kDisabled, + "Can't disable profiling, since it's not starting."); + // Mark the profiling stop. + Mark("_stop_profiler_", nullptr); + g_state = ProfilerState::kDisabled; + + std::vector> all_events = GetAllEvents(); + ParseEvents(all_events, sorted_key); + ResetProfiler(); +} + void ParseEvents(std::vector>& events, EventSortingKey sorted_by) { if (g_profiler_place == "") return; @@ -291,12 +308,12 @@ void ParseEvents(std::vector>& events, } // Print report - PrintProfilingReport(events_table, sorted_domain, max_name_width + 4, 12); + PrintProfiler(events_table, sorted_domain, max_name_width + 4, 12); } -void PrintProfilingReport(std::vector>& events_table, - std::string& sorted_domain, const size_t name_width, - const size_t data_width) { +void PrintProfiler(std::vector>& events_table, + std::string& sorted_domain, const size_t name_width, + const size_t data_width) { // Output header information std::cout << "\n------------------------->" << " Profiling Report " diff --git a/paddle/platform/profiler.h b/paddle/platform/profiler.h index 6df48ef8806e865f473b4317ac0283863c3c6f64..8de1e6ad296d1e15c1659ccf431f1d5013eb608c 100644 --- a/paddle/platform/profiler.h +++ b/paddle/platform/profiler.h @@ -29,7 +29,7 @@ class Event { // The DeviceContext is used to get the cuda stream. // If CPU profiling mode, can pass nullptr. Event(EventKind kind, std::string name, uint32_t thread_id, - DeviceContext* dev_ctx); + const DeviceContext* dev_ctx); std::string kind() const; std::string name() const { return name_; } @@ -84,6 +84,8 @@ struct EventList { return result; } + void Clear() { event_blocks.clear(); } + std::forward_list> event_blocks; }; @@ -93,29 +95,26 @@ enum ProfilerState { kCUDA, // GPU profiling state }; -void Mark(const std::string& name, DeviceContext* dev_ctx); +void Mark(const std::string& name, const DeviceContext* dev_ctx); -void PushEvent(const std::string& name, DeviceContext* dev_ctx); +void PushEvent(const std::string& name, const DeviceContext* dev_ctx); -void PopEvent(const std::string& name, DeviceContext* dev_ctx); +void PopEvent(const std::string& name, const DeviceContext* dev_ctx); struct RecordEvent { - explicit RecordEvent(const std::string& name, DeviceContext* dev_ctx); + explicit RecordEvent(const std::string& name, const DeviceContext* dev_ctx); ~RecordEvent(); // The device context is used by Event to get the current cuda stream. - DeviceContext* dev_ctx_; + const DeviceContext* dev_ctx_; // Event name std::string name_; }; -// Enable the profiling function. -void EnableProfiler(ProfilerState state); - // Return the event list of all threads. Asummed the returned value calls // event_lists, event_lists[i][j] represents the j-th Event of i-th thread. -std::vector> DisableProfiler(); +std::vector> GetAllEvents(); // The information of each event given in the profiling report struct EventItem { @@ -130,13 +129,22 @@ struct EventItem { // Candidate keys to sort the profiling report enum EventSortingKey { kDefault, kCalls, kTotal, kMin, kMax, kAve }; +// Enable the profiling function. +void EnableProfiler(ProfilerState state); + +// Clear the g_all_event_lists, which is total event lists of all threads. +void ResetProfiler(); + +void DisableProfiler(EventSortingKey sorted_key); + // Parse the event list and output the profiling report void ParseEvents(std::vector>&, EventSortingKey sorted_by = EventSortingKey::kDefault); // Print results -void PrintProfilingReport(std::vector>& events_table, - std::string& sorted_domain, const size_t name_width, - const size_t data_width); +void PrintProfiler(std::vector>& events_table, + std::string& sorted_domain, const size_t name_width, + const size_t data_width); + } // namespace platform } // namespace paddle diff --git a/paddle/platform/profiler_test.cc b/paddle/platform/profiler_test.cc index 13dea713c71e147ed5dd8d090e92d86c96256c09..81f10c91342f76910cc780b0ebd0c0df04e9d7bf 100644 --- a/paddle/platform/profiler_test.cc +++ b/paddle/platform/profiler_test.cc @@ -103,18 +103,14 @@ TEST(RecordEvent, RecordEvent) { // Bad Usage: PushEvent("event_without_pop", dev_ctx); PopEvent("event_without_push", dev_ctx); - std::vector> events = paddle::platform::DisableProfiler(); - // Will remove parsing-related code from test later - ParseEvents(events, EventSortingKey::kTotal); + std::vector> events = paddle::platform::GetAllEvents(); int cuda_startup_count = 0; int start_profiler_count = 0; - int stop_profiler_count = 0; for (size_t i = 0; i < events.size(); ++i) { for (size_t j = 0; j < events[i].size(); ++j) { if (events[i][j].name() == "_cuda_startup_") ++cuda_startup_count; if (events[i][j].name() == "_start_profiler_") ++start_profiler_count; - if (events[i][j].name() == "_stop_profiler_") ++stop_profiler_count; if (events[i][j].name() == "push") { EXPECT_EQ(events[i][j + 1].name(), "pop"); #ifdef PADDLE_WITH_CUDA @@ -127,5 +123,7 @@ TEST(RecordEvent, RecordEvent) { } EXPECT_EQ(cuda_startup_count % 5, 0); EXPECT_EQ(start_profiler_count, 1); - EXPECT_EQ(stop_profiler_count, 1); + + // Will remove parsing-related code from test later + DisableProfiler(EventSortingKey::kTotal); } diff --git a/paddle/pybind/CMakeLists.txt b/paddle/pybind/CMakeLists.txt index 7b374307071d2da91a677361b404448f1a3816b0..de53fea0dd692167d61fcca552cc834a7916e209 100644 --- a/paddle/pybind/CMakeLists.txt +++ b/paddle/pybind/CMakeLists.txt @@ -1,7 +1,7 @@ if(WITH_PYTHON) cc_library(paddle_pybind SHARED SRCS pybind.cc exception.cc protobuf.cc const_value.cc - DEPS pybind python backward proto_desc paddle_memory executor prune init + DEPS pybind python backward proto_desc paddle_memory executor prune init profiler feed_fetch_method ${GLOB_OP_LIB}) if(NOT APPLE AND NOT ANDROID) target_link_libraries(paddle_pybind rt) diff --git a/paddle/pybind/print_operators_doc.cc b/paddle/pybind/print_operators_doc.cc index 99694fa592059d979297b72748125d02b2dd70a3..b55ddee17616ced4de659be8e55acd5e072c66b7 100644 --- a/paddle/pybind/print_operators_doc.cc +++ b/paddle/pybind/print_operators_doc.cc @@ -64,6 +64,8 @@ std::string AttrType(paddle::framework::proto::AttrType at) { return "bool array"; case paddle::framework::proto::BLOCK: return "block id"; + case paddle::framework::proto::LONG: + return "long"; } return "UNKNOWN"; // not possible } diff --git a/paddle/pybind/protobuf.cc b/paddle/pybind/protobuf.cc index 4f959481537d29c089be24f9ae306f860c196c0f..371d6119d4ab73e683821d0dc5db5194f44a64ce 100644 --- a/paddle/pybind/protobuf.cc +++ b/paddle/pybind/protobuf.cc @@ -212,6 +212,7 @@ void BindVarDsec(py::module &m) { return name; }, py::return_value_policy::reference) + .def("set_name", &VarDesc::SetName) .def("set_shape", &VarDesc::SetShape) .def("set_dtype", &VarDesc::SetDataType) .def("shape", &VarDesc::Shape, py::return_value_policy::reference) @@ -280,7 +281,8 @@ void BindOpDesc(py::module &m) { .def("check_attrs", &OpDesc::CheckAttrs) .def("infer_shape", &OpDesc::InferShape) .def("infer_var_type", &OpDesc::InferVarType) - .def("serialize_to_string", SerializeMessage); + .def("serialize_to_string", SerializeMessage) + .def("block", &OpDesc::Block, py::return_value_policy::reference); } } // namespace pybind diff --git a/paddle/pybind/protobuf.h b/paddle/pybind/protobuf.h index 089183accc08c3c486a7ae78ccfe060853ec54f5..9e747e9ea60fd95c74937daa283bc7a9eb9368c0 100644 --- a/paddle/pybind/protobuf.h +++ b/paddle/pybind/protobuf.h @@ -17,6 +17,7 @@ limitations under the License. */ #include #include #include +#include "paddle/platform/variant.h" #include "pybind11/numpy.h" #include "pybind11/pybind11.h" #include "pybind11/stl.h" diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc index c5d70bc9f91bc92b28a546cc79b08a9fda150050..a880d9bdbc63aacc1f2cdbc0d7da001a59c7b372 100644 --- a/paddle/pybind/pybind.cc +++ b/paddle/pybind/pybind.cc @@ -30,6 +30,7 @@ limitations under the License. */ #include "paddle/operators/net_op.h" #include "paddle/platform/enforce.h" #include "paddle/platform/place.h" +#include "paddle/platform/profiler.h" #include "paddle/pybind/const_value.h" #include "paddle/pybind/exception.h" #include "paddle/pybind/pybind.h" @@ -52,7 +53,7 @@ static size_t UniqueIntegerGenerator(const std::string &prefix) { return generators[prefix].fetch_add(1); } -bool IsCompileGPU() { +bool IsCompiledWithCUDA() { #ifndef PADDLE_WITH_CUDA return false; #else @@ -123,44 +124,25 @@ PYBIND11_PLUGIN(core) { .def( "__init__", [](LoDTensor &instance, const std::vector> &lod) { -#ifndef PADDLE_WITH_CUDA - new (&instance) LoDTensor(lod); -#else - LoD new_lod; - new_lod.reserve(lod.size()); - std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod)); - new (&instance) LoDTensor(new_lod); -#endif + LoD new_lod; + new_lod.reserve(lod.size()); + std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod)); + new (&instance) LoDTensor(new_lod); }) .def("__init__", [](LoDTensor &instance) { new (&instance) LoDTensor(); }) .def("set_lod", [](LoDTensor &self, const std::vector> &lod) { -#ifndef PADDLE_WITH_CUDA - self.set_lod(lod); -#else LoD new_lod; new_lod.reserve(lod.size()); std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod)); self.set_lod(new_lod); -#endif }) .def("lod", [](LoDTensor &self) -> std::vector> { -#ifndef PADDLE_WITH_CUDA - return self.lod(); -#else - auto lod = self.lod(); - std::vector> new_lod; - new_lod.reserve(lod.size()); - std::transform(lod.begin(), lod.end(), std::back_inserter(new_lod), - [](Vector item) -> - std::vector { - std::vector v; - v.reserve(item.size()); - std::copy(item.begin(), item.end(), std::back_inserter(v)); - return v; - }); - return new_lod; -#endif + auto lod = self.lod(); + std::vector> new_lod; + new_lod.reserve(lod.size()); + std::copy(lod.begin(), lod.end(), std::back_inserter(new_lod)); + return new_lod; }); py::class_(m, "SelectedRows") @@ -423,14 +405,16 @@ All parameter, weight, gradient are variables in Paddle. py::class_(m, "Executor") .def(py::init()) - .def("run", &Executor::Run); + .def("run", + (void (Executor::*)(const ProgramDesc &, Scope *, int, bool, bool)) & + Executor::Run); m.def("unique_integer", UniqueIntegerGenerator); m.def("init_gflags", framework::InitGflags); m.def("init_glog", framework::InitGLOG); m.def("init_devices", &framework::InitDevices); - m.def("is_compile_gpu", IsCompileGPU); + m.def("is_compiled_with_cuda", IsCompiledWithCUDA); m.def("set_feed_variable", framework::SetFeedVariable); m.def("get_fetch_variable", framework::GetFetchVariable); @@ -476,6 +460,24 @@ All parameter, weight, gradient are variables in Paddle. m.def("nvprof_stop", platform::CudaProfilerStop); #endif + py::enum_(m, "ProfilerState", py::arithmetic()) + .value("kDisabled", platform::ProfilerState::kDisabled) + .value("kCPU", platform::ProfilerState::kCPU) + .value("kCUDA", platform::ProfilerState::kCUDA) + .export_values(); + + py::enum_(m, "EventSortingKey", py::arithmetic()) + .value("kDefault", platform::EventSortingKey::kDefault) + .value("kCalls", platform::EventSortingKey::kCalls) + .value("kTotal", platform::EventSortingKey::kTotal) + .value("kMin", platform::EventSortingKey::kMin) + .value("kMax", platform::EventSortingKey::kMax) + .value("kAve", platform::EventSortingKey::kAve) + .export_values(); + + m.def("enable_profiler", platform::EnableProfiler); + m.def("disable_profiler", platform::DisableProfiler); + m.def("reset_profiler", platform::ResetProfiler); return m.ptr(); } } // namespace pybind diff --git a/paddle/scripts/cluster_train/paddle.py b/paddle/scripts/cluster_train/paddle.py index e44bb4505b924b3a955bb5e740aa498f253d7556..ba313ac6a18fe22e1e14d2cce42320ab6d4fe398 100644 --- a/paddle/scripts/cluster_train/paddle.py +++ b/paddle/scripts/cluster_train/paddle.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/python # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved # @@ -93,168 +80,3 @@ def job_prepare(jobdir, data=None): #create job dir run('rm ' + jobdir + ' -fr && ' + 'mkdir -p ' + jobdir) #push data and paddle bin - put(data + "/*", jobdir) - run("mkdir -p " + log) - run('rm -fr ' + log + "/*") - - def set_nodefile(nodeid): - ''' - create nodefile for later usage - ''' - run('echo ' + str(nodeid) + ' > ' + jobdir + '/nodefile') - - execute(job_create_workspace, jobdir, data, hosts=conf.HOSTS) - for i in xrange(len(conf.HOSTS)): - execute(set_nodefile, i, hosts=conf.HOSTS[i]) - #clean rubbish caused by exception - with settings(warn_only=True): - execute(kill_process, hosts=conf.HOSTS) - - -def job_pserver(jobdir, pids=None): - ''' - start all pservers - ''' - pargs = " --num_gradient_servers=" + str(len(conf.HOSTS)) - pargs += (" --nics=" + conf.PADDLE_NIC) - pargs += " --port=" + str(conf.PADDLE_PORT) - pargs += " --ports_num=" + str(conf.PADDLE_PORTS_NUM) - #always start sparse pserver by default - pargs += " --ports_num_for_sparse=" + str(conf.PADDLE_PORTS_NUM_FOR_SPARSE) - pargs += " --comment=" + "paddle_process_by_paddle" - - def start_pserver(jobdir, pargs): - ''' - start pserver process with fabric executor - ''' - with prefix('export LD_LIBRARY_PATH=' + \ - conf.LD_LIBRARY_PATH + \ - ':$LD_LIBRARY_PATH'): - program = 'paddle pserver' - run('cd ' + jobdir + '; ' + \ - 'GLOG_logtostderr=0 GLOG_log_dir="./log" ' + \ - 'nohup ' + \ - program + " " + pargs + ' > ./log/server.log 2>&1 < /dev/null & ', - pty=False) - - execute(start_pserver, jobdir, pargs, hosts=conf.HOSTS) - - -def job_trainer(jobdir, train_args_dict, pids=None): - ''' - start paddle trainer - ''' - args = " --num_gradient_servers=" + str(len(conf.HOSTS)) - args += " --nics=" + conf.PADDLE_NIC - args += " --port=" + str(conf.PADDLE_PORT) - args += " --ports_num=" + str(conf.PADDLE_PORTS_NUM) - args += " --comment=" + "paddle_process_by_paddle" - ip_string = "" - for i in xrange(len(conf.HOSTS)): - host = conf.HOSTS[i] - left = host.find("@") - right = host.find(':') - left = 0 if left == -1 else left + 1 - right = len(host) if right == -1 else right - ip_string += (socket.gethostbyname(host[left:right]) + ",") - ip_string = ip_string.rstrip(",") - args += " --pservers=" + ip_string - - args_ext = "" - for key, value in train_args_dict.items(): - args_ext += (' --' + key + '=' + value) - args += " " + args_ext - - def start_trainer(jobdir, args): - ''' - start trainer process with fabric executor - ''' - with prefix('export LD_LIBRARY_PATH=' + \ - conf.LD_LIBRARY_PATH + \ - ':$LD_LIBRARY_PATH'): - program = 'paddle train' - run('cd ' + jobdir + '; ' + \ - 'GLOG_logtostderr=0 ' - 'GLOG_log_dir="./log" ' - 'nohup ' + \ - program + " " + args + " > ./log/train.log 2>&1 < /dev/null & ", - pty=False) - - for i in xrange(len(conf.HOSTS)): - train_args = copy.deepcopy(args) - train_args += " --trainer_id=" + str(i) - execute(start_trainer, jobdir, train_args, hosts=conf.HOSTS[i]) - - -def job_all(job_package, jobdir=None, train_args_dict=None): - ''' - param job_package - param train_args_dict - ''' - if jobdir is None: - timestamp = time.strftime("%Y%m%d%H%M%S", time.localtime()) - jobdir = conf.ROOT_DIR + "/JOB" + timestamp - job_prepare(jobdir, job_package) - job_pserver(jobdir) - time.sleep(5) #wait until pservers completely start - job_trainer(jobdir, train_args_dict) - job_clean() - - -def job_clean(): - ''' - if starting job failed from paddle internal, the framework always - is launched successfully since these process are daemon processes. - so this job_clean can alway clean job rubbish process with ctrl+c. - ''' - - def signal_handler(signal, frame): - ''' - SIGINT handler - ''' - - def kill_process(): - run("ps aux \ - | grep paddle_process_by_paddle \ - | grep -v grep \ - | awk '{print $2}' \ - | xargs kill > /dev/null 2>&1") - - with settings(warn_only=True): - execute(kill_process, hosts=conf.HOSTS) - - signal.signal(signal.SIGINT, signal_handler) - signal.pause() - - -if __name__ == '__main__': - parser = argparse.ArgumentParser( - prog="paddle.py", description='simple tool for cluster training') - parser.add_argument( - '-j', - '--job_workspace', - required=False, - default=None, - help='job workspace') - parser.add_argument( - '-p', - '--job_dispatch_package', - required=False, - default=None, - help='job package for dispatching to all other nodes') - - args, train_args_list = parser.parse_known_args() - train_args = refine_unknown_args(train_args_list) - train_args_dict = dict(zip(train_args[:-1:2], train_args[1::2])) - - if args.job_workspace is not None: - #if assigned workspace, do not need to dispatch data, - #so job_local_package should be None - assert args.job_dispatch_package is None - job_all(None, args.job_workspace, train_args_dict) - elif args.job_dispatch_package is not None: - assert args.job_workspace is None - assert os.path.isdir(args.job_dispatch_package) - job_all(args.job_dispatch_package, None, train_args_dict) - else: - print "--job_workspace or --job_dispatch_package should be set" diff --git a/paddle/scripts/cpplint.py b/paddle/scripts/cpplint.py index d0cbb070c432f375430de261fd0eb3233fe26df7..dff4339ea33b72e22104a56183e3302067dc583d 100644 --- a/paddle/scripts/cpplint.py +++ b/paddle/scripts/cpplint.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python # # Copyright (c) 2009 Google Inc. All rights reserved. diff --git a/paddle/scripts/docker/README.md b/paddle/scripts/docker/README.md index f0620498cfa6775ce2949cc02fa9f6c9529dec2e..65c46745556bc5ea91fdd4e33060f2535422e8e8 100644 --- a/paddle/scripts/docker/README.md +++ b/paddle/scripts/docker/README.md @@ -56,7 +56,7 @@ Users can specify the following Docker build arguments with either "ON" or "OFF" | ------ | -------- | ----------- | | `WITH_GPU` | OFF | Generates NVIDIA CUDA GPU code and relies on CUDA libraries. | | `WITH_AVX` | OFF | Set to "ON" to enable AVX support. | -| `WITH_TESTING` | ON | Build unit tests binaries. | +| `WITH_TESTING` | OFF | Build unit tests binaries. | | `WITH_MKL` | ON | Build with [Intel® MKL](https://software.intel.com/en-us/mkl) and [Intel® MKL-DNN](https://github.com/01org/mkl-dnn) support. | | `WITH_GOLANG` | ON | Build fault-tolerant parameter server written in go. | | `WITH_SWIG_PY` | ON | Build with SWIG python API support. | diff --git a/paddle/scripts/docker/build.sh b/paddle/scripts/docker/build.sh index e70d04d9017e9e36bbd55d6a28889d9ba7fb2a13..fbae37b2ca063e32cb12ded0da901d93438bc9a2 100644 --- a/paddle/scripts/docker/build.sh +++ b/paddle/scripts/docker/build.sh @@ -32,7 +32,7 @@ function cmake_gen() { cat < #include +#include namespace paddle { namespace string { +inline std::ostream& operator<<(std::ostream& s, const std::type_index& t) { + s << t.name(); + return s; +} + template inline std::string to_string(T v) { std::ostringstream sout; @@ -25,6 +31,11 @@ inline std::string to_string(T v) { return sout.str(); } +template <> +inline std::string to_string(std::type_index t) { + return t.name(); +} + // Faster std::string/const char* type template <> inline std::string to_string(std::string v) { diff --git a/paddle/testing/paddle_gtest_main.cc b/paddle/testing/paddle_gtest_main.cc index a7fb50ee4149a3c36077f83383f45f3106e7e0f1..a2f21e37e415ccaa0d9624656728d89739972905 100644 --- a/paddle/testing/paddle_gtest_main.cc +++ b/paddle/testing/paddle_gtest_main.cc @@ -22,7 +22,9 @@ limitations under the License. */ int main(int argc, char** argv) { std::vector new_argv; std::string gflags_env; - new_argv.push_back(argv[0]); + for (int i = 0; i < argc; ++i) { + new_argv.push_back(argv[i]); + } #ifdef PADDLE_WITH_CUDA new_argv.push_back( strdup("--tryfromenv=fraction_of_gpu_memory_to_use,use_pinned_memory")); diff --git a/paddle/trainer/tests/picojson.h b/paddle/trainer/tests/picojson.h index 4aa64961d096ce94a4187fe94000b05de4080122..eaa8b9baf6e4e753a441ab77811f494cbdab80cf 100644 --- a/paddle/trainer/tests/picojson.h +++ b/paddle/trainer/tests/picojson.h @@ -1,3 +1,17 @@ +// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + /* * Copyright 2009-2010 Cybozu Labs, Inc. * Copyright 2011-2014 Kazuho Oku diff --git a/paddle/trainer/tests/simple_sparse_neural_network.py b/paddle/trainer/tests/simple_sparse_neural_network.py index ba554d5872d5a6e94526eb4d09f62a4f2f21ff30..970fb466dc5061713fe7815d5247cbbde93be821 100644 --- a/paddle/trainer/tests/simple_sparse_neural_network.py +++ b/paddle/trainer/tests/simple_sparse_neural_network.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=17, learning_method=AdaGradOptimizer(), learning_rate=1e-4) diff --git a/paddle/trainer/tests/simple_sparse_neural_network_dp.py b/paddle/trainer/tests/simple_sparse_neural_network_dp.py index 44e96873f0d38b4349f893b9f3daf53f893869cc..49043c91758b7199d063670616826656f7e8b485 100644 --- a/paddle/trainer/tests/simple_sparse_neural_network_dp.py +++ b/paddle/trainer/tests/simple_sparse_neural_network_dp.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer.PyDataProvider2 import provider, integer_sequence, integer_value import random diff --git a/paddle/utils/enable_virtualenv.py b/paddle/utils/enable_virtualenv.py index 29f8deb32455a49baad0ab5a2b5fab0a62917ec3..4e998381e9e2a9254c642e969abb9f976d0e3938 100644 --- a/paddle/utils/enable_virtualenv.py +++ b/paddle/utils/enable_virtualenv.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import os diff --git a/python/paddle/trainer/config_parser.py b/python/paddle/trainer/config_parser.py index 4fdf4090212e31adcccf6b119c937e70d5cbf995..186b91c226accbe1c2d5465d6244b9438eec9979 100644 --- a/python/paddle/trainer/config_parser.py +++ b/python/paddle/trainer/config_parser.py @@ -140,8 +140,13 @@ def init_config_environment( g_submodel_stack=[], g_add_submodel_suffix=False, ): - for k, v in locals().iteritems(): - globals()[k] = copy.deepcopy(v) + # directly iterate through locals().iteritems() will change + # the size of locals() due to introducing k, v into scope + # which will break the process in some env + + local_vars = copy.deepcopy(locals()) + for k, v in local_vars.iteritems(): + globals()[k] = v # Because type is widely used as a variable name in this code. diff --git a/python/paddle/trainer_config_helpers/evaluators.py b/python/paddle/trainer_config_helpers/evaluators.py index 95797fba8f67bacb421f5c2813ad6332bc53cbc9..0eeaf7eabb179f19d2af8dafe821f7baa153fead 100644 --- a/python/paddle/trainer_config_helpers/evaluators.py +++ b/python/paddle/trainer_config_helpers/evaluators.py @@ -16,13 +16,22 @@ from paddle.trainer.config_parser import * from default_decorators import * __all__ = [ - "evaluator_base", "classification_error_evaluator", "auc_evaluator", - "pnpair_evaluator", "precision_recall_evaluator", "ctc_error_evaluator", - "chunk_evaluator", "sum_evaluator", "column_sum_evaluator", - "value_printer_evaluator", "gradient_printer_evaluator", - "maxid_printer_evaluator", "maxframe_printer_evaluator", - "seqtext_printer_evaluator", "classification_error_printer_evaluator", - "detection_map_evaluator" + "evaluator_base", + "classification_error_evaluator", + "auc_evaluator", + "pnpair_evaluator", + "precision_recall_evaluator", + "ctc_error_evaluator", + "chunk_evaluator", + "sum_evaluator", + "column_sum_evaluator", + "value_printer_evaluator", + "gradient_printer_evaluator", + "maxid_printer_evaluator", + "maxframe_printer_evaluator", + "seqtext_printer_evaluator", + "classification_error_printer_evaluator", + "detection_map_evaluator", ] diff --git a/python/paddle/trainer_config_helpers/tests/configs/img_layers.py b/python/paddle/trainer_config_helpers/tests/configs/img_layers.py index c944a96042a96401b0309e123077041724f60246..93b505a6023cb4e2fe14f1208ac12a841ec18f55 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/img_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/img_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-3, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/img_trans_layers.py b/python/paddle/trainer_config_helpers/tests/configs/img_trans_layers.py index 27b11ffdfc7c7f8171c96997cf7fb50fe6b7667d..745f060fa5542293c646a83a81a263e92bda6a48 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/img_trans_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/img_trans_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-3, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/last_first_seq.py b/python/paddle/trainer_config_helpers/tests/configs/last_first_seq.py index 6a900518272d380fa4827161fdf626e5fa2e8fae..b6fc8f70f96487aabf249166b9b339d1b86f986e 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/last_first_seq.py +++ b/python/paddle/trainer_config_helpers/tests/configs/last_first_seq.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/layer_activations.py b/python/paddle/trainer_config_helpers/tests/configs/layer_activations.py index 06115d62e775d843eeca9d11e1def910399fec19..6edc03bba0b286fae5964bbc2bdffb99b175c718 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/layer_activations.py +++ b/python/paddle/trainer_config_helpers/tests/configs/layer_activations.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. ''' Test all activations. ''' diff --git a/python/paddle/trainer_config_helpers/tests/configs/math_ops.py b/python/paddle/trainer_config_helpers/tests/configs/math_ops.py index f5e90fdd89912d103e7404e10e1a3e8762d7c43f..59a71e1cd1d46f6be132fa5758d54edb219e9dd8 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/math_ops.py +++ b/python/paddle/trainer_config_helpers/tests/configs/math_ops.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/projections.py b/python/paddle/trainer_config_helpers/tests/configs/projections.py index c683d378caad278db0d0bcabdd4c7f2c5c62ace7..96f06b40180d5e7707ea363e6311079d81302631 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/projections.py +++ b/python/paddle/trainer_config_helpers/tests/configs/projections.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. ''' Test mixed layer, projections and operators. ''' diff --git a/python/paddle/trainer_config_helpers/tests/configs/shared_fc.py b/python/paddle/trainer_config_helpers/tests/configs/shared_fc.py index bf90d1762c89e96cfdd60ee9729cf33d2c7625c1..69a0a5b8ff53961f07ff3648c63e61d382965519 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/shared_fc.py +++ b/python/paddle/trainer_config_helpers/tests/configs/shared_fc.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/shared_gru.py b/python/paddle/trainer_config_helpers/tests/configs/shared_gru.py index 7cfab838552ab01c55bd8a9cc8eb98f79e81cc68..97b41fb3725ae325d3e67769b2d6702fdbbdb238 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/shared_gru.py +++ b/python/paddle/trainer_config_helpers/tests/configs/shared_gru.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/shared_lstm.py b/python/paddle/trainer_config_helpers/tests/configs/shared_lstm.py index 8a425c7062ef5e09c1da7d9d1e26481276eafd95..4e653dedb9d374ed53148c47582c84c2b6fb532d 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/shared_lstm.py +++ b/python/paddle/trainer_config_helpers/tests/configs/shared_lstm.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/simple_rnn_layers.py b/python/paddle/trainer_config_helpers/tests/configs/simple_rnn_layers.py index 8ee213a493596c24b7264eb8e2e34f87c2502006..dc418325f8857dd8f3ff60a05bb003a2b72e30b4 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/simple_rnn_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/simple_rnn_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-4) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_BatchNorm3D.py b/python/paddle/trainer_config_helpers/tests/configs/test_BatchNorm3D.py index cbd3c3e97f8c31fc0d0eb6778fbbc4a1f09b7358..5b98e3fb348a620e0474322b885112a6ecff3495 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_BatchNorm3D.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_BatchNorm3D.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-4) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_bi_grumemory.py b/python/paddle/trainer_config_helpers/tests/configs/test_bi_grumemory.py index bed9154fe3ed634bb65e994145c1b06c4ee88aa4..f3abdfe1ae71bf728a521c0f46ec773b058acb46 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_bi_grumemory.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_bi_grumemory.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-4) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_bilinear_interp.py b/python/paddle/trainer_config_helpers/tests/configs/test_bilinear_interp.py index 7e1da753f535e9c7c908f24250a4bfa932b3991d..4eb9f207e0cc4d209bdd616e92a88678c148ebff 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_bilinear_interp.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_bilinear_interp.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_clip_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_clip_layer.py index 0a719b073540dc3f79b827d1f9fa1a20fc23c39a..24564c105f9e1321c1a08dea611a601bc94bddb0 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_clip_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_clip_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='input', size=300) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_config_parser_for_non_file_config.py b/python/paddle/trainer_config_helpers/tests/configs/test_config_parser_for_non_file_config.py index 7003872700bca22ccf9653172e3bcc6411f90425..9b791a0222dd60e9ae2fca8b2798cddd13ed1d1c 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_config_parser_for_non_file_config.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_config_parser_for_non_file_config.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved. # diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_conv3d_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_conv3d_layer.py index fb2cacd4433f085241bfdc00c27c8c41f992daa9..35087c42289aa60f34b71b432fbf30df3a0c3167 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_conv3d_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_conv3d_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers.py b/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers.py index a8b5c860ef5abf2c12f8d93c52fba24a9df818e7..b076b89106e0854046603f24164d391b0c668474 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers_with_weight.py b/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers_with_weight.py index eba2e1e4834813be6e5771af0c61c0b54889eb45..fa7a1abe9a129ca24a6b2bb1c13dc30eddd48147 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers_with_weight.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_cost_layers_with_weight.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_crop.py b/python/paddle/trainer_config_helpers/tests/configs/test_crop.py index 870388faf74bed7e749b944a5bc44304cc098f9e..569d747857d8e756e70fea10b3e0b92801c1f4a1 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_crop.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_crop.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_cross_entropy_over_beam.py b/python/paddle/trainer_config_helpers/tests/configs/test_cross_entropy_over_beam.py index 253244dcd42a04b3212f4400ddd1b5320ce3a229..4a5bdf1181dc4538418a8b89b41a1ff713e423c8 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_cross_entropy_over_beam.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_cross_entropy_over_beam.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python #coding=utf-8 diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_deconv3d_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_deconv3d_layer.py index db950093b37a0ec0fa6f4e4fd0dcb5d8270c220b..4f27d9987346198bab06ccc59b63b0b73906e800 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_deconv3d_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_deconv3d_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_detection_output_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_detection_output_layer.py index d304a2985916b9a9508706b712b86258b1236a77..d37954222edb9202ec24bf5d6fed1da1028a276b 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_detection_output_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_detection_output_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_dot_prod_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_dot_prod_layer.py index 2e5dde2da231908462bad73700c6ac5b8bfc233e..63ba0a72b9e19a6a0da9961bd086810905f9129a 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_dot_prod_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_dot_prod_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * vec1 = data_layer(name='vector1', size=10) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_expand_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_expand_layer.py index 345fb2b6aba24b70fcab48a1a9fe466b28cbbfb5..9892bca05d52f870df1c5b2e514240fa06d7a9a4 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_expand_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_expand_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_factorization_machine.py b/python/paddle/trainer_config_helpers/tests/configs/test_factorization_machine.py index 3a489a39da1f6408dbbef02546371f12e4340b62..6fb773d9f738aa3ffa91feb7bace22e2b8e4f312 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_factorization_machine.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_factorization_machine.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='data', size=1024) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_fc.py b/python/paddle/trainer_config_helpers/tests/configs/test_fc.py index 90b0e37270f4f00e19dab35d5baa36a9902c6019..4dd37d024259dfef3e70beb00d39e73287a24473 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_fc.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_fc.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_gated_unit_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_gated_unit_layer.py index 2bd4ab2da4e98726a0f35186c3582d4623ca4779..082646b9d3d60ffd03236df027fec8417e47bda6 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_gated_unit_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_gated_unit_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='input', size=256) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_grumemory_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_grumemory_layer.py index 451909ee183ec4760613b168420433bfc919ebb8..f5271b82804f96d97bca293581951c3d5d291f40 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_grumemory_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_grumemory_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-4) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_hsigmoid.py b/python/paddle/trainer_config_helpers/tests/configs/test_hsigmoid.py index 3ebe40aadc7fb89016ee632d3ff19dc9a4b52fce..ad86d7d5bd598fc241cea50bf4689242467bbb82 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_hsigmoid.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_hsigmoid.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_kmax_seq_socre_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_kmax_seq_socre_layer.py index c762467febc5c0fc29437ca07d3793d971f12045..171da10f75dae03eed7e110d0efd07d6a18e1ecf 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_kmax_seq_socre_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_kmax_seq_socre_layer.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python #coding=utf-8 from paddle.trainer_config_helpers import * diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_l2_distance_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_l2_distance_layer.py index 58bf3de104641cb69b31fd64621f72a5df663408..1796e1c6b6d1d69db85c890a401cef630ed15f79 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_l2_distance_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_l2_distance_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * outputs( diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_lstmemory_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_lstmemory_layer.py index 8d570706dfbf672933b1f6ab78b95f19bc1033f5..7484818ab24d5ffe7e285c39e94e32cca96b2996 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_lstmemory_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_lstmemory_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_maxout.py b/python/paddle/trainer_config_helpers/tests/configs/test_maxout.py index 3b6117d297a2942fda822d5eb71e54a3fc989d06..22788be2e90e65c686b2db046a2407e41dd65fbd 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_maxout.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_maxout.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_multibox_loss_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_multibox_loss_layer.py index 083d0643678bbc53429b9722f33427e5e21df2bb..0dcccc49e43d9222f6a17fd8e25e38e3360d3dcc 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_multibox_loss_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_multibox_loss_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_multiplex_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_multiplex_layer.py index 9c1445584196985759bace73a9b9ba7b91f1870a..046d38741e87dc60e04f88f8e8d599b4f75fa4c2 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_multiplex_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_multiplex_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_ntm_layers.py b/python/paddle/trainer_config_helpers/tests/configs/test_ntm_layers.py index 046698fb4e2a17c738904dd90900d077285ec8e8..d81128c77c3bd4af9e8bbe78db61ea0232a4a837 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_ntm_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_ntm_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_pad.py b/python/paddle/trainer_config_helpers/tests/configs/test_pad.py index 1046db2f098b7eda5f428ad92d740338e791f780..44b0b34d5adb544d81c7ae6e12f3ca89652d7d8e 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_pad.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_pad.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_pooling3D_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_pooling3D_layer.py index 37805d43767ba9fb9f7a2e9790332275b0475b1a..e257e735ad25f154600172ca43d739deb2069f3a 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_pooling3D_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_pooling3D_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=100, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_prelu_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_prelu_layer.py index 10d759f6d90a70aad50561506f9bf49ae9737441..098e2397ececdd847ba76e7d4b65552c873bf2bb 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_prelu_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_prelu_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='input', size=300, height=10, width=10) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_print_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_print_layer.py index 22e0ce3e5acf33291ce71c5ecd622ec378e35b85..714d8893e951267c312ee7b924433892864e4535 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_print_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_print_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_recursive_topology.py b/python/paddle/trainer_config_helpers/tests/configs/test_recursive_topology.py index d1d97f1c5e1373c677871aa7faf3e29b8c67e68c..188a3d2320f6715cac2cd6e0cb3e9935b844452e 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_recursive_topology.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_recursive_topology.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_repeat_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_repeat_layer.py index 6818b91f969c3993cf260e7472942937fe3b9660..93b673afeecbcd4995aaa9fec0380b88c2bdd946 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_repeat_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_repeat_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_resize_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_resize_layer.py index ce8a22ebb170428e97905e941f443b43cff9ee42..3a202974e3d64659ffa3291c764b55810cc3b555 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_resize_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_resize_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='input', size=300) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_rnn_group.py b/python/paddle/trainer_config_helpers/tests/configs/test_rnn_group.py index 79dad5e2508395555bc497c66c7632906fc5cbe0..91074b8fdf3f0865516a301362ca9b5b8c469fbe 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_rnn_group.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_rnn_group.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_roi_pool_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_roi_pool_layer.py index 264341f899e161a4fea0c4ca2ae736f4154a703f..f0a37f7e992e01588c39221efd28be15058ac0af 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_roi_pool_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_roi_pool_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='data', size=3 * 14 * 14, height=14, width=14) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_row_conv.py b/python/paddle/trainer_config_helpers/tests/configs/test_row_conv.py index 342a5029a834325d3851c2f127d3f520e2a91a58..68b1a991f35c9673d5f70f86d2bbedb92dfe7d4f 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_row_conv.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_row_conv.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_row_l2_norm_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_row_l2_norm_layer.py index 9521fa6c471a1e6f396c36553cdcd4602b4c5901..c25393f580efc538dbf865b81715412041469adf 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_row_l2_norm_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_row_l2_norm_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='input', size=300) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_scale_shift_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_scale_shift_layer.py index 698d19d037580ea00911030963b627e4bf560874..3691e8daeaabf9f863d5813f89275dd7dda3039a 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_scale_shift_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_scale_shift_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='data', size=100) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_scale_sub_region_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_scale_sub_region_layer.py index 22fb25d0f23d8bf401682812b58a1a1d9165d272..426afcf3a000011a32a129b894f285de565cb9cf 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_scale_sub_region_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_scale_sub_region_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_seq_concat_reshape.py b/python/paddle/trainer_config_helpers/tests/configs/test_seq_concat_reshape.py index 1883ed9d4ed428b5a986596cc8e2cda19a400375..72960818573162a1033714cf3e36626b4b0814fa 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_seq_concat_reshape.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_seq_concat_reshape.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_seq_slice_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_seq_slice_layer.py index 12d7f1f33b1ce58abcaa415ce8ac827671a023e1..510ad3220893fddac278ba691307d00d57e440a3 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_seq_slice_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_seq_slice_layer.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python #coding=utf-8 from paddle.trainer_config_helpers import * diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_sequence_pooling.py b/python/paddle/trainer_config_helpers/tests/configs/test_sequence_pooling.py index 8cf5fd70e3fd09a66f8b5d78d09c259112a9dec3..d13a5a842990b49426120457ba0370e01815bd09 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_sequence_pooling.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_sequence_pooling.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_smooth_l1.py b/python/paddle/trainer_config_helpers/tests/configs/test_smooth_l1.py index 7188d82a534e6d37d2d4be7af35433fab19e260b..42225b85058ee82c22ee829f8d3aafe6d7d3a329 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_smooth_l1.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_smooth_l1.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * data = data_layer(name='input', size=300) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_split_datasource.py b/python/paddle/trainer_config_helpers/tests/configs/test_split_datasource.py index a628272196a018f756f3e467fd8fb5db2aa56284..7ebdf7408db17e69be481f305657051d944f8c81 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_split_datasource.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_split_datasource.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * define_py_data_sources2( diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_spp_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_spp_layer.py index 58c1675e6b6ea6288376a44d7ce9541d788187e0..1f19ea77adf3fe378e8274e9d1ee2a99552b94b1 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_spp_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_spp_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=100, learning_rate=1e-5) diff --git a/python/paddle/trainer_config_helpers/tests/configs/test_sub_nested_seq_select_layer.py b/python/paddle/trainer_config_helpers/tests/configs/test_sub_nested_seq_select_layer.py index 64d1d7b6eebe9cf18df82c2ca800794181b701b4..6d1c3175ba9801d69f3f9cb9e754858253192270 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/test_sub_nested_seq_select_layer.py +++ b/python/paddle/trainer_config_helpers/tests/configs/test_sub_nested_seq_select_layer.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/env python #coding=utf-8 from paddle.trainer_config_helpers import * diff --git a/python/paddle/trainer_config_helpers/tests/configs/unused_layers.py b/python/paddle/trainer_config_helpers/tests/configs/unused_layers.py index 6294cb04ef5165eb4fc79dd241b03dda1453cc98..8581ba60ab9e382ee9bfdd2307921b4c94d132ed 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/unused_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/unused_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(batch_size=1000, learning_rate=1e-4) diff --git a/python/paddle/trainer_config_helpers/tests/configs/util_layers.py b/python/paddle/trainer_config_helpers/tests/configs/util_layers.py index 89b881b3611d97f55e3299f761d1dcda946ba076..a66c9515c77bc04b3a72ceb31b7dbc43050d7627 100644 --- a/python/paddle/trainer_config_helpers/tests/configs/util_layers.py +++ b/python/paddle/trainer_config_helpers/tests/configs/util_layers.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.trainer_config_helpers import * settings(learning_rate=1e-4, batch_size=1000) diff --git a/python/paddle/trainer_config_helpers/tests/test_reset_hook.py b/python/paddle/trainer_config_helpers/tests/test_reset_hook.py index 0423babdb720191d8e9dfc67f1af3be339dbe27d..81186dedd20824234040e57967ce4cb53b07a8f6 100644 --- a/python/paddle/trainer_config_helpers/tests/test_reset_hook.py +++ b/python/paddle/trainer_config_helpers/tests/test_reset_hook.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import unittest from paddle.trainer.config_parser import parse_config diff --git a/python/paddle/utils/image_multiproc.py b/python/paddle/utils/image_multiproc.py index 1acf40df58e7bf5db1a76376037fd06372c9fa8b..fdbefef9ff75669becadaa1291891c0d6b7fb268 100644 --- a/python/paddle/utils/image_multiproc.py +++ b/python/paddle/utils/image_multiproc.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import os, sys import numpy as np from PIL import Image diff --git a/python/paddle/utils/plotcurve.py b/python/paddle/utils/plotcurve.py index 27a69b6a5c8b42e6b4c829e3ceebe11bed63ad15..27bd8157d39632913e2fa3278f3af20ddea61da7 100644 --- a/python/paddle/utils/plotcurve.py +++ b/python/paddle/utils/plotcurve.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. #!/usr/bin/python # Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved # diff --git a/python/paddle/v2/dataset/__init__.py b/python/paddle/v2/dataset/__init__.py index 90830515c1e8e6f5260cfca631e02a3a52cedbe5..c1acbecd9c313b02d6d33d2d04fd33fc1a8b026e 100644 --- a/python/paddle/v2/dataset/__init__.py +++ b/python/paddle/v2/dataset/__init__.py @@ -24,11 +24,23 @@ import conll05 import uci_housing import sentiment import wmt14 +import wmt16 import mq2007 import flowers import voc2012 __all__ = [ - 'mnist', 'imikolov', 'imdb', 'cifar', 'movielens', 'conll05', 'sentiment' - 'uci_housing', 'wmt14', 'mq2007', 'flowers', 'voc2012' + 'mnist', + 'imikolov', + 'imdb', + 'cifar', + 'movielens', + 'conll05', + 'sentiment' + 'uci_housing', + 'wmt14', + 'wmt16', + 'mq2007', + 'flowers', + 'voc2012', ] diff --git a/python/paddle/v2/dataset/common.py b/python/paddle/v2/dataset/common.py index fab8a68b0beee8b813bee2a05047e2da526a9c9b..9aba35a6481e3ad3ab37c8d4de0f998c9f0a1f07 100644 --- a/python/paddle/v2/dataset/common.py +++ b/python/paddle/v2/dataset/common.py @@ -25,8 +25,12 @@ import glob import cPickle as pickle __all__ = [ - 'DATA_HOME', 'download', 'md5file', 'split', 'cluster_files_reader', - 'convert' + 'DATA_HOME', + 'download', + 'md5file', + 'split', + 'cluster_files_reader', + 'convert', ] DATA_HOME = os.path.expanduser('~/.cache/paddle/dataset') @@ -58,12 +62,15 @@ def md5file(fname): return hash_md5.hexdigest() -def download(url, module_name, md5sum): +def download(url, module_name, md5sum, save_name=None): dirname = os.path.join(DATA_HOME, module_name) if not os.path.exists(dirname): os.makedirs(dirname) - filename = os.path.join(dirname, url.split('/')[-1]) + filename = os.path.join(dirname, + url.split('/')[-1] + if save_name is None else save_name) + retry = 0 retry_limit = 3 while not (os.path.exists(filename) and md5file(filename) == md5sum): @@ -196,9 +203,11 @@ def convert(output_path, reader, line_count, name_prefix): Convert data from reader to recordio format files. :param output_path: directory in which output files will be saved. - :param reader: a data reader, from which the convert program will read data instances. + :param reader: a data reader, from which the convert program will read + data instances. :param name_prefix: the name prefix of generated files. - :param max_lines_to_shuffle: the max lines numbers to shuffle before writing. + :param max_lines_to_shuffle: the max lines numbers to shuffle before + writing. """ assert line_count >= 1 diff --git a/python/paddle/v2/dataset/sentiment.py b/python/paddle/v2/dataset/sentiment.py index 7174413018cc29216a99c9291dee5ef723ea95f1..b0b9757c1a75d215cf8945b5cedbb1239fd43af7 100644 --- a/python/paddle/v2/dataset/sentiment.py +++ b/python/paddle/v2/dataset/sentiment.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. # /usr/bin/env python # -*- coding:utf-8 -*- diff --git a/python/paddle/v2/dataset/tests/imikolov_test.py b/python/paddle/v2/dataset/tests/imikolov_test.py index 9b3ab72feb0d30cbd5c86d3add31ee46ccd4d653..eed1458244562634ef44e8c9653059aaddbae6b2 100644 --- a/python/paddle/v2/dataset/tests/imikolov_test.py +++ b/python/paddle/v2/dataset/tests/imikolov_test.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.dataset.imikolov import unittest diff --git a/python/paddle/v2/dataset/tests/test_sentiment.py b/python/paddle/v2/dataset/tests/test_sentiment.py index f107948801d09f2e4490c8187217322bd57d5edb..407405290734609059c1767600748d530e8a13a6 100644 --- a/python/paddle/v2/dataset/tests/test_sentiment.py +++ b/python/paddle/v2/dataset/tests/test_sentiment.py @@ -1,16 +1,3 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. # /usr/bin/env python # -*- coding:utf-8 -*- diff --git a/python/paddle/v2/dataset/tests/wmt16_test.py b/python/paddle/v2/dataset/tests/wmt16_test.py new file mode 100644 index 0000000000000000000000000000000000000000..cef6c3216e7de8d9785a063976e63f88d90b24df --- /dev/null +++ b/python/paddle/v2/dataset/tests/wmt16_test.py @@ -0,0 +1,66 @@ +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import paddle.v2.dataset.wmt16 +import unittest + + +class TestWMT16(unittest.TestCase): + def checkout_one_sample(self, sample): + # train data has 3 field: source language word indices, + # target language word indices, and target next word indices. + self.assertEqual(len(sample), 3) + + # test start mark and end mark in source word indices. + self.assertEqual(sample[0][0], 0) + self.assertEqual(sample[0][-1], 1) + + # test start mask in target word indices + self.assertEqual(sample[1][0], 0) + + # test en mask in target next word indices + self.assertEqual(sample[2][-1], 1) + + def test_train(self): + for idx, sample in enumerate( + paddle.v2.dataset.wmt16.train( + src_dict_size=100000, trg_dict_size=100000)()): + if idx >= 10: break + self.checkout_one_sample(sample) + + def test_test(self): + for idx, sample in enumerate( + paddle.v2.dataset.wmt16.test( + src_dict_size=1000, trg_dict_size=1000)()): + if idx >= 10: break + self.checkout_one_sample(sample) + + def test_val(self): + for idx, sample in enumerate( + paddle.v2.dataset.wmt16.validation( + src_dict_size=1000, trg_dict_size=1000)()): + if idx >= 10: break + self.checkout_one_sample(sample) + + def test_get_dict(self): + dict_size = 1000 + word_dict = paddle.v2.dataset.wmt16.get_dict("en", dict_size, True) + self.assertEqual(len(word_dict), dict_size) + self.assertEqual(word_dict[0], "") + self.assertEqual(word_dict[1], "") + self.assertEqual(word_dict[2], "") + + +if __name__ == "__main__": + unittest.main() diff --git a/python/paddle/v2/dataset/wmt14.py b/python/paddle/v2/dataset/wmt14.py index 95a35d97ce9d9503153974cc167ee60829244d5f..5104e29051e4480f3a7eb18421f1b519841b009b 100644 --- a/python/paddle/v2/dataset/wmt14.py +++ b/python/paddle/v2/dataset/wmt14.py @@ -25,12 +25,20 @@ import gzip import paddle.v2.dataset.common from paddle.v2.parameters import Parameters -__all__ = ['train', 'test', 'build_dict', 'convert'] - -URL_DEV_TEST = 'http://www-lium.univ-lemans.fr/~schwenk/cslm_joint_paper/data/dev+test.tgz' +__all__ = [ + 'train', + 'test', + 'get_dict', + 'convert', +] + +URL_DEV_TEST = ('http://www-lium.univ-lemans.fr/~schwenk/' + 'cslm_joint_paper/data/dev+test.tgz') MD5_DEV_TEST = '7d7897317ddd8ba0ae5c5fa7248d3ff5' -# this is a small set of data for test. The original data is too large and will be add later. -URL_TRAIN = 'http://paddlepaddle.cdn.bcebos.com/demo/wmt_shrinked_data/wmt14.tgz' +# this is a small set of data for test. The original data is too large and +# will be add later. +URL_TRAIN = ('http://paddlepaddle.cdn.bcebos.com/demo/' + 'wmt_shrinked_data/wmt14.tgz') MD5_TRAIN = '0791583d57d5beb693b9414c5b36798c' # BLEU of this trained model is 26.92 URL_MODEL = 'http://paddlepaddle.bj.bcebos.com/demo/wmt_14/wmt14_model.tar.gz' @@ -42,8 +50,8 @@ UNK = "" UNK_IDX = 2 -def __read_to_dict__(tar_file, dict_size): - def __to_dict__(fd, size): +def __read_to_dict(tar_file, dict_size): + def __to_dict(fd, size): out_dict = dict() for line_count, line in enumerate(fd): if line_count < size: @@ -58,19 +66,19 @@ def __read_to_dict__(tar_file, dict_size): if each_item.name.endswith("src.dict") ] assert len(names) == 1 - src_dict = __to_dict__(f.extractfile(names[0]), dict_size) + src_dict = __to_dict(f.extractfile(names[0]), dict_size) names = [ each_item.name for each_item in f if each_item.name.endswith("trg.dict") ] assert len(names) == 1 - trg_dict = __to_dict__(f.extractfile(names[0]), dict_size) + trg_dict = __to_dict(f.extractfile(names[0]), dict_size) return src_dict, trg_dict def reader_creator(tar_file, file_name, dict_size): def reader(): - src_dict, trg_dict = __read_to_dict__(tar_file, dict_size) + src_dict, trg_dict = __read_to_dict(tar_file, dict_size) with tarfile.open(tar_file, mode='r') as f: names = [ each_item.name for each_item in f @@ -152,7 +160,7 @@ def get_dict(dict_size, reverse=True): # if reverse = False, return dict = {'a':'001', 'b':'002', ...} # else reverse = true, return dict = {'001':'a', '002':'b', ...} tar_file = paddle.v2.dataset.common.download(URL_TRAIN, 'wmt14', MD5_TRAIN) - src_dict, trg_dict = __read_to_dict__(tar_file, dict_size) + src_dict, trg_dict = __read_to_dict(tar_file, dict_size) if reverse: src_dict = {v: k for k, v in src_dict.items()} trg_dict = {v: k for k, v in trg_dict.items()} diff --git a/python/paddle/v2/dataset/wmt16.py b/python/paddle/v2/dataset/wmt16.py new file mode 100644 index 0000000000000000000000000000000000000000..c8818f715beadd9499ae588f2c19a57fbf26f372 --- /dev/null +++ b/python/paddle/v2/dataset/wmt16.py @@ -0,0 +1,349 @@ +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +""" +ACL2016 Multimodal Machine Translation. Please see this website for more +details: http://www.statmt.org/wmt16/multimodal-task.html#task1 + +If you use the dataset created for your task, please cite the following paper: +Multi30K: Multilingual English-German Image Descriptions. + +@article{elliott-EtAl:2016:VL16, + author = {{Elliott}, D. and {Frank}, S. and {Sima"an}, K. and {Specia}, L.}, + title = {Multi30K: Multilingual English-German Image Descriptions}, + booktitle = {Proceedings of the 6th Workshop on Vision and Language}, + year = {2016}, + pages = {70--74}, + year = 2016 +} +""" + +import os +import tarfile +import gzip +from collections import defaultdict + +import paddle.v2.dataset.common + +__all__ = [ + "train", + "test", + "validation", + "convert", + "fetch", + "get_dict", +] + +DATA_URL = ("http://cloud.dlnel.org/filepub/" + "?uuid=46a0808e-ddd8-427c-bacd-0dbc6d045fed") +DATA_MD5 = "0c38be43600334966403524a40dcd81e" + +TOTAL_EN_WORDS = 11250 +TOTAL_DE_WORDS = 19220 + +START_MARK = "" +END_MARK = "" +UNK_MARK = "" + + +def __build_dict(tar_file, dict_size, save_path, lang): + word_dict = defaultdict(int) + with tarfile.open(tar_file, mode="r") as f: + for line in f.extractfile("wmt16/train"): + line_split = line.strip().split("\t") + if len(line_split) != 2: continue + sen = line_split[0] if lang == "en" else line_split[1] + for w in sen.split(): + word_dict[w] += 1 + + with open(save_path, "w") as fout: + fout.write("%s\n%s\n%s\n" % (START_MARK, END_MARK, UNK_MARK)) + for idx, word in enumerate( + sorted( + word_dict.iteritems(), key=lambda x: x[1], reverse=True)): + if idx + 3 == dict_size: break + fout.write("%s\n" % (word[0])) + + +def __load_dict(tar_file, dict_size, lang, reverse=False): + dict_path = os.path.join(paddle.v2.dataset.common.DATA_HOME, + "wmt16/%s_%d.dict" % (lang, dict_size)) + if not os.path.exists(dict_path) or ( + len(open(dict_path, "r").readlines()) != dict_size): + __build_dict(tar_file, dict_size, dict_path, lang) + + word_dict = {} + with open(dict_path, "r") as fdict: + for idx, line in enumerate(fdict): + if reverse: + word_dict[idx] = line.strip() + else: + word_dict[line.strip()] = idx + return word_dict + + +def __get_dict_size(src_dict_size, trg_dict_size, src_lang): + src_dict_size = min(src_dict_size, (TOTAL_EN_WORDS if src_lang == "en" else + TOTAL_DE_WORDS)) + trg_dict_size = min(trg_dict_size, (TOTAL_DE_WORDS if src_lang == "en" else + TOTAL_ENG_WORDS)) + return src_dict_size, trg_dict_size + + +def reader_creator(tar_file, file_name, src_dict_size, trg_dict_size, src_lang): + def reader(): + src_dict = __load_dict(tar_file, src_dict_size, src_lang) + trg_dict = __load_dict(tar_file, trg_dict_size, + ("de" if src_lang == "en" else "en")) + + # the indice for start mark, end mark, and unk are the same in source + # language and target language. Here uses the source language + # dictionary to determine their indices. + start_id = src_dict[START_MARK] + end_id = src_dict[END_MARK] + unk_id = src_dict[UNK_MARK] + + src_col = 0 if src_lang == "en" else 1 + trg_col = 1 - src_col + + with tarfile.open(tar_file, mode="r") as f: + for line in f.extractfile(file_name): + line_split = line.strip().split("\t") + if len(line_split) != 2: + continue + src_words = line_split[src_col].split() + src_ids = [start_id] + [ + src_dict.get(w, unk_id) for w in src_words + ] + [end_id] + + trg_words = line_split[trg_col].split() + trg_ids = [trg_dict.get(w, unk_id) for w in trg_words] + + trg_ids_next = trg_ids + [end_id] + trg_ids = [start_id] + trg_ids + + yield src_ids, trg_ids, trg_ids_next + + return reader + + +def train(src_dict_size, trg_dict_size, src_lang="en"): + """ + WMT16 train set reader. + + This function returns the reader for train data. Each sample the reader + returns is made up of three fields: the source language word index sequence, + target language word index sequence and next word index sequence. + + + NOTE: + The original like for training data is: + http://www.quest.dcs.shef.ac.uk/wmt16_files_mmt/training.tar.gz + + paddle.dataset.wmt16 provides a tokenized version of the original dataset by + using moses's tokenization script: + https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/tokenizer.perl + + Args: + src_dict_size(int): Size of the source language dictionary. Three + special tokens will be added into the dictionary: + for start mark, for end mark, and for + unknown word. + trg_dict_size(int): Size of the target language dictionary. Three + special tokens will be added into the dictionary: + for start mark, for end mark, and for + unknown word. + src_lang(string): A string indicating which language is the source + language. Available options are: "en" for English + and "de" for Germany. + + Returns: + callable: The train reader. + """ + + if src_lang not in ["en", "de"]: + raise ValueError("An error language type. Only support: " + "en (for English); de(for Germany).") + src_dict_size, trg_dict_size = __get_dict_size(src_dict_size, trg_dict_size, + src_lang) + + return reader_creator( + tar_file=paddle.v2.dataset.common.download(DATA_URL, "wmt16", DATA_MD5, + "wmt16.tar.gz"), + file_name="wmt16/train", + src_dict_size=src_dict_size, + trg_dict_size=trg_dict_size, + src_lang=src_lang) + + +def test(src_dict_size, trg_dict_size, src_lang="en"): + """ + WMT16 test set reader. + + This function returns the reader for test data. Each sample the reader + returns is made up of three fields: the source language word index sequence, + target language word index sequence and next word index sequence. + + NOTE: + The original like for test data is: + http://www.quest.dcs.shef.ac.uk/wmt16_files_mmt/mmt16_task1_test.tar.gz + + paddle.dataset.wmt16 provides a tokenized version of the original dataset by + using moses's tokenization script: + https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/tokenizer.perl + + Args: + src_dict_size(int): Size of the source language dictionary. Three + special tokens will be added into the dictionary: + for start mark, for end mark, and for + unknown word. + trg_dict_size(int): Size of the target language dictionary. Three + special tokens will be added into the dictionary: + for start mark, for end mark, and for + unknown word. + src_lang(string): A string indicating which language is the source + language. Available options are: "en" for English + and "de" for Germany. + + Returns: + callable: The test reader. + """ + + if src_lang not in ["en", "de"]: + raise ValueError("An error language type. " + "Only support: en (for English); de(for Germany).") + + src_dict_size, trg_dict_size = __get_dict_size(src_dict_size, trg_dict_size, + src_lang) + + return reader_creator( + tar_file=paddle.v2.dataset.common.download(DATA_URL, "wmt16", DATA_MD5, + "wmt16.tar.gz"), + file_name="wmt16/test", + src_dict_size=src_dict_size, + trg_dict_size=trg_dict_size, + src_lang=src_lang) + + +def validation(src_dict_size, trg_dict_size, src_lang="en"): + """ + WMT16 validation set reader. + + This function returns the reader for validation data. Each sample the reader + returns is made up of three fields: the source language word index sequence, + target language word index sequence and next word index sequence. + + NOTE: + The original like for validation data is: + http://www.quest.dcs.shef.ac.uk/wmt16_files_mmt/validation.tar.gz + + paddle.dataset.wmt16 provides a tokenized version of the original dataset by + using moses's tokenization script: + https://github.com/moses-smt/mosesdecoder/blob/master/scripts/tokenizer/tokenizer.perl + + Args: + src_dict_size(int): Size of the source language dictionary. Three + special tokens will be added into the dictionary: + for start mark, for end mark, and for + unknown word. + trg_dict_size(int): Size of the target language dictionary. Three + special tokens will be added into the dictionary: + for start mark, for end mark, and for + unknown word. + src_lang(string): A string indicating which language is the source + language. Available options are: "en" for English + and "de" for Germany. + + Returns: + callable: The validation reader. + """ + if src_lang not in ["en", "de"]: + raise ValueError("An error language type. " + "Only support: en (for English); de(for Germany).") + src_dict_size, trg_dict_size = __get_dict_size(src_dict_size, trg_dict_size, + src_lang) + + return reader_creator( + tar_file=paddle.v2.dataset.common.download(DATA_URL, "wmt16", DATA_MD5, + "wmt16.tar.gz"), + file_name="wmt16/val", + src_dict_size=src_dict_size, + trg_dict_size=trg_dict_size, + src_lang=src_lang) + + +def get_dict(lang, dict_size, reverse=False): + """ + return the word dictionary for the specified language. + + Args: + lang(string): A string indicating which language is the source + language. Available options are: "en" for English + and "de" for Germany. + dict_size(int): Size of the specified language dictionary. + reverse(bool): If reverse is set to False, the returned python + dictionary will use word as key and use index as value. + If reverse is set to True, the returned python + dictionary will use index as key and word as value. + + Returns: + dict: The word dictionary for the specific language. + """ + + if lang == "en": dict_size = min(dict_size, TOTAL_EN_WORDS) + else: dict_size = min(dict_size, TOTAL_DE_WORDS) + + dict_path = os.path.join(paddle.v2.dataset.common.DATA_HOME, + "wmt16/%s_%d.dict" % (lang, dict_size)) + assert os.path.exists(dict_path), "Word dictionary does not exist. " + "Please invoke paddle.dataset.wmt16.train/test/validation first " + "to build the dictionary." + tar_file = os.path.join(paddle.v2.dataset.common.DATA_HOME, "wmt16.tar.gz") + return __load_dict(tar_file, dict_size, lang, reverse) + + +def fetch(): + """download the entire dataset. + """ + paddle.v4.dataset.common.download(DATA_URL, "wmt16", DATA_MD5, + "wmt16.tar.gz") + + +def convert(path, src_dict_size, trg_dict_size, src_lang): + """Converts dataset to recordio format. + """ + + paddle.v2.dataset.common.convert( + path, + train( + src_dict_size=src_dict_size, + trg_dict_size=trg_dict_size, + src_lang=src_lang), + 1000, + "wmt16_train") + paddle.v2.dataset.common.convert( + path, + test( + src_dict_size=src_dict_size, + trg_dict_size=trg_dict_size, + src_lang=src_lang), + 1000, + "wmt16_test") + paddle.v2.dataset.common.convert( + path, + validation( + src_dict_size=src_dict_size, + trg_dict_size=trg_dict_size, + src_lang=src_lang), + 1000, + "wmt16_validation") diff --git a/python/paddle/v2/event.py b/python/paddle/v2/event.py index f322bffe133e9a726668495826eb145062f52ac7..01067ef426d426e2921e51e8b6f620313609ab2c 100644 --- a/python/paddle/v2/event.py +++ b/python/paddle/v2/event.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. """ Testing and training events. diff --git a/python/paddle/v2/fluid/__init__.py b/python/paddle/v2/fluid/__init__.py index 8c29ee741cbb8f484531c450dd99cf183f78178e..f52346c3b59264370f46844d0e6b1e2d489299c7 100644 --- a/python/paddle/v2/fluid/__init__.py +++ b/python/paddle/v2/fluid/__init__.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function # import all class inside framework into fluid module import framework @@ -25,6 +26,7 @@ import initializer import layers import nets import optimizer +import learning_rate_decay import backward import regularizer from param_attr import ParamAttr @@ -34,26 +36,16 @@ from distribute_transpiler import DistributeTranspiler from distribute_transpiler_simple import SimpleDistributeTranspiler import clip from memory_optimization_transpiler import memory_optimize +import profiler Tensor = LoDTensor + __all__ = framework.__all__ + executor.__all__ + [ - 'io', - 'initializer', - 'layers', - 'nets', - 'optimizer', - 'backward', - 'regularizer', - 'LoDTensor', - 'CPUPlace', - 'CUDAPlace', - 'Tensor', + 'io', 'initializer', 'layers', 'nets', 'optimizer', 'learning_rate_decay', + 'backward', 'regularizer', 'LoDTensor', 'CPUPlace', 'CUDAPlace', 'Tensor', 'ParamAttr' - 'DataFeeder', - 'clip', - 'SimpleDistributeTranspiler', - 'DistributeTranspiler', - 'memory_optimize', + 'DataFeeder', 'clip', 'SimpleDistributeTranspiler', 'DistributeTranspiler', + 'memory_optimize', 'profiler' ] @@ -84,13 +76,14 @@ def __bootstrap__(): os.environ['OMP_NUM_THREADS'] = str(num_threads) - read_env_flags = ['use_pinned_memory', 'check_nan_inf'] - if core.is_compile_gpu(): - read_env_flags += ['fraction_of_gpu_memory_to_use', 'op_sync'] + read_env_flags = ['use_pinned_memory', 'check_nan_inf', 'benchmark'] + if core.is_compiled_with_cuda(): + read_env_flags += ['fraction_of_gpu_memory_to_use'] core.init_gflags([sys.argv[0]] + ["--tryfromenv=" + ",".join(read_env_flags)]) core.init_glog(sys.argv[0]) core.init_devices() +layers.monkey_patch_variable() __bootstrap__() diff --git a/python/paddle/v2/fluid/backward.py b/python/paddle/v2/fluid/backward.py index 27cf637c48bbd5fb8578d1526530b484f3a2c523..29243c90e872ca4a7d1ce6f84f6297b865655da1 100644 --- a/python/paddle/v2/fluid/backward.py +++ b/python/paddle/v2/fluid/backward.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.v2.fluid import framework as framework from . import core import collections @@ -177,7 +178,7 @@ def _remove_no_grad_branch_(op_descs, no_grad_set): if _all_in_set_( filter(lambda name: name.find(core.grad_var_suffix()) != -1, op_desc.input_arg_names()), no_grad_set): - no_grad_set.union(out_arg_names) + no_grad_set.update(out_arg_names) return True return False diff --git a/python/paddle/v2/fluid/clip.py b/python/paddle/v2/fluid/clip.py index e4d9ed599ee297b2f3cf85f3eed716b1c49578d3..fdbc8524abb7d6687983b026ca8e65e61c3dfd1a 100644 --- a/python/paddle/v2/fluid/clip.py +++ b/python/paddle/v2/fluid/clip.py @@ -1,29 +1,38 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import copy + import functools import layers +import framework from . import core __all__ = [ - 'GradientClipByValue', 'ErrorClipByValue', + 'GradientClipByValue', + 'GradientClipByNorm', + 'GradientClipByGlobalNorm', 'append_gradient_clip_ops', 'error_clip_callback', ] class BaseErrorClipAttr(object): + def __str__(self): + raise NotImplementedError() + def append_clip_op(self, block, grad_name): raise NotImplementedError() @@ -38,6 +47,9 @@ class ErrorClipByValue(BaseErrorClipAttr): self.max = max self.min = min + def __str__(self): + return "ByValue, min=%f, max=%f" % (self.min, self.max) + def append_clip_op(self, block, grad_name): clip_op_desc = block.desc.append_op() clip_op_desc.set_type("clip") @@ -65,7 +77,10 @@ def error_clip_callback(block, context): class BaseGradientClipAttr(object): - def process_context(self, context, p_g): + def __str__(self): + raise NotImplementedError() + + def process_context(self, context, param, grad): raise NotImplementedError() def create_operators(self, param, grad): @@ -73,7 +88,10 @@ class BaseGradientClipAttr(object): class NullGradientClipAttr(BaseGradientClipAttr): - def process_context(self, context, p_g): + def __str__(self): + return "Null" + + def process_context(self, context, param, grad): pass def create_operators(self, param, grad): @@ -90,7 +108,10 @@ class GradientClipByValue(BaseGradientClipAttr): self.max = max self.min = min - def process_context(self, context, p_g): + def __str__(self): + return "ByValue, min=%f, max=%f" % (self.min, self.max) + + def process_context(self, context, param, grad): pass def create_operators(self, param, grad): @@ -98,19 +119,111 @@ class GradientClipByValue(BaseGradientClipAttr): return param, new_grad +class GradientClipByNorm(BaseGradientClipAttr): + def __init__(self, clip_norm): + self.clip_norm = clip_norm + + def __str__(self): + return "ByNorm, clip_norm=%f" % self.clip_norm + + def process_context(self, context, param, grad): + pass + + def create_operators(self, param, grad): + new_grad = layers.clip_by_norm(x=grad, max_norm=self.clip_norm) + return param, new_grad + + +class GradientClipByGlobalNorm(BaseGradientClipAttr): + def __init__(self, clip_norm, group_name="default_group"): + if not isinstance(group_name, basestring): + raise TypeError("'group_name' must be a basestring.") + + self.clip_norm = clip_norm + self.group_name = group_name + + def __str__(self): + return "ByGlobalNorm, group_name=%s, clip_norm=%f" % (self.group_name, + self.clip_norm) + + def process_context(self, context, param, grad): + if self.group_name not in context: + context[self.group_name] = [] + context[self.group_name + "_clip_value"] = self.clip_norm + context[self.group_name + "_clip"] = layers.fill_constant( + shape=[1], dtype="float32", value=self.clip_norm) + else: + if not self.clip_norm == context[self.group_name + "_clip_value"]: + raise ValueError( + "All parameters' 'clip_norm' of a same group should be the same" + ) + + local_norm_var = layers.reduce_sum(input=layers.pow(x=grad, factor=2.0)) + context[self.group_name].append(local_norm_var) + + self.context = context + + def create_operators(self, param, grad): + group_scale_name = self.group_name + "_scale" + if group_scale_name not in self.context: + group_norm_var = layers.sums(input=self.context[self.group_name]) + layers.sqrt(x=group_norm_var, out=group_norm_var) + clip_var = self.context[self.group_name + "_clip"] + group_scale_var = layers.elementwise_div( + x=clip_var, + y=layers.elementwise_max( + x=clip_var, y=group_norm_var)) + assert group_scale_var.shape == (1L, ) + self.context[group_scale_name] = group_scale_var + + new_grad = layers.elementwise_mul( + x=grad, y=self.context[group_scale_name]) + return param, new_grad + + +def set_gradient_clip(clip, param_list=None, program=None): + """ + To specify parameters that require gradient clip. + Args: + clip(BaseGradientClipAttr): An instance of some derived class of BaseGradientClipAttr, + which describes the type and detailed attributes of required gradient clip. + param_list(list, None by default): Parameters that require gradient clip. + It can be a list of parameter or a list of parameter's name. + When it's None, all parameters in the program will be included. + program(Program, None by default): The program where parameters are. + Will be the default main program when assigned with None. + """ + if not isinstance(clip, BaseGradientClipAttr): + raise TypeError( + "'clip' should be an instance of BaseGradientClipAttr's derived class" + ) + if program is None: + program = framework.default_main_program() + if param_list is None: + param_list = program.block(0).all_parameters() + if all(isinstance(elem, basestring) for elem in param_list): + param_list = [program.block(0).var(elem) for elem in param_list] + if not all(isinstance(elem, framework.Parameter) for elem in param_list): + raise TypeError( + "'param_list' should be a list of Parameter or basestring(parameter's name)." + ) + + for param in param_list: + param.gradient_clip_attr = copy.deepcopy(clip) + + def append_gradient_clip_ops(param_grad): context = dict() create_op_callbacks = [] for p, g in param_grad: - clip_attr = getattr(p, 'clip_attr', NullGradientClipAttr()) + clip_attr = getattr(p, 'gradient_clip_attr', NullGradientClipAttr()) if clip_attr is None: clip_attr = NullGradientClipAttr() if not isinstance(clip_attr, BaseGradientClipAttr): raise TypeError( - "clip attribute should be an instance of BaseGradientClippingAttr" - ) + "clip attribute should be an instance of BaseGradientClipAttr") - clip_attr.process_context(context=context, p_g=param_grad) + clip_attr.process_context(context=context, param=p, grad=g) create_op_callbacks.append( functools.partial( clip_attr.create_operators, param=p, grad=g)) @@ -119,3 +232,5 @@ def append_gradient_clip_ops(param_grad): ClipByValue = GradientClipByValue +ClipByNorm = GradientClipByNorm +ClipByGlobalNorm = GradientClipByGlobalNorm diff --git a/python/paddle/v2/fluid/data_feeder.py b/python/paddle/v2/fluid/data_feeder.py index bfdd00e3ef77f5977926d452a648df5259f58823..a3b22a8633eae02c570f2a4c4caf4a6152649ef8 100644 --- a/python/paddle/v2/fluid/data_feeder.py +++ b/python/paddle/v2/fluid/data_feeder.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import core import numpy diff --git a/python/paddle/v2/fluid/default_scope_funcs.py b/python/paddle/v2/fluid/default_scope_funcs.py index 2218bb140ac57bd3d7fa9cd0b97a9bcea48aa201..a27280208b8184c6539274afb1ddd6bda2861205 100644 --- a/python/paddle/v2/fluid/default_scope_funcs.py +++ b/python/paddle/v2/fluid/default_scope_funcs.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. """ Default scope function. diff --git a/python/paddle/v2/fluid/distribute_transpiler.py b/python/paddle/v2/fluid/distribute_transpiler.py index 06a7b6fb02f5e38e0762d112492854f027fe66ad..a4464a281aae714d79a531ec8a2cf793d6330a12 100644 --- a/python/paddle/v2/fluid/distribute_transpiler.py +++ b/python/paddle/v2/fluid/distribute_transpiler.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import framework from framework import Program, default_main_program, Parameter, Variable @@ -18,6 +19,7 @@ import optimizer from layer_helper import LayerHelper from distributed_spliter import * import math +from . import core class VarBlock: @@ -31,19 +33,23 @@ class VarBlock: return "%s:%d:%d" % (self.varname, self.offset, self.size) +def same_or_split_var(p_name, var_name): + return p_name == var_name or p_name.startswith(var_name + ".block") + + def split_dense_variable(var_list, pserver_count, min_block_size=1024, max_block_size=1048576): """ - We may need to split dense tensor to one or several blocks and put + We may need to split dense tensor to one or more blocks and put them equally onto parameter server. One block is a sub-tensor aligned by dim[0] of the tensor. - + We need to have a minimal block size so that the calculations in the parameter server side can gain better performance. By default - mininum block size is 1024. The max block size is used to prevent - too large block that may causing send error. + minimum block size is 1024. The max block size is used to prevent + very large blocks that may cause send error. """ blocks = [] for var in var_list: @@ -62,7 +68,7 @@ def split_dense_variable(var_list, remains = block_size % dim1 if remains != 0: block_size += dim1 - remains - # update split_count after align + # update split_count after aligning split_count = int(math.ceil(var_numel / float(block_size))) for block_id in xrange(split_count): curr_block_size = min(block_size, var_numel - ( @@ -81,18 +87,18 @@ class DistributeTranspiler: trainers=1, split_method=round_robin): """ - Transpile the program to a distributed data-parallelism programs. - The main_program will be transform to use a remote parameter server + Transpile the program to distributed data-parallelism programs. + The main_program will be transformed to use a remote parameter server to do parameter optimization. And the optimization graph will be put - in to a parameter server program. + into a parameter server program. - Use different methods to split trainable varialbles to different + Use different methods to split trainable variables to different parameter servers. :param optimize_ops: op list of optimization, should be the return value of Optimizer.minimize :type optimize_ops: list - :param program: program to optimize, default default_main_program + :param program: program to optimize, default is default_main_program :param pservers: parameter server endpoints like "m1:6174,m2:6174" :type pservers: string :return: return a list of programs @@ -104,11 +110,11 @@ class DistributeTranspiler: self.trainers = trainers self.optimize_ops = optimize_ops # steps to transpile: - # 1. split variable to multiple blocks, align by product(dim[1:]) (width). + # 1. split variable to multiple blocks, aligned by product(dim[1:]) (width). # 2. modify trainer program add split_op to each Grad. # 3. append send_op to trainer. # 4. append concat_op to trainer to update local weights. - # 5. create new program as parameter server. + # 5. create new program for parameter server. # 6. create parameter server program by split_method generated endpoint->VarBlock pserver_endpoints = pservers.split(",") @@ -134,10 +140,10 @@ class DistributeTranspiler: for b in param_blocks: varname, block_id, _ = b.split(":") send_outputs.append(param_var_mapping[varname][int(block_id)]) - # let send_op know which endpoint to send which var, eplist is of the same - # order of send_inputs. + # let send_op know which endpoint to send which var to, eplist has the same + # order as send_inputs. eplist = split_method(send_inputs, pserver_endpoints) - # create mapping of endpoint -> splited var to create pserver side program + # create mapping of endpoint -> split var to create pserver side program self.param_grad_ep_mapping = dict() for i, ep in enumerate(eplist): param = send_outputs[i] @@ -147,10 +153,18 @@ class DistributeTranspiler: self.param_grad_ep_mapping[ep]["params"].append(param) self.param_grad_ep_mapping[ep]["grads"].append(grad) + rpc_client_var = program.global_block().create_var( + name="RPC_CLIENT_VAR", + psersistable=True, + dtype='float32', # dtype and shape is not used in fact + shape=[0]) + + # create send_op send_op = program.global_block().append_op( type="send", inputs={"X": send_inputs}, - outputs={"Out": send_outputs}, + outputs={"Out": send_outputs, + "RPCClient": rpc_client_var}, attrs={"endpoints": pserver_endpoints, "epmap": eplist}) # step4 @@ -165,6 +179,7 @@ class DistributeTranspiler: attrs={"axis": 0}) def _create_vars_from_blocklist(self, program, block_list): + # Create respective variables using the block_list block_map = dict() var_mapping = dict() for block_str in block_list: @@ -205,26 +220,40 @@ class DistributeTranspiler: dtype=var.dtype, type=var.type, lod_level=var.lod_level, - # HACK: let all param in pserver persistable so child + # HACK: let all param in pserver be persistable so the child # program in recv can get them persistable=True) def _append_split_op(self, program, gradblocks): + # Split variables that need to be split and append respective ops var_mapping = self._create_vars_from_blocklist(program, gradblocks) for varname, splited_vars in var_mapping.iteritems(): # variable that don't need to split have empty splited_vars if len(splited_vars) <= 1: continue orig_var = program.global_block().vars[varname] - sections = [] - for v in splited_vars: - sections.append(v.shape[0]) - program.global_block().append_op( - type="split", - inputs={"X": orig_var}, - outputs={"Out": splited_vars}, - attrs={"sections": sections} # assume split evenly - ) + if orig_var.type == core.VarDesc.VarType.SELECTED_ROWS: + height_sections = [] + for v in splited_vars: + height_sections.append(v.shape[0]) + program.global_block().append_op( + type="split_selected_rows", + inputs={"X": orig_var}, + outputs={"Out": splited_vars}, + attrs={"height_sections": height_sections}) + elif orig_var.type == core.VarDesc.VarType.LOD_TENSOR: + sections = [] + for v in splited_vars: + sections.append(v.shape[0]) + program.global_block().append_op( + type="split", + inputs={"X": orig_var}, + outputs={"Out": splited_vars}, + attrs={"sections": sections} # assume split evenly + ) + else: + AssertionError("Variable type should be in set " + "[LOD_TENSOR, SELECTED_ROWS]") return var_mapping def get_trainer_program(self): @@ -233,6 +262,7 @@ class DistributeTranspiler: return self.program def _create_var_for_trainers(self, block, var, trainers): + # For each trainer, create the necessary variables var_list = [] for i in xrange(trainers): var_each = block.create_var( @@ -247,7 +277,7 @@ class DistributeTranspiler: param_shape): """ Returns the shape for optimizer inputs that need to be reshaped when - Param and Grad is splited to multiple servers. + Param and Grad is split to multiple servers. """ # HACK(typhoonzero): Should use functions of corresponding optimizer in # optimizer.py to get the shape, do not bind this in the transpiler. @@ -284,8 +314,8 @@ class DistributeTranspiler: return True else: for n in param_names: - if n.startswith(op.inputs["Param"].name+".block") and \ - n != op.inputs["Param"].name: + if same_or_split_var(n, op.inputs[ + "Param"].name) and n != op.inputs["Param"].name: return True return False else: @@ -316,7 +346,7 @@ class DistributeTranspiler: if key == "Grad": grad_block = None for g in self.param_grad_ep_mapping[endpoint]["grads"]: - if g.name.startswith(var.name): + if same_or_split_var(g.name, var.name): grad_block = g break if not grad_block: @@ -346,7 +376,7 @@ class DistributeTranspiler: # param is already created on global program param_block = None for p in self.param_grad_ep_mapping[endpoint]["params"]: - if p.name.startswith(var.name): + if same_or_split_var(p.name, var.name): param_block = p break if not param_block: @@ -381,7 +411,7 @@ class DistributeTranspiler: dtype=var.dtype, shape=new_shape) - # change outputs ParamOut variable + # change output's ParamOut variable opt_op.outputs["ParamOut"] = new_inputs["Param"] program.global_block().append_op( type=opt_op.type, @@ -390,6 +420,7 @@ class DistributeTranspiler: attrs=opt_op.attrs) def _append_pserver_non_opt_ops(self, program, pserver_program, opt_op): + # Append the ops for parameters that do not need to be optimized/updated for _, var in opt_op.inputs.iteritems(): program.global_block().create_var( name=var.name, @@ -407,9 +438,9 @@ class DistributeTranspiler: outputs=opt_op.outputs, attrs=opt_op.attrs) - def get_pserver_program(self, endpoint, optimize_ops): + def get_pserver_program(self, endpoint): """ - get pserver side program by endpoint + Get pserver side program using the endpoint NOTE: assume blocks of the same variable is not distributed on the same pserver, only change param/grad varnames for @@ -420,11 +451,25 @@ class DistributeTranspiler: pserver_program = Program() for v in self.param_grad_ep_mapping[endpoint]["params"]: self._clone_var(pserver_program.global_block(), v) + for v in self.param_grad_ep_mapping[endpoint]["grads"]: + # create vars for each trainer in global scope, so + # we don't need to create them when grad arrives. + pserver_program.global_block().create_var( + name=v.name, persistable=True, dtype=v.dtype, shape=v.shape) + for trainer_id in xrange(self.trainers): + print("create variable for program: %s.trainer_%d" % + (v.name, trainer_id)) + pserver_program.global_block().create_var( + name="%s.trainer_%d" % (v.name, trainer_id), + persistable=True, + dtype=v.dtype, + shape=v.shape) # step6 optimize_sub_program = Program() - for idx, opt_op in enumerate(optimize_ops): - is_op_on_pserver = self._is_op_on_pserver(endpoint, optimize_ops, - idx) + # Iterate through the ops and append ops as needed + for idx, opt_op in enumerate(self.optimize_ops): + is_op_on_pserver = self._is_op_on_pserver(endpoint, + self.optimize_ops, idx) if not is_op_on_pserver: continue if opt_op.inputs.has_key("Grad"): @@ -433,13 +478,13 @@ class DistributeTranspiler: else: self._append_pserver_non_opt_ops(optimize_sub_program, pserver_program, opt_op) + # Append the recv op pserver_program.global_block().append_op( type="recv", - inputs={"RX": self.param_grad_ep_mapping[endpoint]["grads"] - }, # grads to recv + inputs={}, outputs={}, attrs={ - "OptimizeProgram": optimize_sub_program.desc, + "OptimizeBlock": optimize_sub_program.global_block(), "endpoint": endpoint, "ParamList": [ p.name @@ -449,7 +494,7 @@ class DistributeTranspiler: p.name for p in self.param_grad_ep_mapping[endpoint]["grads"] ], - "Trainers": self.trainers + "Fanin": self.trainers }) pserver_program.sync_with_cpp() return pserver_program @@ -458,7 +503,7 @@ class DistributeTranspiler: """ Get startup program for current parameter server. Modify operator input variables if there are variables that - was splited to several blocks. + were split to several blocks. """ s_prog = Program() orig_s_prog = framework.default_startup_program() @@ -467,7 +512,7 @@ class DistributeTranspiler: def _get_splited_name_and_shape(varname): for idx, splited_param in enumerate(params): pname = splited_param.name - if pname.startswith(varname) and varname != pname: + if same_or_split_var(pname, varname) and varname != pname: return pname, splited_param.shape return "", [] diff --git a/python/paddle/v2/fluid/distribute_transpiler_simple.py b/python/paddle/v2/fluid/distribute_transpiler_simple.py index bd88f02bde0c6a58138e20db2b07cbd06cd40ba3..73d9bed1ae9d81d66eb32675ea8473da248076f0 100644 --- a/python/paddle/v2/fluid/distribute_transpiler_simple.py +++ b/python/paddle/v2/fluid/distribute_transpiler_simple.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import framework from framework import Program, default_main_program, Parameter, Variable import optimizer @@ -243,7 +244,7 @@ class SimpleDistributeTranspiler: self.param_grad_map[endpoint]["grads"]}, # grads to recv outputs={}, attrs={ - "OptimizeProgram": optimize_sub_program.desc, + "OptimizeBlock": optimize_sub_program.global_block(), "endpoint": endpoint, "ParamList": [p.name for p in self.param_grad_map[endpoint]["params"]], diff --git a/python/paddle/v2/fluid/distributed_spliter.py b/python/paddle/v2/fluid/distributed_spliter.py index e647f760e9d3d400e28f54215b684079b2279ffc..8cf0b06786f2ccb5601af10aab39e7d0c22ae624 100644 --- a/python/paddle/v2/fluid/distributed_spliter.py +++ b/python/paddle/v2/fluid/distributed_spliter.py @@ -1,16 +1,18 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + + def hash_name(varlist, pserver_endpoints): """ hash variable names to several endpoints. diff --git a/python/paddle/v2/fluid/evaluator.py b/python/paddle/v2/fluid/evaluator.py index adf174a07daeea521fa3a1c97273ec68b3a9a67f..2686a5bdfcf0e2d26ce8f58cceff1967b06d835b 100644 --- a/python/paddle/v2/fluid/evaluator.py +++ b/python/paddle/v2/fluid/evaluator.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import layers @@ -204,3 +205,63 @@ class ChunkEvaluator(Evaluator): [precision], dtype='float32'), np.array( [recall], dtype='float32'), np.array( [f1_score], dtype='float32') + + +class EditDistance(Evaluator): + """ + Accumulate edit distance sum and sequence number from mini-batches and + compute the average edit_distance of all batches. + + Args: + input: the sequences predicted by network. + label: the target sequences which must has same sequence count + with input. + ignored_tokens(list of int): Tokens that should be removed before + calculating edit distance. + + Example: + + exe = fluid.executor(place) + distance_evaluator = fluid.Evaluator.EditDistance(input, label) + for epoch in PASS_NUM: + distance_evaluator.reset(exe) + for data in batches: + loss, sum_distance = exe.run(fetch_list=[cost] + distance_evaluator.metrics) + avg_distance = distance_evaluator.eval(exe) + pass_distance = distance_evaluator.eval(exe) + + In the above example: + 'sum_distance' is the sum of the batch's edit distance. + 'avg_distance' is the average of edit distance from the firt batch to the current batch. + 'pass_distance' is the average of edit distance from all the pass. + + """ + + def __init__(self, input, label, ignored_tokens=None, **kwargs): + super(EditDistance, self).__init__("edit_distance", **kwargs) + main_program = self.helper.main_program + if main_program.current_block().idx != 0: + raise ValueError("You can only invoke Evaluator in root block") + + self.total_error = self.create_state( + dtype='float32', shape=[1], suffix='total_error') + self.seq_num = self.create_state( + dtype='int64', shape=[1], suffix='seq_num') + error, seq_num = layers.edit_distance( + input=input, label=label, ignored_tokens=ignored_tokens) + #error = layers.cast(x=error, dtype='float32') + sum_error = layers.reduce_sum(error) + layers.sums(input=[self.total_error, sum_error], out=self.total_error) + layers.sums(input=[self.seq_num, seq_num], out=self.seq_num) + self.metrics.append(sum_error) + + def eval(self, executor, eval_program=None): + if eval_program is None: + eval_program = Program() + block = eval_program.current_block() + with program_guard(main_program=eval_program): + total_error = _clone_var_(block, self.total_error) + seq_num = _clone_var_(block, self.seq_num) + seq_num = layers.cast(x=seq_num, dtype='float32') + out = layers.elementwise_div(x=total_error, y=seq_num) + return np.array(executor.run(eval_program, fetch_list=[out])[0]) diff --git a/python/paddle/v2/fluid/executor.py b/python/paddle/v2/fluid/executor.py index a99c5157b285d7edbf06398c00df3e7ec514cd91..9f48815b8b84426c7d539af4e7d45ea47e69d4d9 100644 --- a/python/paddle/v2/fluid/executor.py +++ b/python/paddle/v2/fluid/executor.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import contextlib from framework import Program, default_main_program @@ -67,6 +68,84 @@ def as_numpy(tensor): return ans +def has_feed_operators(block, feed_targets, feed_holder_name): + """ Check whether the block already has feed operators. + + Return false if the block does not have any feed operators. + If some feed operators have been prepended to the block, check that + the info contained in these feed operators matches the feed_targets + and feed_holder_name. Raise exception when any mismatch is found. + Return true when the block has feed operators with matching info. + + Args: + block: a block instance (typically global block of a program) + feed_targets: a dictionary of {feed_target_name: feed_target_data} + feed_holder_name: the name of the variable that holds the data of + all feed targets. The type of this feed_holder variable is + FEED_MINIBATCH, which is essentially vector. + + Returns: + A boolean value that indicates whether a block has feed operators + that match the info contained in feed_targets and feed_holder_name. + """ + + feed_count = 0 + for op in block.ops: + if op.desc.type() == 'feed': + feed_count += 1 + assert op.desc.input('X')[0] == feed_holder_name + feed_target_name = op.desc.output('Out')[0] + if feed_target_name not in feed_targets: + raise Exception("'feed_targets' does not have {} variable". + format(feed_target_name)) + else: + break + if feed_count > 0 and feed_count != len(feed_targets): + raise Exception( + "Feed operators in program desc do not match 'feed_targets'") + return feed_count > 0 + + +def has_fetch_operators(block, fetch_targets, fetch_holder_name): + """ Check whether the block already has fetch operators. + + Return false if the block does not have any fetch operators. + If some fetch operators have been appended to the block, check that + the info contained in these fetch operators matches the fetch_targets + and fetch_holder_name. Raise exception when any mismatch is found. + Return true when the block has fetch operators with matching info. + + Args: + block: a block instance (typically global block of a program) + fetch_targets: a dictionary of {fetch_target_name: fetch_target_data} + fetch_holder_name: the name of the variable that holds the data of + all fetch targets. The type of this fetch_holder variable is + FETCH_LIST, which is essentially vector. + + Return: + A boolean value that indicates whether a block has fetch operators + that match the info contained in fetch_targets and fetch_holder_name. + """ + + fetch_count = 0 + for op in block.ops: + if op.desc.type() == 'fetch': + fetch_count += 1 + assert op.desc.output('Out')[0] == fetch_holder_name + fetch_target_name = op.desc.input('X')[0] + if fetch_target_name not in [ + var.desc.name() for var in fetch_targets + ]: + raise Exception("'fetch_targets' does not have {} variable". + format(fetch_target_name)) + idx = op.desc.attr('col') + assert fetch_target_name == fetch_targets[idx].desc.name() + if fetch_count > 0 and fetch_count != len(fetch_targets): + raise Exception( + "Fetch operators in program desc do not match 'fetch_targets'") + return fetch_count > 0 + + class Executor(object): def __init__(self, places): if not isinstance(places, list) and not isinstance(places, tuple): @@ -146,33 +225,50 @@ class Executor(object): program = program.clone() global_block = program.global_block() - feed_var = global_block.create_var( - name=feed_var_name, - type=core.VarDesc.VarType.FEED_MINIBATCH, - persistable=True) - - for i, name in enumerate(feed): - out = global_block.var(name) - global_block.prepend_op( - 'feed', - inputs={'X': [feed_var]}, - outputs={'Out': [out]}, - attrs={'col': i}) - cur_feed = feed[name] - if not isinstance(cur_feed, core.LoDTensor): - cur_feed = self.aslodtensor(cur_feed) - core.set_feed_variable(scope, cur_feed, feed_var.name, i) - - fetch_var = global_block.create_var( - name=fetch_var_name, - type=core.VarDesc.VarType.FETCH_LIST, - persistable=True) - for i, var in enumerate(fetch_list): - global_block.append_op( - type='fetch', - inputs={'X': [var]}, - outputs={'Out': [fetch_var]}, - attrs={'col': i}) + + if feed_var_name in global_block.vars: + feed_var = global_block.var(feed_var_name) + else: + feed_var = global_block.create_var( + name=feed_var_name, + type=core.VarDesc.VarType.FEED_MINIBATCH, + persistable=True) + + if fetch_var_name in global_block.vars: + fetch_var = global_block.var(fetch_var_name) + else: + fetch_var = global_block.create_var( + name=fetch_var_name, + type=core.VarDesc.VarType.FETCH_LIST, + persistable=True) + + if not has_feed_operators(global_block, feed, feed_var_name): + for i, name in enumerate(feed): + out = global_block.var(name) + global_block.prepend_op( + type='feed', + inputs={'X': [feed_var]}, + outputs={'Out': [out]}, + attrs={'col': i}) + + for op in global_block.ops: + if op.desc.type() == 'feed': + feed_target_name = op.desc.output('Out')[0] + cur_feed = feed[feed_target_name] + if not isinstance(cur_feed, core.LoDTensor): + cur_feed = self.aslodtensor(cur_feed) + idx = op.desc.attr('col') + core.set_feed_variable(scope, cur_feed, feed_var_name, idx) + else: + break + + if not has_fetch_operators(global_block, fetch_list, fetch_var_name): + for i, var in enumerate(fetch_list): + global_block.append_op( + type='fetch', + inputs={'X': [var]}, + outputs={'Out': [fetch_var]}, + attrs={'col': i}) self.executor.run(program.desc, scope, 0, True, True) outs = [ diff --git a/python/paddle/v2/fluid/framework.py b/python/paddle/v2/fluid/framework.py index 8042febfed7ed7db2f6d1507142b17079aa00fd8..8bf545e2ecc3939b00ba25d003a6b3887a54f860 100644 --- a/python/paddle/v2/fluid/framework.py +++ b/python/paddle/v2/fluid/framework.py @@ -1,18 +1,20 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import collections import contextlib +import re import numpy as np @@ -116,8 +118,8 @@ def _debug_string_(proto, throw_on_error=True): """ error_fields = list() if not proto.IsInitialized(error_fields) and throw_on_error: - raise ValueError("{0} are not initialized\nThe message is {1}".format( - error_fields, proto)) + raise ValueError("{0} are not initialized.\nThe message is {1}:\n". + format(error_fields, proto)) return proto.__str__() @@ -238,20 +240,30 @@ class Variable(object): def __str__(self): return self.to_string(True) - def to_string(self, throw_on_error): + def to_string(self, throw_on_error, with_details=False): """ Get debug string. Args: throw_on_error(bool): True if raise an exception when self is not intialized. + with_details(bool): more details about variables and parameters + (e.g. trainable, optimize_attr, ...) will be printed when with_details is True Returns(str): The debug string. """ + assert isinstance(throw_on_error, bool) and isinstance(with_details, + bool) protostr = self.desc.serialize_to_string() proto = framework_pb2.VarDesc.FromString(str(protostr)) - return _debug_string_(proto, throw_on_error) + res_str = _debug_string_(proto, throw_on_error) + if with_details: + additional_attr = ("error_clip", "stop_gradient") + for attr_name in additional_attr: + res_str += "%s: %s\n" % (attr_name, + str(getattr(self, attr_name))) + return res_str __repr__ = __str__ @@ -374,12 +386,13 @@ class Operator(object): >>> outputs={"Out": [var1]}) Args: - block(Block): The block has the current operator - desc(core.OpDesc): The protobuf description + block(Block): The block has the current operator. + desc(core.OpDesc): The protobuf description. type(str): The type of operator. inputs(dict): The input dictionary. Key is the input parameter name. Value is a list of variables. - outputs(dict): The output dictionary. Has same format with inputs + outputs(dict): The output dictionary which has the same format with + inputs. attrs(dict): The attributes dictionary. Key is attribute name. Value is the attribute value. The attribute type should be as same as the type registered in C++ @@ -436,10 +449,11 @@ class Operator(object): for m in proto.outputs: need.add(m.name) if not given == need: - raise ValueError( - "Incorrect setting for output(s) of operator \"%s\". Need: [%s] Given: [%s]" - % (type, ", ".join(str(e) for e in need), ", ".join( - str(e) for e in given))) + raise ValueError(("Incorrect setting for output(s) of " + "operator \"%s\". Need: [%s] Given: [%s]") % + (type, ", ".join(str(e) + for e in need), ", ".join( + str(e) for e in given))) for out_proto in proto.outputs: out_args = outputs[out_proto.name] @@ -626,10 +640,36 @@ class Block(object): def __str__(self): return self.to_string(True) - def to_string(self, throw_on_error): - protostr = self.desc.serialize_to_string() - proto = framework_pb2.BlockDesc.FromString(str(protostr)) - return _debug_string_(proto, throw_on_error) + def to_string(self, throw_on_error, with_details=False): + """ + To debug string. + Args: + throw_on_error(bool): raise exception when self is not initialized + when throw_on_error is True + with_details(bool): more details about variables and parameters + (e.g. trainable, optimize_attr, ...) will be printed when with_details is True + + Returns(str): The debug string. + + """ + assert isinstance(throw_on_error, bool) and isinstance(with_details, + bool) + if with_details: + re_add_indent = re.compile(r"\n(.)") + res_str = "blocks {\n idx: %d\n parent_idx: %d" % ( + self.idx, self.parent_idx) + for var in self.vars.itervalues(): + res_str += "\n vars {\n %s }" % re_add_indent.sub( + r"\n \1", var.to_string(throw_on_error, with_details)) + for op in self.ops: + res_str += "\n ops {\n %s }" % re_add_indent.sub( + r"\n \1", op.to_string(throw_on_error)) + res_str += "\n}" + else: + protostr = self.desc.serialize_to_string() + proto = framework_pb2.BlockDesc.FromString(str(protostr)) + res_str = _debug_string_(proto, throw_on_error) + return res_str __repr__ = __str__ @@ -777,7 +817,7 @@ class Block(object): trainable=p.trainable, optimize_attr=p.optimize_attr, regularizer=p.regularizer, - clip_attr=p.clip_attr, + gradient_clip_attr=p.gradient_clip_attr, error_clip=p.error_clip, name=v.name) self.vars[new_p.name] = new_p @@ -793,10 +833,29 @@ class Program(object): def __str__(self): return self.to_string(True) - def to_string(self, throw_on_error): - protostr = self.desc.serialize_to_string() - proto = framework_pb2.ProgramDesc.FromString(str(protostr)) - return _debug_string_(proto, throw_on_error) + def to_string(self, throw_on_error, with_details=False): + """ + To debug string. + Args: + throw_on_error(bool): raise exception when self is not initialized + when throw_on_error is True + with_details(bool): more details about variables and parameters + (e.g. trainable, optimize_attr, ...) will be printed when with_details is True + + Returns(str): The debug string. + + """ + assert isinstance(throw_on_error, bool) and isinstance(with_details, + bool) + if with_details: + res_str = "" + for block in self.blocks: + res_str += block.to_string(throw_on_error, with_details) + else: + protostr = self.desc.serialize_to_string() + proto = framework_pb2.ProgramDesc.FromString(str(protostr)) + res_str = _debug_string_(proto, throw_on_error) + return res_str def get_desc(self): return self.desc @@ -818,9 +877,8 @@ class Program(object): if isinstance(t, Variable): t = t.op else: - raise ValueError( - "All targets of prune() can only be Variable or Operator." - ) + raise ValueError(("All targets of prune() can only be " + "Variable or Operator.")) targets_idx.append([t.block.idx, t.idx]) res = Program() @@ -946,7 +1004,37 @@ class Parameter(Variable): self.regularizer = kwargs.get('regularizer', None) - self.clip_attr = kwargs.get('clip_attr', None) + self.gradient_clip_attr = kwargs.get('gradient_clip_attr', None) + + def __str__(self): + return self.to_string(True) + + def to_string(self, throw_on_error, with_details=False): + """ + To debug string. + Args: + throw_on_error(bool): raise exception when self is not initialized + when throw_on_error is True + with_details(bool): more details about variables and parameters + (e.g. trainable, optimize_attr, ...) will be printed when with_details is True + + Returns(str): The debug string. + + """ + assert isinstance(throw_on_error, bool) and isinstance(with_details, + bool) + if with_details: + res_str = Variable.to_string(self, throw_on_error, True) + additional_attr = ("trainable", "optimize_attr", "regularizer", + "gradient_clip_attr") + for attr_name in additional_attr: + res_str += "%s: %s\n" % (attr_name, + str(getattr(self, attr_name))) + else: + res_str = Variable.to_string(self, throw_on_error, False) + return res_str + + __repr__ = __str__ # program is a global instance. diff --git a/python/paddle/v2/fluid/initializer.py b/python/paddle/v2/fluid/initializer.py index 2e8cfa3177ba00b48a1297b1deec7350588745f2..b9c0d12ad6cf09e66df6b1a8da09df275c79a3f6 100644 --- a/python/paddle/v2/fluid/initializer.py +++ b/python/paddle/v2/fluid/initializer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import framework import numpy as np diff --git a/python/paddle/v2/fluid/io.py b/python/paddle/v2/fluid/io.py index 499df05e592855f63f41ec8ceb939edf0e4d435c..d56ec45c538b580f5520bc060b4b339bb1be0539 100644 --- a/python/paddle/v2/fluid/io.py +++ b/python/paddle/v2/fluid/io.py @@ -1,20 +1,22 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import os -import cPickle as pickle +from paddle.v2.fluid.evaluator import Evaluator from paddle.v2.fluid.framework import Program, Parameter, default_main_program, Variable +from . import core __all__ = [ 'save_vars', @@ -185,12 +187,53 @@ def get_inference_program(target_vars, main_program=None): main_program = default_main_program() if not isinstance(target_vars, list): target_vars = [target_vars] - - pruned_program = main_program.prune(targets=target_vars) + vars = [] + for var in target_vars: + if isinstance(var, Evaluator): + vars.extend(var.states) + vars.extend(var.metrics) + else: + vars.append(var) + pruned_program = main_program.prune(targets=vars) inference_program = pruned_program.inference_optimize() return inference_program +def prepend_feed_ops(inference_program, + feed_target_names, + feed_holder_name='feed'): + global_block = inference_program.global_block() + feed_var = global_block.create_var( + name=feed_holder_name, + type=core.VarDesc.VarType.FEED_MINIBATCH, + persistable=True) + + for i, name in enumerate(feed_target_names): + out = global_block.var(name) + global_block.prepend_op( + type='feed', + inputs={'X': [feed_var]}, + outputs={'Out': [out]}, + attrs={'col': i}) + + +def append_fetch_ops(inference_program, + fetch_target_names, + fetch_holder_name='fetch'): + global_block = inference_program.global_block() + fetch_var = global_block.create_var( + name=fetch_holder_name, + type=core.VarDesc.VarType.FETCH_LIST, + persistable=True) + + for i, name in enumerate(fetch_target_names): + global_block.append_op( + type='fetch', + inputs={'X': [name]}, + outputs={'Out': [fetch_var]}, + attrs={'col': i}) + + def save_inference_model(dirname, feeded_var_names, target_vars, @@ -233,18 +276,12 @@ def save_inference_model(dirname, inference_program = pruned_program.inference_optimize() fetch_var_names = [v.name for v in target_vars] + prepend_feed_ops(inference_program, feeded_var_names) + append_fetch_ops(inference_program, fetch_var_names) + model_file_name = dirname + "/__model__" - with open(model_file_name, "w") as f: - pickle.dump({ - "program_desc_str": inference_program.desc.serialize_to_string(), - "feed_var_names": feeded_var_names, - "fetch_var_names": fetch_var_names - }, f, -1) - - # Save only programDesc of inference_program in binary format - # in another file: __model__.dat - with open(model_file_name + ".dat", "wb") as fp: - fp.write(inference_program.desc.serialize_to_string()) + with open(model_file_name, "wb") as f: + f.write(inference_program.desc.serialize_to_string()) save_params(executor, dirname, main_program) @@ -267,6 +304,24 @@ def load_persistables_if_exist(executor, dirname, main_program=None): predicate=_is_presistable_and_exist_) +def get_feed_targets_names(program): + feed_targets_names = [] + global_block = program.global_block() + for op in global_block.ops: + if op.desc.type() == 'feed': + feed_targets_names.insert(0, op.desc.output('Out')[0]) + return feed_targets_names + + +def get_fetch_targets_names(program): + fetch_targets_names = [] + global_block = program.global_block() + for op in global_block.ops: + if op.desc.type() == 'fetch': + fetch_targets_names.append(op.desc.input('X')[0]) + return fetch_targets_names + + def load_inference_model(dirname, executor): """ Load inference model from a directory @@ -274,24 +329,28 @@ def load_inference_model(dirname, executor): :param dirname: directory path :param executor: executor that load inference model - :return: [program, feed_var_names, fetch_var_names] + :return: [program, feed_target_names, fetch_targets] program: program especially for inference. - feeded_var_names: Names of variables that need to feed data - fetch_vars: Variables from which we can get inference results. + feed_target_names: Names of variables that need to feed data + fetch_targets: Variables from which we can get inference results. """ if not os.path.isdir(dirname): raise ValueError("There is no directory named '%s'", dirname) model_file_name = dirname + "/__model__" - model = pickle.load(open(model_file_name, "r")) - program_desc_str = model["program_desc_str"] - feed_var_names = model["feed_var_names"] - fetch_var_names = model["fetch_var_names"] + with open(model_file_name, "rb") as f: + program_desc_str = f.read() + program = Program.parse_from_string(program_desc_str) load_persistables_if_exist(executor, dirname, program) - fetch_vars = [program.global_block().var(name) for name in fetch_var_names] - return [program, feed_var_names, fetch_vars] + feed_target_names = get_feed_targets_names(program) + fetch_target_names = get_fetch_targets_names(program) + fetch_targets = [ + program.global_block().var(name) for name in fetch_target_names + ] + + return [program, feed_target_names, fetch_targets] def get_parameter_value(para, executor): diff --git a/python/paddle/v2/fluid/layer_helper.py b/python/paddle/v2/fluid/layer_helper.py index 191d2349b5e692f1f8ad9c068daf5592031433ad..2119ca12c8dea6463934aa68cb1b46ec687e3f72 100644 --- a/python/paddle/v2/fluid/layer_helper.py +++ b/python/paddle/v2/fluid/layer_helper.py @@ -1,23 +1,24 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import copy import itertools from framework import Variable, Parameter, default_main_program, default_startup_program, \ unique_name, dtype_is_floating from paddle.v2.fluid.initializer import Constant, Xavier -from param_attr import ParamAttr +from param_attr import ParamAttr, WeightNormParamAttr class LayerHelper(object): @@ -99,9 +100,181 @@ class LayerHelper(object): if dtype is None: dtype = each.dtype elif dtype != each.dtype: - raise ValueError("Data Type mismatch") + raise ValueError("Data Type mismatch: %d to %d" % + (dtype, each.dtype)) return dtype + def _create_weight_normalize(self, attr, shape, dtype): + from .layers import elementwise_mul, elementwise_div, reshape + + # Remove these ops when LayerHelper and layers support indicating + # program and block. + def __norm_op(x, + out=None, + p=2, + dim=None, + keep_dim=False, + block=self.startup_program.global_block()): + if out is None: + out = block.create_var( + name=unique_name(".".join([self.name, 'weight_norm_norm'])), + dtype=dtype, + persistable=False) + abs_out = block.create_var( + name=unique_name(".".join([self.name, 'weight_norm_abs'])), + dtype=dtype, + persistable=False) + block.append_op( + type='abs', inputs={'X': x}, outputs={'Out': abs_out}) + pow_out = block.create_var( + name=unique_name(".".join([self.name, 'weight_norm_pow'])), + dtype=dtype, + persistable=False) + block.append_op( + type='pow', + inputs={'X': abs_out}, + outputs={'Out': pow_out}, + attrs={'factor': float(p)}) + sum_out = block.create_var( + name=unique_name(".".join([self.name, 'weight_norm_sum'])), + dtype=dtype, + persistable=False) + block.append_op( + type='reduce_sum', + inputs={'X': pow_out}, + outputs={'Out': sum_out}, + attrs={ + 'dim': dim, + 'keep_dim': keep_dim, + 'reduce_all': True if dim is None else False + }) + block.append_op( + type='pow', + inputs={'X': sum_out}, + outputs={'Out': out}, + attrs={'factor': 1. / p}) + return out + + def __reshape_op(x, + shape, + out=None, + block=self.startup_program.global_block()): + if out is None: + out = block.create_var( + name=unique_name(".".join( + [self.name, 'weight_norm_reshape'])), + dtype=dtype, + persistable=False) + block.append_op( + type='reshape', + inputs={'X': x}, + outputs={'Out': out}, + attrs={'shape': shape}) + return out + + def __transpose_op(x, + axis, + out=None, + block=self.startup_program.global_block()): + if out is None: + out = block.create_var( + name=unique_name(".".join( + [self.name, 'weight_norm_transpose'])), + dtype=dtype, + persistable=False) + block.append_op( + type='transpose', + inputs={'X': x}, + outputs={'Out': out}, + attrs={'axis': axis}) + return out + + def __norm_except_dim(x, + out=None, + dim=None, + block=self.startup_program.global_block()): + """Computes the norm over all dimensions except dim""" + if out is None: + out = block.create_var( + name=unique_name(".".join([self.name, 'weight_norm_norm'])), + dtype=dtype, + persistable=False) + if dim is None: + __norm_op(x, out, dim=dim, block=block) + elif dim == 0: + out_shape = [x.shape[0]] + [1] * (len(x.shape) - 1) + reshape = __reshape_op(x, shape=[x.shape[0], -1], block=block) + norm = __norm_op(reshape, dim=1, block=block) + __reshape_op(norm, out=out, shape=out_shape, block=block) + elif dim == len(x.shape) - 1: + out_shape = [1] * (len(x.shape) - 1) + [x.shape[-1]] + reshape = __reshape_op(x, shape=[-1, x.shape[-1]], block=block) + norm = __norm_op(reshape, dim=0, block=block) + __reshape_op(norm, out=out, shape=out_shape, block=block) + else: + perm = range(len(x.shape)) + perm[0], perm[dim] = dim, 0 + transpose = __transpose_op(x, perm, block=block) + norm = __norm_op(transpose, dim=0, block=block) + __transpose_op(norm, perm, out=out, block=block) + return out + + def __weight_normalize(g, v, dim): + """Calculations for weight normalization""" + norm = __norm_except_dim( + v, dim=dim, block=self.main_program.current_block()) + scale = elementwise_div( + x=g, y=norm) # The shapes of g and norm are the same. + # Currently, elementwise_mul only support broadcast when the shape + # of y is a subset of the shape of x. Thus, we reshape y to squeeze + # to achive the subset. + w = elementwise_mul( + x=v, + y=scale if dim is None else reshape( + x=scale, shape=[v.shape[dim]]), + axis=-1 if dim is None else dim) + # To serialize the original parameter for inference, maybe a + # parameter rather than a variable should be returned. + return w + + g_param_attr = copy.deepcopy(attr) + g_param_attr.name = attr.name + '_g' + g_param_shape = [1] * len(shape) + if attr.dim is not None: + g_param_shape[attr.dim] = shape[attr.dim] + v_param_attr = copy.deepcopy(attr) + v_param_attr.name = attr.name + '_v' + v_param_shape = shape + + # Add to startup_program to initialize g and v. + # Try to reconstruct the initializer of w by initializing g and v. + # Set the initializers of g and v as below, then the distribution + # of w is the same as initializing w with the given initializer. + # For Data-Dependent Initialization, please compute the init-values + # of g and v in external and then feed the values to g and v by + # executing an extra program. + g_param = self.startup_program.global_block().create_parameter( + dtype=dtype, + shape=g_param_shape, + **g_param_attr.to_kwargs(with_initializer=False)) + v_param = self.startup_program.global_block().create_parameter( + dtype=dtype, + shape=v_param_shape, + **v_param_attr.to_kwargs(with_initializer=True)) + __norm_except_dim( + x=v_param, + out=g_param, + dim=attr.dim, + block=self.startup_program.global_block()) + + # Add weight normalization to main_program + g_param = self.main_program.global_block().create_parameter( + dtype=dtype, shape=g_param_shape, **g_param_attr.to_kwargs()) + v_param = self.main_program.global_block().create_parameter( + dtype=dtype, shape=v_param_shape, **v_param_attr.to_kwargs()) + w_param = __weight_normalize(g_param, v_param, dim=attr.dim) + return w_param + def create_parameter(self, attr, shape, @@ -109,18 +282,26 @@ class LayerHelper(object): is_bias=False, default_initializer=None): # Deepcopy the attr so that parameters can be shared in program + attr = copy.deepcopy(attr) assert isinstance(attr, ParamAttr) suffix = 'b' if is_bias else 'w' + if attr.name is None: + attr.name = unique_name(".".join([self.name, suffix])) - if default_initializer is None: + if default_initializer is None and attr.initializer is None: if is_bias: attr.set_default_bias_initializer() else: attr.set_default_param_initializer() else: attr.set_default_initializer(default_initializer) - if attr.name is None: - attr.name = unique_name(".".join([self.name, suffix])) + + # If weight normalization is set, insert extra parameters and ops. + # Refer to https://arxiv.org/pdf/1602.07868.pdf + if isinstance(attr, WeightNormParamAttr): + param = self._create_weight_normalize(attr, shape, dtype) + WeightNormParamAttr.params_with_weight_norm.append(param) + return param self.startup_program.global_block().create_parameter( dtype=dtype, shape=shape, **attr.to_kwargs(with_initializer=True)) diff --git a/python/paddle/v2/fluid/layers/__init__.py b/python/paddle/v2/fluid/layers/__init__.py index c190af3329409e5b87b182a11a84dc87dfc46d6e..a83dd3db74aed548a324a1c605723c957fca8604 100644 --- a/python/paddle/v2/fluid/layers/__init__.py +++ b/python/paddle/v2/fluid/layers/__init__.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import ops from ops import * import nn @@ -23,6 +24,8 @@ import control_flow from control_flow import * import device from device import * +import math_op_patch +from math_op_patch import * __all__ = [] __all__ += nn.__all__ @@ -31,3 +34,4 @@ __all__ += tensor.__all__ __all__ += control_flow.__all__ __all__ += ops.__all__ __all__ += device.__all__ +__all__ += math_op_patch.__all__ diff --git a/python/paddle/v2/fluid/layers/control_flow.py b/python/paddle/v2/fluid/layers/control_flow.py index e72b22c83f65789b9e5d81611bec602d8d78be6b..0fcbfe0e2f2f9686366139e84b7fdcc158bf0aa7 100644 --- a/python/paddle/v2/fluid/layers/control_flow.py +++ b/python/paddle/v2/fluid/layers/control_flow.py @@ -1,31 +1,51 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from ..layer_helper import LayerHelper, unique_name -from ..framework import Program, Variable, Operator -from .. import core -from tensor import assign, fill_constant +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import contextlib -from ..registry import autodoc + +from layer_function_generator import autodoc +from tensor import assign, fill_constant +from .. import core +from ..framework import Program, Variable, Operator +from ..layer_helper import LayerHelper, unique_name __all__ = [ - 'split_lod_tensor', 'merge_lod_tensor', 'BlockGuard', - 'BlockGuardWithCompletion', 'StaticRNNMemoryLink', 'WhileGuard', 'While', - 'lod_rank_table', 'max_sequence_len', 'topk', 'lod_tensor_to_array', - 'array_to_lod_tensor', 'increment', 'array_write', 'create_array', - 'less_than', 'array_read', 'shrink_memory', 'array_length', 'IfElse', - 'DynamicRNN', 'ConditionalBlock', 'StaticRNN', 'reorder_lod_tensor_by_rank', - 'ParallelDo', 'Print' + 'split_lod_tensor', + 'merge_lod_tensor', + 'BlockGuard', + 'BlockGuardWithCompletion', + 'StaticRNNMemoryLink', + 'WhileGuard', + 'While', + 'lod_rank_table', + 'max_sequence_len', + 'topk', + 'lod_tensor_to_array', + 'array_to_lod_tensor', + 'increment', + 'array_write', + 'create_array', + 'less_than', + 'array_read', + 'shrink_memory', + 'array_length', + 'IfElse', + 'DynamicRNN', + 'ConditionalBlock', + 'StaticRNN', + 'reorder_lod_tensor_by_rank', + 'ParallelDo', + 'Print', ] @@ -269,6 +289,7 @@ class ParallelDo(object): for in_var_name in op.input(iname): if in_var_name not in local_inputs: params.append(in_var_name) + params = list(set(params)) return [parent_block.var(name) for name in params] @@ -749,7 +770,7 @@ def topk(input, k): array = fluid.layers.topk(x, k) """ helper = LayerHelper('topk', **locals()) - topk_out = helper.create_tmp_variable(dtype=input.data_type) + topk_out = helper.create_tmp_variable(dtype=input.dtype) topk_indices = helper.create_tmp_variable(dtype='int64') helper.append_op( type='top_k', @@ -1457,7 +1478,7 @@ class DynamicRNN(object): method)) -@autodoc +@autodoc() def reorder_lod_tensor_by_rank(x, rank_table): helper = LayerHelper('reorder_lod_tensor_by_rank', **locals()) helper.is_instance('x', Variable) diff --git a/python/paddle/v2/fluid/layers/device.py b/python/paddle/v2/fluid/layers/device.py index ef74b2b2f08ae446b612a5e5552344b3901d8178..107511b5f4ab1108610bc1326f30e5d9ab407853 100644 --- a/python/paddle/v2/fluid/layers/device.py +++ b/python/paddle/v2/fluid/layers/device.py @@ -1,28 +1,28 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. """ All util layers. """ -from ..layer_helper import LayerHelper +from layer_function_generator import autodoc from ..framework import unique_name -from ..registry import autodoc +from ..layer_helper import LayerHelper __all__ = ['get_places'] -@autodoc +@autodoc() def get_places(device_count=None, device_type=None): helper = LayerHelper('get_places', **locals()) out_places = helper.create_variable(name=unique_name(helper.name + ".out")) diff --git a/python/paddle/v2/fluid/layers/io.py b/python/paddle/v2/fluid/layers/io.py index 6177f0b4d71cdda8637fc40e3e65c72842bf7439..b7b2cf2296cc8868dd0b5eb6cd6d58b9ae795d5d 100644 --- a/python/paddle/v2/fluid/layers/io.py +++ b/python/paddle/v2/fluid/layers/io.py @@ -1,20 +1,23 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from .. import core from ..layer_helper import LayerHelper +from control_flow import BlockGuard +from ..layer_helper import LayerHelper -__all__ = ['data'] +__all__ = ['data', 'BlockGuardServ', 'ListenAndServ', 'Send'] def data(name, @@ -28,9 +31,9 @@ def data(name, **Data Layer** This function takes in the input and based on whether data has - to be returned back as a minibatch, it creates the global variable using + to be returned back as a minibatch, it creates the global variable by using the helper functions. The global variables can be accessed by all the - following operations and layers in the graph. + following operators in the graph. All the input variables of this function are passed in as local variables to the LayerHelper constructor. @@ -73,3 +76,123 @@ def data(name, type=type, stop_gradient=stop_gradient, lod_level=lod_level) + + +class BlockGuardServ(BlockGuard): + """ + BlockGuardServ class. + + BlockGuardServ class is used to create an op with a block in a program. + """ + + def __init__(self, server): + if not (isinstance(server, ListenAndServ)): + raise TypeError("BlockGuardServ takes a ListenAndServ") + super(BlockGuardServ, self).__init__(server.helper.main_program) + self.server = server + + def __exit__(self, exc_type, exc_val, exc_tb): + if exc_type is not None: + return False + + self.server.complete_op() + return super(BlockGuardServ, self).__exit__(exc_type, exc_val, exc_tb) + + +class ListenAndServ(object): + """ + ListenAndServ class. + + ListenAndServ class is used to wrap listen_and_serv op to create a server + which can receive variables from clients and run a block. + """ + + def __init__(self, endpoint, fan_in=1, optimizer_mode=True): + self.helper = LayerHelper("recv") + self.inputs = [] + self.outputs = [] + self.endpoint = endpoint + self.fan_in = fan_in + # FIXME(typhoonzero): add optimizer_mode is stupid, should make it more + # general. + self.optimizer_mode = optimizer_mode + + def do(self): + return BlockGuardServ(self) + + def get_params_and_grads(self): + main_program = self.helper.main_program + current_block = main_program.current_block() + parent_block = self.parent_block() + # params and grads in the same order. + params = list() + grads = list() + for op in current_block.ops: + # FIXME(typhoonzero): op.inputs is None if it's cloned. + if self.optimizer_mode: + if "Grad" in op.inputs and "Param" in op.inputs: + params.append(op.inputs["Param"].name) + grads.append(op.inputs["Grad"].name) + else: + # simple recv mode, recv operators inputs. + for iname in op.input_names: + for in_var_name in op.input(iname): + params.append(parent_block.var(in_var_name)) + grads.append(parent_block.var(in_var_name)) + + return params, grads + + def parent_block(self): + prog = self.helper.main_program + parent_idx = prog.current_block().parent_idx + assert parent_idx >= 0 + parent_block = prog.block(parent_idx) + return parent_block + + def complete_op(self): + main_program = self.helper.main_program + current_block = main_program.current_block() + parent_block = self.parent_block() + + params, grads = self.get_params_and_grads() + param_names = [p.name for p in params] + grad_names = [g.name for g in grads] + parent_block.append_op( + type='recv', + inputs={}, + outputs={}, + attrs={ + 'endpoint': self.endpoint, + 'Fanin': self.fan_in, + 'ParamList': param_names, + 'GradList': grad_names, + 'OptimizeBlock': current_block + }) + + +def Send(endpoints, send_vars, get_vars): + """ + Send layer + + Args: + endpoints: comma seperated IP:PORT pairs in the order + of send_vars to send + send_vars: vars to send + get_vars: vars to get from server after send completes. + + Send variables to the server side, and get vars from server + side when server have finished running server side program. + """ + assert (type(send_vars) == list) + assert (type(get_vars) == list) + + epmap = endpoints.split(",") + endpoints = list(set(epmap)) + + helper = LayerHelper("Send", **locals()) + helper.append_op( + type="send", + inputs={"X": send_vars}, + outputs={"Out": get_vars}, + attrs={"endpoints": endpoints, + "epmap": epmap}) diff --git a/python/paddle/v2/fluid/registry.py b/python/paddle/v2/fluid/layers/layer_function_generator.py similarity index 81% rename from python/paddle/v2/fluid/registry.py rename to python/paddle/v2/fluid/layers/layer_function_generator.py index 6c0c3a35185391873fe5bb98d1ed5ee1cf13aa15..b0e4d1635f7b5d0afdfa677e6ec1e8f9245a9d54 100644 --- a/python/paddle/v2/fluid/registry.py +++ b/python/paddle/v2/fluid/layers/layer_function_generator.py @@ -1,29 +1,31 @@ # Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import re import cStringIO -import warnings import functools -import inspect +import warnings + +from .. import proto -import proto.framework_pb2 as framework_pb2 -from framework import OpProtoHolder, Variable, Program, Operator -from paddle.v2.fluid.layer_helper import LayerHelper, unique_name +framework_pb2 = proto.framework_pb2 + +from ..framework import OpProtoHolder, Variable +from ..layer_helper import LayerHelper __all__ = [ 'deprecated', - 'register_layer', + 'generate_layer_fn', 'autodoc', ] @@ -96,7 +98,7 @@ def _generate_doc_string_(op_proto): return buf.getvalue() -def register_layer(op_type): +def generate_layer_fn(op_type): """Register the Python layer for an Operator. Args: @@ -167,13 +169,18 @@ def register_layer(op_type): inputs[ipt.name] = val outputs = dict() - out = helper.create_tmp_variable(dtype=dtype) - outputs[o_name] = [out] + out = kwargs.pop(_convert_(o_name), []) + if out: + out_var = out[0] if (isinstance(out, list) or + isinstance(out, tuple)) else out + else: + out_var = helper.create_tmp_variable(dtype=dtype) + outputs[o_name] = [out_var] for name in intermediate_output_names: outputs[name] = [helper.create_tmp_variable(dtype=dtype)] helper.append_op( type=op_type, inputs=inputs, outputs=outputs, attrs=kwargs) - return helper.append_activation(out) + return helper.append_activation(out_var) func.__name__ = op_type func.__doc__ = _generate_doc_string_(op_proto) @@ -202,7 +209,10 @@ def deprecated(func_or_class): return func_wrapper -def autodoc(func): - func.__doc__ = _generate_doc_string_(OpProtoHolder.instance().get_op_proto( - func.__name__)) - return func +def autodoc(comment=""): + def __impl__(func): + func.__doc__ = _generate_doc_string_(OpProtoHolder.instance( + ).get_op_proto(func.__name__)) + comment + return func + + return __impl__ diff --git a/python/paddle/v2/fluid/layers/math_op_patch.py b/python/paddle/v2/fluid/layers/math_op_patch.py new file mode 100644 index 0000000000000000000000000000000000000000..79a130a3eb148e6c5a8fa3cdf174780b354c23c9 --- /dev/null +++ b/python/paddle/v2/fluid/layers/math_op_patch.py @@ -0,0 +1,154 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from ..framework import Variable, unique_name +from layer_function_generator import OpProtoHolder + +__all__ = ['monkey_patch_variable'] + + +def monkey_patch_variable(): + def unique_tmp_name(): + return unique_name("tmp") + + def safe_get_dtype(var): + try: + dtype = var.dtype + except: + raise ValueError("Cannot get data type from %s", var.name) + return dtype + + def create_tensor(block, value, dtype, shape): + value = float(value) + tmp_name = unique_tmp_name() + var = block.create_var(name=tmp_name, shape=shape, dtype=dtype) + block.append_op( + type="fill_constant", + outputs={'Out': [var]}, + attrs={'dtype': var.dtype, + 'shape': shape, + 'value': value}) + return var + + def create_scalar(block, value, dtype): + return create_tensor(block, value, dtype, shape=[1]) + + def create_tensor_with_batchsize(ref_var, value, dtype): + assert isinstance(ref_var, Variable) + value = float(value) + tmp_name = unique_tmp_name() + var = ref_var.block.create_var(name=tmp_name, dtype=dtype) + ref_var.block.append_op( + type='fill_constant_batch_size_like', + outputs={'Out': [var]}, + inputs={'Input': [ref_var]}, + attrs={'shape': ref_var.shape, + 'value': value}) + return var + + def astype(self, dtype): + """ + Cast a variable to a specified data type. + NOTE: The variable must be a Tensor + Args: + self(Variable): The source variable + dtype: The target dtype + + Returns: + Variable with new dtype + """ + tmp_name = unique_tmp_name() + out = self.block.create_var(name=tmp_name, dtype=dtype) + self.block.append_op( + type="cast", + inputs={"X": [self]}, + outputs={"Out": [out]}, + attrs={"in_dtype": self.dtype, + "out_dtype": out.dtype}) + return out + + def _elemwise_method_creator_(method_name, op_type, reverse=False): + def __impl__(self, other_var): + lhs_dtype = safe_get_dtype(self) + + if not isinstance(other_var, Variable): + if reverse: + has_batch_size = False + for elem in self.shape: + if elem < 0: + has_batch_size = True + break + if not has_batch_size: + other_var = create_tensor( + self.block, + other_var, + dtype=lhs_dtype, + shape=self.shape) + else: + other_var = create_tensor_with_batchsize( + self, other_var, lhs_dtype) + else: + # add fill_op to self.block + other_var = create_scalar( + self.block, value=other_var, dtype=lhs_dtype) + + rhs_dtype = safe_get_dtype(other_var) + if lhs_dtype != rhs_dtype: + other_var = astype(other_var, lhs_dtype) + if reverse: + tmp = self + self = other_var + other_var = tmp + + tmp_name = unique_tmp_name() + out = self.block.create_var(name=tmp_name, dtype=lhs_dtype) + self.block.append_op( + type=op_type, + inputs={'X': [self], + 'Y': [other_var]}, + outputs={'Out': out}) + return out + + comment = OpProtoHolder.instance().get_op_proto(op_type).comment + + __impl__.__doc__ = """ + {0} + Args: + self(Variable): left hand variable + other_var(Variable|float|int): right hand variable + + Returns: + Variable + """.format(comment) + __impl__.__name__ = method_name + return __impl__ + + # inject methods + for method_name, op_type, reverse in ( + ("__add__", "elementwise_add", False), + # a+b == b+a. Do not need to reverse explicitly + ("__radd__", "elementwise_add", False), + ("__sub__", "elementwise_sub", False), + ("__rsub__", "elementwise_sub", True), + ("__mul__", "elementwise_mul", False), + # a*b == b*a. Do not need to reverse explicitly + ("__rmul__", "elementwise_mul", False), + ("__div__", "elementwise_div", False), + ("__rdiv__", "elementwise_div", True), + ("__pow__", "elementwise_pow", False), + ("__rpow__", "elementwise_pow", True)): + setattr(Variable, method_name, + _elemwise_method_creator_(method_name, op_type, reverse)) + + Variable.astype = astype diff --git a/python/paddle/v2/fluid/layers/nn.py b/python/paddle/v2/fluid/layers/nn.py index 4e8fd407c9983e2827d3137fa4f49a8425d5dce2..c38e21087de1bf7076ce5aaf23d4d4faaebb50a7 100644 --- a/python/paddle/v2/fluid/layers/nn.py +++ b/python/paddle/v2/fluid/layers/nn.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. """ All layers just related to the neural network. """ @@ -19,12 +19,15 @@ from ..layer_helper import LayerHelper from ..initializer import Normal, Constant from ..framework import Variable from ..param_attr import ParamAttr +from layer_function_generator import autodoc from tensor import concat __all__ = [ 'fc', 'embedding', 'dynamic_lstm', + 'dynamic_lstmp', + 'dynamic_gru', 'gru_unit', 'linear_chain_crf', 'crf_decoding', @@ -50,6 +53,18 @@ __all__ = [ 'sequence_last_step', 'dropout', 'split', + 'ctc_greedy_decoder', + 'edit_distance', + 'l2_normalize', + 'matmul', + 'warpctc', + 'sequence_reshape', + 'transpose', + 'im2sequence', + 'nce', + 'beam_search', + 'row_conv', + 'multiplex', ] @@ -64,14 +79,14 @@ def fc(input, **Fully Connected Layer** The fully connected layer can take multiple tensors as its inputs. It - creates a variable (one for each input tensor) called weights for each input - tensor, which represents a fully connected weight matrix from each input - unit to each output unit. The fully connected layer multiplies each input - tensor with its coresponding weight to produce an output Tensor. If - multiple input tensors are given, the results of multiple multiplications - will be sumed up. If bias_attr is not None, a biases variable will be - created and added to the output. Finally, if activation is not None, - it will be applied to the output as well. + creates a variable (one for each input tensor) called weights for each + input tensor, which represents a fully connected weight matrix from + each input unit to each output unit. The fully connected layer + multiplies each input tensor with its coresponding weight to produce + an output Tensor. If multiple input tensors are given, the results of + multiple multiplications will be sumed up. If bias_attr is not None, + a biases variable will be created and added to the output. Finally, + if activation is not None, it will be applied to the output as well. This process can be formulated as follows: @@ -97,16 +112,17 @@ def fc(input, into a 2-dimensional matrix. The parameter `num_flatten_dims` determines how the input tensor is flattened: the first `num_flatten_dims` - dimensions will be flatten to form the first - dimension of the final matrix (height of the - matrix), and the rest `rank(X) - num_flatten_dims` - dimensions are flattened to form the second - dimension of the final matrix (width of the matrix). - For example, suppose `X` is a 6-dimensional tensor - with a shape [2, 3, 4, 5, 6], and - `num_flatten_dims` = 3. Then, the flattened matrix - will have a shape [2 x 3 x 4, 5 x 6] = [24, 30]. - By default, `num_flatten_dims` is set to 1. + (inclusive, index starts from 1) dimensions will + be flatten to form the first dimension of the + final matrix (height of the matrix), and the rest + `rank(X) - num_flatten_dims` dimensions are + flattened to form the second dimension of the + final matrix (width of the matrix). For example, + suppose `X` is a 6-dimensional tensor with a shape + [2, 3, 4, 5, 6], and `num_flatten_dims` = 3. Then, + the flattened matrix will have a shape + [2 x 3 x 4, 5 x 6] = [24, 30]. By default, + `num_flatten_dims` is set to 1. param_attr(ParamAttr|list): The parameter attribute for learnable parameters/weights of the fully connected layer. @@ -147,15 +163,14 @@ def fc(input, param_shape = [ reduce(lambda a, b: a * b, input_shape[num_flatten_dims:], 1) ] + [size] + w = helper.create_parameter( attr=param_attr, shape=param_shape, dtype=dtype, is_bias=False) tmp = helper.create_tmp_variable(dtype) helper.append_op( type="mul", - inputs={ - "X": input_var, - "Y": w, - }, + inputs={"X": input_var, + "Y": w}, outputs={"Out": tmp}, attrs={"x_num_col_dims": num_flatten_dims, "y_num_col_dims": 1}) @@ -174,22 +189,35 @@ def fc(input, return helper.append_activation(pre_activation) -def embedding(input, size, is_sparse=False, param_attr=None, dtype='float32'): +def embedding(input, + size, + is_sparse=False, + padding_idx=None, + param_attr=None, + dtype='float32'): """ **Embedding Layer** - This layer is used to lookup a vector of IDs, provided by *input*, in a lookup table. - The result of this lookup is the embedding of each ID in the *input*. + This layer is used to lookup embeddings of IDs, provided by :attr:`input`, in + a lookup table. The result of this lookup is the embedding of each ID in the + :attr:`input`. All the input variables are passed in as local variables to the LayerHelper constructor. Args: - input(Variable): Input to the function - size(tuple|list|None): Shape of the look up table parameter - is_sparse(bool): Boolean flag that specifying whether the input is sparse - param_attr(ParamAttr): Parameters for this layer - dtype(np.dtype|core.DataType|str): The type of data : float32, float_16, int etc + input(Variable): The tensor variable containing the IDs. + size(tuple|list): The shape of the look up table parameter. It should + have two elements which indicate the size of the dictionary of + embeddings and the size of each embedding vector respectively. + is_sparse(bool): The flag indicating whether to use sparse update. + padding_idx(int|long|None): If :attr:`None`, it makes no effect to lookup. + Otherwise the given :attr:`padding_idx` indicates padding the output + with zeros whenever lookup encounters it in :attr:`input`. If + :math:`padding_idx < 0`, the padding_idx to use in lookup is + :math:`size[0] + dim`. + param_attr(ParamAttr): Parameters for this layer + dtype(np.dtype|core.DataType|str): The type of data : float32, float_16, int etc Returns: Variable: The tensor variable storing the embeddings of the \ @@ -207,12 +235,15 @@ def embedding(input, size, is_sparse=False, param_attr=None, dtype='float32'): w = helper.create_parameter( attr=helper.param_attr, shape=size, dtype=dtype, is_bias=False) tmp = helper.create_tmp_variable(dtype) + padding_idx = -1 if padding_idx is None else padding_idx if padding_idx >= 0 else ( + size[0] + padding_idx) helper.append_op( type='lookup_table', inputs={'Ids': input, 'W': w}, outputs={'Out': tmp}, - attrs={'is_sparse': is_sparse}) + attrs={'is_sparse': is_sparse, + 'padding_idx': padding_idx}) return tmp @@ -226,7 +257,107 @@ def dynamic_lstm(input, gate_activation='sigmoid', cell_activation='tanh', candidate_activation='tanh', - dtype='float32'): + dtype='float32', + name=None): + """ + **Dynamic LSTM Layer** + + The defalut implementation is diagonal/peephole connection + (https://arxiv.org/pdf/1402.1128.pdf), the formula is as follows: + + .. math:: + + i_t & = \sigma(W_{ix}x_{t} + W_{ih}h_{t-1} + W_{ic}c_{t-1} + b_i) + + f_t & = \sigma(W_{fx}x_{t} + W_{fh}h_{t-1} + W_{fc}c_{t-1} + b_f) + + \\tilde{c_t} & = act_g(W_{cx}x_t + W_{ch}h_{t-1} + b_c) + + o_t & = \sigma(W_{ox}x_{t} + W_{oh}h_{t-1} + W_{oc}c_t + b_o) + + c_t & = f_t \odot c_{t-1} + i_t \odot \\tilde{c_t} + + h_t & = o_t \odot act_h(c_t) + + where the :math:`W` terms denote weight matrices (e.g. :math:`W_{xi}` is + the matrix of weights from the input gate to the input), :math:`W_{ic}, \ + W_{fc}, W_{oc}` are diagonal weight matrices for peephole connections. In + our implementation, we use vectors to reprenset these diagonal weight + matrices. The :math:`b` terms denote bias vectors (:math:`b_i` is the input + gate bias vector), :math:`\sigma` is the non-linear activations, such as + logistic sigmoid function, and :math:`i, f, o` and :math:`c` are the input + gate, forget gate, output gate, and cell activation vectors, respectively, + all of which have the same size as the cell output activation vector :math:`h`. + + The :math:`\odot` is the element-wise product of the vectors. :math:`act_g` + and :math:`act_h` are the cell input and cell output activation functions + and `tanh` is usually used for them. :math:`\\tilde{c_t}` is also called + candidate hidden state, which is computed based on the current input and + the previous hidden state. + + Set `use_peepholes` to `False` to disable peephole connection. The formula + is omitted here, please refer to the paper + http://www.bioinf.jku.at/publications/older/2604.pdf for details. + + Note that these :math:`W_{xi}x_{t}, W_{xf}x_{t}, W_{xc}x_{t}, W_{xo}x_{t}` + operations on the input :math:`x_{t}` are NOT included in this operator. + Users can choose to use fully-connect layer before LSTM layer. + + Args: + input(Variable): The input of dynamic_lstm layer, which supports + variable-time length input sequence. The underlying + tensor in this Variable is a matrix with shape + (T X 4D), where T is the total time steps in this + mini-batch, D is the hidden size. + size(int): 4 * hidden size. + param_attr(ParamAttr|None): The parameter attribute for the learnable + hidden-hidden weights. + + - Weights = {:math:`W_{ch}, W_{ih}, \ + W_{fh}, W_{oh}`} + - The shape is (D x 4D), where D is the hidden + size. + bias_attr(ParamAttr|None): The bias attribute for the learnable bias + weights, which contains two parts, input-hidden + bias weights and peephole connections weights if + setting `use_peepholes` to `True`. + + 1. `use_peepholes = False` + - Biases = {:math:`b_c, b_i, b_f, b_o`}. + - The shape is (1 x 4D). + 2. `use_peepholes = True` + - Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic}, \ + W_{fc}, W_{oc}`}. + - The shape is (1 x 7D). + use_peepholes(bool): Whether to enable diagonal/peephole connections, + default `True`. + is_reverse(bool): Whether to compute reversed LSTM, default `False`. + gate_activation(str): The activation for input gate, forget gate and + output gate. Choices = ["sigmoid", "tanh", "relu", + "identity"], default "sigmoid". + cell_activation(str): The activation for cell output. Choices = ["sigmoid", + "tanh", "relu", "identity"], default "tanh". + candidate_activation(str): The activation for candidate hidden state. + Choices = ["sigmoid", "tanh", "relu", "identity"], + default "tanh". + dtype(str): Data type. Choices = ["float32", "float64"], default "float32". + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. + + Returns: + tuple: The hidden state, and cell state of LSTM. The shape of both \ + is (T x D), and lod is the same with the `input`. + + Examples: + .. code-block:: python + + hidden_dim = 512 + forward_proj = fluid.layers.fc(input=input_seq, size=hidden_dim * 4, + act=None, bias_attr=None) + forward, _ = fluid.layers.dynamic_lstm( + input=forward_proj, size=hidden_dim * 4, use_peepholes=False) + """ + helper = LayerHelper('lstm', **locals()) size = size / 4 weight = helper.create_parameter( @@ -263,6 +394,299 @@ def dynamic_lstm(input, return hidden, cell +def dynamic_lstmp(input, + size, + proj_size, + param_attr=None, + bias_attr=None, + use_peepholes=True, + is_reverse=False, + gate_activation='sigmoid', + cell_activation='tanh', + candidate_activation='tanh', + proj_activation='tanh', + dtype='float32', + name=None): + """ + **Dynamic LSTMP Layer** + + LSTMP (LSTM with recurrent projection) layer has a separate projection + layer after the LSTM layer, projecting the original hidden state to a + lower-dimensional one, which is proposed to reduce the number of total + parameters and furthermore computational complexity for the LSTM, + espeacially for the case that the size of output units is relative + large (https://research.google.com/pubs/archive/43905.pdf). + + The formula is as follows: + + .. math:: + + i_t & = \sigma(W_{ix}x_{t} + W_{ir}r_{t-1} + W_{ic}c_{t-1} + b_i) + + f_t & = \sigma(W_{fx}x_{t} + W_{fr}r_{t-1} + W_{fc}c_{t-1} + b_f) + + \\tilde{c_t} & = act_g(W_{cx}x_t + W_{cr}r_{t-1} + b_c) + + o_t & = \sigma(W_{ox}x_{t} + W_{or}r_{t-1} + W_{oc}c_t + b_o) + + c_t & = f_t \odot c_{t-1} + i_t \odot \\tilde{c_t} + + h_t & = o_t \odot act_h(c_t) + + r_t & = \overline{act_h}(W_{rh}h_t) + + In the above formula: + + * :math:`W`: Denotes weight matrices (e.g. :math:`W_{xi}` is \ + the matrix of weights from the input gate to the input). + * :math:`W_{ic}`, :math:`W_{fc}`, :math:`W_{oc}`: Diagonal weight \ + matrices for peephole connections. In our implementation, \ + we use vectors to reprenset these diagonal weight matrices. + * :math:`b`: Denotes bias vectors (e.g. :math:`b_i` is the input gate \ + bias vector). + * :math:`\sigma`: The activation, such as logistic sigmoid function. + * :math:`i, f, o` and :math:`c`: The input gate, forget gate, output \ + gate, and cell activation vectors, respectively, all of which have \ + the same size as the cell output activation vector :math:`h`. + * :math:`h`: The hidden state. + * :math:`r`: The recurrent projection of the hidden state. + * :math:`\\tilde{c_t}`: The candidate hidden state, whose \ + computation is based on the current input and previous hidden state. + * :math:`\odot`: The element-wise product of the vectors. + * :math:`act_g` and :math:`act_h`: The cell input and cell output \ + activation functions and `tanh` is usually used for them. + * :math:`\overline{act_h}`: The activation function for the projection \ + output, usually using `identity` or same as :math:`act_h`. + + Set `use_peepholes` to `False` to disable peephole connection. The formula + is omitted here, please refer to the paper + http://www.bioinf.jku.at/publications/older/2604.pdf for details. + + Note that these :math:`W_{xi}x_{t}, W_{xf}x_{t}, W_{xc}x_{t}, W_{xo}x_{t}` + operations on the input :math:`x_{t}` are NOT included in this operator. + Users can choose to use fully-connected layer before LSTMP layer. + + Args: + input(Variable): The input of dynamic_lstmp layer, which supports + variable-time length input sequence. The underlying + tensor in this Variable is a matrix with shape + (T X 4D), where T is the total time steps in this + mini-batch, D is the hidden size. + size(int): 4 * hidden size. + proj_size(int): The size of projection output. + param_attr(ParamAttr|None): The parameter attribute for the learnable + hidden-hidden weight and projection weight. + + - Hidden-hidden weight = {:math:`W_{ch}, W_{ih}, \ + W_{fh}, W_{oh}`}. + - The shape of hidden-hidden weight is (P x 4D), + where P is the projection size and D the hidden + size. + - Projection weight = {:math:`W_{rh}`}. + - The shape of projection weight is (D x P). + bias_attr(ParamAttr|None): The bias attribute for the learnable bias + weights, which contains two parts, input-hidden + bias weights and peephole connections weights if + setting `use_peepholes` to `True`. + + 1. `use_peepholes = False` + - Biases = {:math:`b_c, b_i, b_f, b_o`}. + - The shape is (1 x 4D). + 2. `use_peepholes = True` + - Biases = { :math:`b_c, b_i, b_f, b_o, W_{ic}, \ + W_{fc}, W_{oc}`}. + - The shape is (1 x 7D). + use_peepholes(bool): Whether to enable diagonal/peephole connections, + default `True`. + is_reverse(bool): Whether to compute reversed LSTM, default `False`. + gate_activation(str): The activation for input gate, forget gate and + output gate. Choices = ["sigmoid", "tanh", "relu", + "identity"], default "sigmoid". + cell_activation(str): The activation for cell output. Choices = ["sigmoid", + "tanh", "relu", "identity"], default "tanh". + candidate_activation(str): The activation for candidate hidden state. + Choices = ["sigmoid", "tanh", "relu", "identity"], + default "tanh". + proj_activation(str): The activation for projection output. + Choices = ["sigmoid", "tanh", "relu", "identity"], + default "tanh". + dtype(str): Data type. Choices = ["float32", "float64"], default "float32". + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. + + Returns: + tuple: The projection of hidden state, and cell state of LSTMP. The \ + shape of projection is (T x P), for the cell state which is \ + (T x D), and both LoD is the same with the `input`. + + Examples: + .. code-block:: python + + hidden_dim, proj_dim = 512, 256 + fc_out = fluid.layers.fc(input=input_seq, size=hidden_dim * 4, + act=None, bias_attr=None) + proj_out, _ = fluid.layers.dynamic_lstmp(input=fc_out, + size=hidden_dim * 4, + proj_size=proj_dim, + use_peepholes=False, + is_reverse=True, + cell_activation="tanh", + proj_activation="tanh") + """ + + helper = LayerHelper('lstmp', **locals()) + size = size / 4 + weight = helper.create_parameter( + attr=helper.param_attr, shape=[proj_size, 4 * size], dtype=dtype) + proj_weight = helper.create_parameter( + attr=helper.param_attr, shape=[size, proj_size], dtype=dtype) + bias_size = [1, 7 * size] + if not use_peepholes: + bias_size[1] = 4 * size + bias = helper.create_parameter( + attr=helper.bias_attr, shape=bias_size, dtype=dtype, is_bias=True) + + projection = helper.create_tmp_variable(dtype) + cell = helper.create_tmp_variable(dtype) + ordered_proj0 = helper.create_tmp_variable(dtype) + batch_hidden = helper.create_tmp_variable(dtype) + batch_gate = helper.create_tmp_variable(dtype) + batch_cell_pre_act = helper.create_tmp_variable(dtype) + + helper.append_op( + type='lstmp', + inputs={ + 'Input': input, + 'Weight': weight, + 'ProjWeight': proj_weight, + 'Bias': bias + }, + outputs={ + 'Projection': projection, + 'Cell': cell, + 'OrderedP0': ordered_proj0, + 'BatchHidden': batch_hidden, + 'BatchGate': batch_gate, + 'BatchCellPreAct': batch_cell_pre_act + }, + attrs={ + 'use_peepholes': use_peepholes, + 'is_reverse': is_reverse, + 'gate_activation': gate_activation, + 'cell_activation': cell_activation, + 'candidate_activation': candidate_activation, + 'proj_activation': proj_activation + }) + return projection, cell + + +def dynamic_gru(input, + size, + param_attr=None, + bias_attr=None, + is_reverse=False, + gate_activation='sigmoid', + candidate_activation='tanh', + h_0=None): + """ + **Dynamic GRU Layer** + + Refer to `Empirical Evaluation of Gated Recurrent Neural Networks on + Sequence Modeling `_ + + The formula is as follows: + + .. math:: + + u_t & = act_g(W_{ux}x_{t} + W_{uh}h_{t-1} + b_u) + + r_t & = act_g(W_{rx}x_{t} + W_{rh}h_{t-1} + b_r) + + \\tilde{h_t} & = act_c(W_{cx}x_{t} + W_{ch}(r_t \odot h_{t-1}) + b_c) + + h_t & = (1-u_t) \odot h_{t-1} + u_t \odot \\tilde{h_t} + + The :math:`\odot` is the element-wise product of the vectors. :math:`act_g` + is the update gate and reset gate activation function and :math:`sigmoid` + is usually used for it. :math:`act_c` is the activation function for + candidate hidden state and :math:`tanh` is usually used for it. + + Note that these :math:`W_{ux}x_{t}, W_{rx}x_{t}, W_{cx}x_{t}` operations on + the input :math:`x_{t}` are NOT included in this operator. Users can choose + to use fully-connect layer before GRU layer. + + Args: + input(Variable): The input of dynamic_gru layer, which supports + variable-time length input sequence. The underlying tensor in this + Variable is a matrix with shape :math:`(T \\times 3D)`, where + :math:`T` is the total time steps in this mini-batch, :math:`D` + is the hidden size. + size(int): The dimension of the gru cell. + param_attr(ParamAttr|None): The parameter attribute for the learnable + hidden-hidden weight matrix. Note: + + - The shape of the weight matrix is :math:`(T \\times 3D)`, where + :math:`D` is the hidden size. + - All elements in the weight matrix can be divided into two parts. + The first part are weights of the update gate and reset gate with + shape :math:`(D \\times 2D)`, and the second part are weights for + candidate hidden state with shape :math:`(D \\times D)`. + bias_attr(ParamAttr): The parameter attribute for learnable the + hidden-hidden bias. + is_reverse(bool): Whether to compute reversed GRU, default + :attr:`False`. + gate_activation(str): The activation for update gate and reset gate. + Choices = ["sigmoid", "tanh", "relu", "identity"], default "sigmoid". + activation(str): The activation for candidate hidden state. + Choices = ["sigmoid", "tanh", "relu", "identity"], default "tanh". + + Returns: + Variable: The hidden state of GRU. The shape is (T \\times D), and lod \ + is the same with the input. + + Examples: + .. code-block:: python + + hidden_dim = 512 + x = fluid.layers.fc(input=data, size=hidden_dim * 3) + hidden = fluid.layers.dynamic_gru(input=x, dim=hidden_dim) + """ + + helper = LayerHelper('gru', **locals()) + dtype = helper.input_dtype() + + weight = helper.create_parameter( + attr=helper.param_attr, shape=[size, 3 * size], dtype=dtype) + bias = helper.create_parameter( + attr=helper.bias_attr, shape=[1, 3 * size], dtype=dtype, is_bias=True) + inputs = {'Input': input, 'Weight': weight, 'Bias': bias} + if h_0 != None: + assert h_0.shape == ( + size, size), 'The shape of h0 should be(%d, %d)' % (size, size) + inputs['h0'] = h_0 + + hidden = helper.create_tmp_variable(dtype) + batch_gate = helper.create_tmp_variable(dtype) + batch_reset_hidden_prev = helper.create_tmp_variable(dtype) + batch_hidden = helper.create_tmp_variable(dtype) + + helper.append_op( + type='gru', + inputs=inputs, + outputs={ + 'Hidden': hidden, + 'BatchGate': batch_gate, + 'BatchResetHiddenPrev': batch_reset_hidden_prev, + 'BatchHidden': batch_hidden + }, + attrs={ + 'is_reverse': is_reverse, + 'gate_activation': gate_activation, + 'activation': candidate_activation + }) + return hidden + + def gru_unit(input, hidden, size, @@ -300,8 +724,10 @@ def gru_unit(input, size (integer): The input dimension value. weight (ParamAttr): The weight parameters for gru unit. Default: None bias (ParamAttr): The bias parameters for gru unit. Default: None - activation (string): The activation type for cell (actNode). Default: 'tanh' - gate_activation (string): The activation type for gates (actGate). Default: 'sigmoid' + activation (string): The activation type for cell (actNode). + Default: 'tanh' + gate_activation (string): The activation type for gates (actGate). + Default: 'sigmoid' Returns: tuple: The hidden value, reset-hidden value and gate values. @@ -421,7 +847,35 @@ def cos_sim(X, Y, **kwargs): return out -def dropout(x, dropout_prob, is_test=False, seed=0, **kwargs): +def dropout(x, dropout_prob, is_test=False, seed=None, **kwargs): + """ + Computes dropout. + + Drop or keep each element of `x` independently. Dropout is a regularization + technique for reducing overfitting by preventing neuron co-adaption during + training. The dropout operator randomly set (according to the given dropout + probability) the outputs of some units to zero, while others are remain + unchanged. + + Args: + x(variable): The input tensor. + dropout_prob(float): Probability of setting units to zero. + is_test(bool): A flag indicating whether it is in test phrase or not. + seed(int): A Python integer used to create random seeds. If this + parameter is set to None, a random seed is used. + NOTE: If an integer seed is given, always the same output + units will be dropped. DO NOT use a fixed seed in training. + + Returns: + Variable: A tensor variable. + + Examples: + .. code-block:: python + + x = fluid.layers.data(name="data", shape=[32, 32], dtype="float32") + droped = fluid.layers.dropout(input=x, dropout_rate=0.5) + """ + helper = LayerHelper('dropout', **kwargs) out = helper.create_tmp_variable(dtype=x.dtype) mask = helper.create_tmp_variable(dtype=x.dtype, stop_gradient=True) @@ -430,9 +884,12 @@ def dropout(x, dropout_prob, is_test=False, seed=0, **kwargs): inputs={'X': [x]}, outputs={'Out': [out], 'Mask': [mask]}, - attrs={'dropout_prob': dropout_prob, - 'is_test': is_test, - 'seed': seed}) + attrs={ + 'dropout_prob': dropout_prob, + 'is_test': is_test, + 'fix_seed': seed is not None, + 'seed': seed if seed is not None else 0 + }) return out @@ -440,8 +897,9 @@ def cross_entropy(input, label, **kwargs): """ **Cross Entropy Layer** - This layer computes the cross entropy between `input` and `label`. It supports - both standard cross-entropy and soft-label cross-entropy loss computation. + This layer computes the cross entropy between `input` and `label`. It + supports both standard cross-entropy and soft-label cross-entropy loss + computation. 1) One-hot cross-entropy: `soft_label = False`, `Label[i, 0]` indicates the class index for sample i: @@ -468,23 +926,28 @@ def cross_entropy(input, label, **kwargs): Args: input (Variable|list): a 2-D tensor with shape [N x D], where N is the - batch size and D is the number of classes. This input is a probability - computed by the previous operator, which is almost always the result - of a softmax operator. + batch size and D is the number of classes. This + input is a probability computed by the previous + operator, which is almost always the result of + a softmax operator. label (Variable|list): the ground truth which is a 2-D tensor. When - `soft_label` is set to `False`, `label` is a tensor with shape - [N x 1]. When `soft_label` is set to `True`, `label` is a - tensor with shape [N x D]. - soft_label (bool, via `**kwargs`): a flag indicating whether to interpretate - the given labels as soft labels, default `False`. + `soft_label` is set to `False`, `label` is a + tensor with shape [N x 1]. When + `soft_label` is set to `True`, `label` is a + tensor with shape [N x D]. + soft_label (bool, via `**kwargs`): a flag indicating whether to + interpretate the given labels as soft + labels, default `False`. Returns: A 2-D tensor with shape [N x 1], the cross entropy loss. Raises: - `ValueError`: 1) the 1st dimension of `input` and `label` are not equal; 2) when \ - `soft_label == True`, and the 2nd dimension of `input` and `label` are not \ - equal; 3) when `soft_label == False`, and the 2nd dimension of `label` is not 1. + `ValueError`: 1) the 1st dimension of `input` and `label` are not equal. + 2) when `soft_label == True`, and the 2nd dimension of + `input` and `label` are not equal. + 3) when `soft_label == False`, and the 2nd dimension of + `label` is not 1. Examples: .. code-block:: python @@ -507,7 +970,9 @@ def square_error_cost(input, label, **kwargs): """ **Square error cost layer** - This layer accepts input predictions and target label and returns the squared error cost. + This layer accepts input predictions and target label and returns the + squared error cost. + For predictions, :math:`X`, and target labels, :math:`Y`, the equation is: .. math:: @@ -525,8 +990,8 @@ def square_error_cost(input, label, **kwargs): label(Variable): Label tensor, has target labels. Returns: - Variable: The tensor variable storing the element-wise squared error difference \ - of input and label. + Variable: The tensor variable storing the element-wise squared error + difference of input and label. Examples: .. code-block:: python @@ -622,7 +1087,8 @@ def chunk_eval(input, "chunk_scheme": chunk_scheme, "excluded_chunk_types": excluded_chunk_types or [] }) - return precision, recall, f1_score, num_infer_chunks, num_label_chunks, num_correct_chunks + return (precision, recall, f1_score, num_infer_chunks, num_label_chunks, + num_correct_chunks) def sequence_conv(input, @@ -674,20 +1140,22 @@ def conv2d(input, groups=None, param_attr=None, bias_attr=None, + use_cudnn=True, act=None): """ **Convlution2D Layer** The convolution2D layer calculates the output based on the input, filter - and strides, paddings, dilations, groups parameters. Input(Input) and Output(Output) - are in NCHW format. Where N is batch size, C is the number of channels, H is the height - of the feature, and W is the width of the feature. + and strides, paddings, dilations, groups parameters. Input(Input) and + Output(Output) are in NCHW format. Where N is batch size, C is the number of + channels, H is the height of the feature, and W is the width of the feature. The details of convolution layer, please refer UFLDL's `convolution, `_ . - If bias attribution and activation type are provided, bias is added to the output of the convolution, - and the corresponding activation function is applied to the final result. - For each input :math:`X`, the equation is: + If bias attribution and activation type are provided, bias is added to the + output of the convolution, and the corresponding activation function is + applied to the final result. + For each input :math:`X`, the equation is: .. math:: @@ -695,64 +1163,72 @@ def conv2d(input, In the above equation: - * :math:`X`: Input value, a tensor with NCHW format. - * :math:`W`: Filter value, a tensor with MCHW format. - * :math:`\\ast`: Convolution operation. - * :math:`b`: Bias value, a 2-D tensor with shape [M, 1]. - * :math:`\\sigma`: Activation function. - * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different. + * :math:`X`: Input value, a tensor with NCHW format. + * :math:`W`: Filter value, a tensor with MCHW format. + * :math:`\\ast`: Convolution operation. + * :math:`b`: Bias value, a 2-D tensor with shape [M, 1]. + * :math:`\\sigma`: Activation function. + * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be + different. Example: - Input: - Input shape: $(N, C_{in}, H_{in}, W_{in})$ + - Input: + + Input shape: $(N, C_{in}, H_{in}, W_{in})$ - Filter shape: $(C_{out}, C_{in}, H_f, W_f)$ + Filter shape: $(C_{out}, C_{in}, H_f, W_f)$ + + - Output: + Output shape: $(N, C_{out}, H_{out}, W_{out})$ - Output: - Output shape: $(N, C_{out}, H_{out}, W_{out})$ Where - .. math:: + + .. math:: H_{out}&= \\frac{(H_{in} + 2 * paddings[0] - (dilations[0] * (H_f - 1) + 1))}{strides[0]} + 1 \\\\ W_{out}&= \\frac{(W_{in} + 2 * paddings[1] - (dilations[1] * (W_f - 1) + 1))}{strides[1]} + 1 Args: - input(Variable): The input image with [N, C, H, W] format. - num_filters(int): The number of filter. It is as same as the output - image channel. - filter_size(int|tuple|None): The filter size. If filter_size is a tuple, - it must contain two integers, (filter_size_H, filter_size_W). - Otherwise, the filter will be a square. - stride(int|tuple): The stride size. If stride is a tuple, it must - contain two integers, (stride_H, stride_W). Otherwise, the - stride_H = stride_W = stride. Default: stride = 1. - padding(int|tuple): The padding size. If padding is a tuple, it must - contain two integers, (padding_H, padding_W). Otherwise, the - padding_H = padding_W = padding. Default: padding = 0. - groups(int): The groups number of the Conv2d Layer. According to grouped - convolution in Alex Krizhevsky's Deep CNN paper: when group=2, - the first half of the filters is only connected to the first half - of the input channels, while the second half of the filters is only - connected to the second half of the input channels. Default: groups=1 - param_attr(ParamAttr): The parameters to the Conv2d Layer. Default: None - bias_attr(ParamAttr): Bias parameter for the Conv2d layer. Default: None - act(str): Activation type. Default: None + input(Variable): The input image with [N, C, H, W] format. + num_filters(int): The number of filter. It is as same as the output + image channel. + filter_size(int|tuple|None): The filter size. If filter_size is a tuple, + it must contain two integers, (filter_size_H, filter_size_W). + Otherwise, the filter will be a square. + stride(int|tuple): The stride size. If stride is a tuple, it must + contain two integers, (stride_H, stride_W). Otherwise, the + stride_H = stride_W = stride. Default: stride = 1. + padding(int|tuple): The padding size. If padding is a tuple, it must + contain two integers, (padding_H, padding_W). Otherwise, the + padding_H = padding_W = padding. Default: padding = 0. + groups(int): The groups number of the Conv2d Layer. According to grouped + convolution in Alex Krizhevsky's Deep CNN paper: when group=2, + the first half of the filters is only connected to the first half + of the input channels, while the second half of the filters is only + connected to the second half of the input channels. Default: groups=1 + param_attr(ParamAttr): The parameters to the Conv2d Layer. Default: None + bias_attr(ParamAttr): Bias parameter for the Conv2d layer. Default: None + use_cudnn(bool): Use cudnn kernel or not, it is valid only when the cudnn + library is installed. Default: True + act(str): Activation type. Default: None Returns: - Variable: The tensor variable storing the convolution and \ + Variable: The tensor variable storing the convolution and non-linearity activation result. Raises: - ValueError: If the shapes of input, filter_size, stride, padding and groups mismatch. + ValueError: If the shapes of input, filter_size, stride, padding and + groups mismatch. Examples: .. code-block:: python - data = fluid.layers.data(name='data', shape=[3, 32, 32], dtype='float32') - conv2d = fluid.layers.conv2d(input=data, num_filters=2, filter_size=3, act="relu") + data = fluid.layers.data( + name='data', shape=[3, 32, 32], dtype='float32') + conv2d = fluid.layers.conv2d( + input=data, num_filters=2, filter_size=3, act="relu") """ - if stride is None: stride = [1, 1] helper = LayerHelper('conv2d', **locals()) @@ -772,6 +1248,8 @@ def conv2d(input, stride = [stride, stride] if isinstance(padding, int): padding = [padding, padding] + if not isinstance(use_cudnn, bool): + raise ValueError("use_cudnn should be True or False") input_shape = input.shape filter_shape = [num_filters, num_filter_channels] + filter_size @@ -795,9 +1273,12 @@ def conv2d(input, 'Filter': filter_param, }, outputs={"Output": pre_bias}, - attrs={'strides': stride, - 'paddings': padding, - 'groups': groups}) + attrs={ + 'strides': stride, + 'paddings': padding, + 'groups': groups, + 'use_cudnn': use_cudnn + }) pre_act = helper.append_bias_op(pre_bias, dim_start=1, dim_end=2) @@ -945,7 +1426,9 @@ def pool2d(input, pool_type, pool_stride=None, pool_padding=None, - global_pooling=False): + global_pooling=False, + use_cudnn=True, + name=None): """ This function adds the operator for pooling in 2 dimensions, using the pooling configurations mentioned in input parameters. @@ -964,6 +1447,8 @@ def pool2d(input, pool_stride = [pool_stride, pool_stride] if isinstance(pool_padding, int): pool_padding = [pool_padding, pool_padding] + if not isinstance(use_cudnn, bool): + raise ValueError("use_cudnn should be True or False") helper = LayerHelper('pool2d', **locals()) dtype = helper.input_dtype() @@ -978,7 +1463,8 @@ def pool2d(input, "ksize": pool_size, "global_pooling": global_pooling, "strides": pool_stride, - "paddings": pool_padding + "paddings": pool_padding, + "use_cudnn": use_cudnn }) return pool_out @@ -991,7 +1477,8 @@ def batch_norm(input, epsilon=1e-05, param_attr=None, bias_attr=None, - data_layout='NCHW'): + data_layout='NCHW', + name=None): """ This function helps create an operator to implement the BatchNorm layer using the configurations from the input parameters. @@ -1067,7 +1554,7 @@ def batch_norm(input, return helper.append_activation(batch_norm_out) -def beam_search_decode(ids, scores): +def beam_search_decode(ids, scores, name=None): helper = LayerHelper('beam_search_decode', **locals()) sentence_ids = helper.create_tmp_variable(dtype=ids.dtype) sentence_scores = helper.create_tmp_variable(dtype=ids.dtype) @@ -1091,38 +1578,95 @@ def conv2d_transpose(input, padding=None, stride=None, dilation=None, - param_attr=None): + param_attr=None, + use_cudnn=True, + name=None): """ - The transpose of conv2d layer. + **Convlution2D transpose layer** + + The convolution2D transpose layer calculates the output based on the input, + filter, and dilations, strides, paddings. Input(Input) and output(Output) + are in NCHW format. Where N is batch size, C is the number of channels, + H is the height of the feature, and W is the width of the feature. + Parameters(dilations, strides, paddings) are two elements. These two elements + represent height and width, respectively. The details of convolution transpose + layer, please refer to the following explanation and references + `therein `_. + + For each input :math:`X`, the equation is: + + .. math:: + + Out = W \\ast X + + In the above equation: + + * :math:`X`: Input value, a tensor with NCHW format. + * :math:`W`: Filter value, a tensor with MCHW format. + * :math:`\\ast` : Convolution transpose operation. + * :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be + different. + + Example: + + - Input: + + Input shape: $(N, C_{in}, H_{in}, W_{in})$ - This layer is also known as deconvolution layer. + Filter shape: $(C_{in}, C_{out}, H_f, W_f)$ + + - Output: + + Output shape: $(N, C_{out}, H_{out}, W_{out})$ + + Where + + .. math:: + + H_{out} &= (H_{in} - 1) * strides[0] - 2 * paddings[0] + dilations[0] * (H_f - 1) + 1 \\\\ + W_{out} &= (W_{in} - 1) * strides[1] - 2 * paddings[1] + dilations[1] * (W_f - 1) + 1 Args: - input(Variable): The input image with [N, C, H, W] format. - num_filters(int): The number of filter. It is as same as the output - image channel. - output_size(int|tuple|None): The output image size. If output size is a - tuple, it must contain two integers, (image_H, image_W). This - parameter only works when filter_size is None. - filter_size(int|tuple|None): The filter size. If filter_size is a tuple, - it must contain two integers, (filter_size_H, filter_size_W). - Otherwise, the filter will be a square. None if use output size to - calculate filter_size - padding(int|tuple): The padding size. If padding is a tuple, it must - contain two integers, (padding_H, padding_W). Otherwise, the - padding_H = padding_W = padding. - stride(int|tuple): The stride size. If stride is a tuple, it must - contain two integers, (stride_H, stride_W). Otherwise, the - stride_H = stride_W = stride. - dilation(int|tuple): The dilation size. If dilation is a tuple, it must - contain two integers, (dilation_H, dilation_W). Otherwise, the - dilation_H = dilation_W = dilation. - param_attr: Parameter Attribute. - main_program(Program): the main program - startup_program(Program): the startup program + input(Variable): The input image with [N, C, H, W] format. + num_filters(int): The number of the filter. It is as same as the output + image channel. + output_size(int|tuple|None): The output image size. If output size is a + tuple, it must contain two integers, (image_H, image_W). This + parameter only works when filter_size is None. + filter_size(int|tuple|None): The filter size. If filter_size is a tuple, + it must contain two integers, (filter_size_H, filter_size_W). + Otherwise, the filter will be a square. None if use output size to + calculate filter_size. + padding(int|tuple): The padding size. If padding is a tuple, it must + contain two integers, (padding_H, padding_W). Otherwise, the + padding_H = padding_W = padding. Default: padding = 0. + stride(int|tuple): The stride size. If stride is a tuple, it must + contain two integers, (stride_H, stride_W). Otherwise, the + stride_H = stride_W = stride. Default: stride = 1. + dilation(int|tuple): The dilation size. If dilation is a tuple, it must + contain two integers, (dilation_H, dilation_W). Otherwise, the + dilation_H = dilation_W = dilation. Default: dilation = 1. + param_attr(ParamAttr): The parameters to the Conv2d_transpose Layer. + Default: None + use_cudnn(bool): Use cudnn kernel or not, it is valid only when the cudnn + library is installed. Default: True + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: - Variable: Output image. + Variable: The tensor variable storing the convolution transpose result. + + Raises: + ValueError: If the shapes of input, filter_size, stride, padding and + groups mismatch. + + Examples: + .. code-block:: python + + data = fluid.layers.data( + name='data', shape=[3, 32, 32], dtype='float32') + conv2d_transpose = fluid.layers.conv2d_transpose( + input=data, num_filters=2, filter_size=3) """ helper = LayerHelper("conv2d_transpose", **locals()) if not isinstance(input, Variable): @@ -1146,6 +1690,10 @@ def conv2d_transpose(input, elif dilation is not None: op_attr['dilations'] = dilation + if not isinstance(use_cudnn, bool): + raise ValueError("use_cudnn should be True or False") + op_attr['use_cudnn'] = use_cudnn + if filter_size is None: if output_size is None: raise ValueError("output_size must be set when filter_size is None") @@ -1183,7 +1731,7 @@ def conv2d_transpose(input, return out -def sequence_expand(x, y): +def sequence_expand(x, y, name=None): """Sequence Expand Layer. This layer will expand the input variable **x** according to LoD information of **y**. And the following examples will explain how sequence_expand works: @@ -1227,6 +1775,8 @@ def sequence_expand(x, y): Args: x (Variable): The input variable which is a Tensor or LoDTensor. y (Variable): The input variable which is a LoDTensor. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: Variable: The expanded variable which is a LoDTensor. @@ -1248,12 +1798,45 @@ def sequence_expand(x, y): return tmp +def beam_search(pre_ids, ids, scores, beam_size, end_id, level=0): + ''' + This function implements the beam search algorithm. + ''' + helper = LayerHelper('beam_search', **locals()) + score_type = scores.dtype + id_type = ids.dtype + + selected_scores = helper.create_tmp_variable(dtype=score_type) + selected_ids = helper.create_tmp_variable(dtype=id_type) + + helper.append_op( + type='beam_search', + inputs={ + 'pre_ids': pre_ids, + 'ids': ids, + 'scores': scores, + }, + outputs={ + 'selected_ids': selected_ids, + 'selected_scores': selected_scores, + }, + attrs={ + # TODO(ChunweiYan) to assure other value support + 'level': level, + 'beam_size': beam_size, + 'end_id': end_id, + }) + + return selected_ids, selected_scores + + def lstm_unit(x_t, hidden_t_prev, cell_t_prev, forget_bias=0.0, param_attr=None, - bias_attr=None): + bias_attr=None, + name=None): """Lstm unit layer. The equation of a lstm step is: .. math:: @@ -1300,15 +1883,17 @@ def lstm_unit(x_t, initializer, name etc. bias_attr (ParamAttr): The attributes of bias weights, if not False, bias weights will be created and be set to default value. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: tuple: The hidden value and cell value of lstm unit. Raises: - ValueError: The ranks of **x_t**, **hidden_t_prev** and **cell_t_prev**\ - not be 2 or the 1st dimensions of **x_t**, **hidden_t_prev** \ - and **cell_t_prev** not be the same or the 2nd dimensions of \ - **hidden_t_prev** and **cell_t_prev** not be the same. + ValueError: The ranks of **x_t**, **hidden_t_prev** and **cell_t_prev** + not be 2 or the 1st dimensions of **x_t**, **hidden_t_prev** + and **cell_t_prev** not be the same or the 2nd dimensions of + **hidden_t_prev** and **cell_t_prev** not be the same. Examples: @@ -1365,7 +1950,7 @@ def lstm_unit(x_t, return h, c -def reduce_sum(input, dim=None, keep_dim=False): +def reduce_sum(input, dim=None, keep_dim=False, name=None): """ Computes the sum of tensor elements over the given dimension. @@ -1379,6 +1964,8 @@ def reduce_sum(input, dim=None, keep_dim=False): keep_dim (bool): Whether to reserve the reduced dimension in the output Tensor. The result tensor will have one fewer dimension than the :attr:`input` unless :attr:`keep_dim` is true. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: Variable: The reduced Tensor variable. @@ -1409,7 +1996,7 @@ def reduce_sum(input, dim=None, keep_dim=False): return out -def reduce_mean(input, dim=None, keep_dim=False): +def reduce_mean(input, dim=None, keep_dim=False, name=None): """ Computes the mean of tensor elements over the given dimension. @@ -1423,6 +2010,8 @@ def reduce_mean(input, dim=None, keep_dim=False): keep_dim (bool): Whether to reserve the reduced dimension in the output Tensor. The result tensor will have one fewer dimension than the :attr:`input` unless :attr:`keep_dim` is true. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: Variable: The reduced Tensor variable. @@ -1453,7 +2042,7 @@ def reduce_mean(input, dim=None, keep_dim=False): return out -def reduce_max(input, dim=None, keep_dim=False): +def reduce_max(input, dim=None, keep_dim=False, name=None): """ Computes the maximum of tensor elements over the given dimension. @@ -1467,6 +2056,8 @@ def reduce_max(input, dim=None, keep_dim=False): keep_dim (bool): Whether to reserve the reduced dimension in the output Tensor. The result tensor will have one fewer dimension than the :attr:`input` unless :attr:`keep_dim` is true. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: Variable: The reduced Tensor variable. @@ -1497,7 +2088,7 @@ def reduce_max(input, dim=None, keep_dim=False): return out -def reduce_min(input, dim=None, keep_dim=False): +def reduce_min(input, dim=None, keep_dim=False, name=None): """ Computes the minimum of tensor elements over the given dimension. @@ -1511,6 +2102,8 @@ def reduce_min(input, dim=None, keep_dim=False): keep_dim (bool): Whether to reserve the reduced dimension in the output Tensor. The result tensor will have one fewer dimension than the :attr:`input` unless :attr:`keep_dim` is true. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: Variable: The reduced Tensor variable. @@ -1541,20 +2134,22 @@ def reduce_min(input, dim=None, keep_dim=False): return out -def split(input, num_or_sections, dim=-1): +def split(input, num_or_sections, dim=-1, name=None): """ - Splits the tensor into multiple sub-tensors. + Split the input tensor into multiple sub-tensors. Args: input (Variable): The input variable which is a Tensor or LoDTensor. - num_or_sections (int|list): If :attr:`num_or_sections` is an integer, - then the integer indicates the number of equal sized sub-tensors - that the tensor will be divided into. If :attr:`num_or_sections` - is a list of integers, the length of list indicates the number of - sub-tensors and the integers indicate the sizes of sub-tensors' + num_or_sections (int|list): If :attr:`num_or_sections` is an integer, + then the integer indicates the number of equal sized sub-tensors + that the tensor will be divided into. If :attr:`num_or_sections` + is a list of integers, the length of list indicates the number of + sub-tensors and the integers indicate the sizes of sub-tensors' :attr:`dim` dimension orderly. - dim (int): The dimension along which to split. If :math:`dim < 0`, the + dim (int): The dimension along which to split. If :math:`dim < 0`, the dimension to split along is :math:`rank(input) + dim`. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. Returns: List: The list of segmented tensor variables. @@ -1597,3 +2192,793 @@ def split(input, num_or_sections, dim=-1): 'axis': dim }) return outs + + +def l2_normalize(x, axis, epsilon=1e-12, name=None): + """ + **L2 normalize Layer** + + The l2 normalize layer normalizes `x` along dimension `axis` using an L2 + norm. For a 1-D tensor (`dim` is fixed to 0), this layer computes + + output = x / sqrt(max(sum(x**2), epsilon)) + + For `x` with more dimensions, this layer independently normalizes each 1-D + slice along dimension `axis`. + + Args: + x(Variable|list): The input tensor to l2_normalize layer. + axis(int): Dimension along which to normalize the input. + epsilon(float): A lower bound value for `x`'s l2 norm. sqrt(epsilon) will + be used as the divisor if the l2 norm of `x` is less than + sqrt(epsilon). + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. + + + Returns: + Variable: The output tensor variable. + + Examples: + .. code-block:: python + + data = fluid.layers.data(name="data", + shape=(3, 17, 13), + dtype="float32") + normed = fluid.layers.l2_normalize(x=data, axis=1) + """ + + if len(x.shape) == 1: axis = 0 + + helper = LayerHelper("l2_normalize", **locals()) + + square = helper.create_tmp_variable(dtype=x.dtype) + helper.append_op(type="square", inputs={"X": x}, outputs={"Out": square}) + + reduced_sum = helper.create_tmp_variable(dtype=x.dtype) + helper.append_op( + type="reduce_sum", + inputs={"X": square}, + outputs={"Out": reduced_sum}, + attrs={ + "dim": 1 if axis is None else axis, + "keep_dim": True, + "reduce_all": False + }) + + # TODO(caoying) A lower bound value epsilon for the norm is needed to + # imporve the numeric stability of reciprocal. This requires a maximum_op. + rsquare = helper.create_tmp_variable(dtype=x.dtype) + helper.append_op( + type="reciprocal", inputs={"X": reduced_sum}, outputs={"Out": rsquare}) + + # TODO(caoying) the current elementwise_mul operator does not support a + # general broadcast rule which broadcasts input(Y) to have the same + # dimension with Input(X) starting from a specified dimension. So this + # exanpsion is requred. Once a general broadcast rule is spported, this + # expanding canbe removed. + rsquare_expanded = helper.create_tmp_variable(dtype=x.dtype) + expand_times = [1] * len(x.shape) + expand_times[axis] = int(x.shape[axis]) + helper.append_op( + type="expand", + inputs={"X": rsquare}, + outputs={"Out": rsquare_expanded}, + attrs={"expand_times": expand_times}) + + out = helper.create_tmp_variable(dtype=x.dtype) + helper.append_op( + type="elementwise_mul", + inputs={"X": x, + "Y": rsquare_expanded}, + outputs={"Out": out}) + return out + + +def matmul(x, y, transpose_x=False, transpose_y=False, name=None): + """ + Applies matrix multiplication to two tensors. + + Currently, the input tensors' rank can be any, but when the rank of any + inputs is bigger than 3, this two inputs' rank should be equal. + + The actual behavior depends on the shapes of :math:`x`, :math:`y` and the + flag values of :attr:`transpose_x`, :attr:`transpose_y`. Specifically: + + - If a transpose flag is specified, the last two dimensions of the tensor + are transposed. If the tensor is rank-1 of shape :math:`[D]`, then for + :math:`x` it is treated as :math:`[1, D]` in nontransposed form and as + :math:`[D, 1]` in transposed form, whereas for :math:`y` it is the + opposite: It is treated as :math:`[D, 1]` in nontransposed form and as + :math:`[1, D]` in transposed form. + + - After transpose, the two tensors are 2-D or n-D and matrix multiplication + performs in the following way. + + - If both are 2-D, they are multiplied like conventional matrices. + - If either is n-D, it is treated as a stack of matrices residing in the + last two dimensions and a batched matrix multiply supporting broadcast + applies on the two tensors. + + Also note that if the raw tensor :math:`x` or :math:`y` is rank-1 and + nontransposed, the prepended or appended dimension :math:`1` will be + removed after matrix multiplication. + + Args: + x (Variable): The input variable which is a Tensor or LoDTensor. + y (Variable): The input variable which is a Tensor or LoDTensor. + transpose_x (bool): Whether to transpose :math:`x` before multiplication. + transpose_y (bool): Whether to transpose :math:`y` before multiplication. + name(str|None): A name for this layer(optional). If set None, the layer + will be named automatically. + + Returns: + Variable: The product Tensor variable. + + Examples: + .. code-block:: python + + # Examples to clarify shapes of the inputs and output + # x: [B, ..., M, K], y: [B, ..., K, N] + fluid.layers.matmul(x, y) # out: [B, ..., M, N] + + # x: [B, M, K], y: [B, K, N] + fluid.layers.matmul(x, y) # out: [B, M, N] + + # x: [B, M, K], y: [K, N] + fluid.layers.matmul(x, y) # out: [B, M, N] + + # x: [M, K], y: [K, N] + fluid.layers.matmul(x, y) # out: [M, N] + + # x: [B, M, K], y: [K] + fluid.layers.matmul(x, y) # out: [B, M] + + # x: [K], y: [K] + fluid.layers.matmul(x, y) # out: [1] + + # x: [M], y: [N] + fluid.layers.matmul(x, y, True, True) # out: [M, N] + """ + + def __check_input(x, y): + if len(y.shape) > len(x.shape): + raise ValueError( + "Invalid inputs for matmul. " + "x's rank should be always greater than or equal to y'rank.") + + x_shape = list(x.shape) + y_shape = list(y.shape) + if len(x_shape) == 1: + x_shape = [1] + x_shape + if len(y_shape) == 1: + y_shape = y_shape + [1] + + # check the inner 2 dimensions + if transpose_x: + x_shape[-2], x_shape[-1] = x_shape[-1], x_shape[-2] + if transpose_y: + y_shape[-2], y_shape[-1] = y_shape[-1], y_shape[-2] + if x_shape[-1] != y_shape[-2]: + raise ValueError("Invalid inputs for matmul.") + + if len(y_shape) > 2: + for i, dim_x in enumerate(x_shape[:-2]): + if dim_x != y_shape[i]: + raise ValueError("Invalid inputs for matmul.") + + __check_input(x, y) + + helper = LayerHelper('matmul', **locals()) + out = helper.create_tmp_variable(dtype=x.dtype) + helper.append_op( + type='matmul', + inputs={'X': x, + 'Y': y}, + outputs={'Out': out}, + attrs={'transpose_X': transpose_x, + 'transpose_Y': transpose_y}) + return out + + +def edit_distance(input, + label, + normalized=False, + ignored_tokens=None, + name=None): + """ + EditDistance operator computes the edit distances between a batch of + hypothesis strings and their references. Edit distance, also called + Levenshtein distance, measures how dissimilar two strings are by counting + the minimum number of operations to transform one string into anthor. + Here the operations include insertion, deletion, and substitution. + + For example, given hypothesis string A = "kitten" and reference + B = "sitting", the edit distance is 3 for A will be transformed into B + at least after two substitutions and one insertion: + + "kitten" -> "sitten" -> "sittin" -> "sitting" + + Input(Hyps) is a LoDTensor consisting of all the hypothesis strings with + the total number denoted by `batch_size`, and the separation is specified + by the LoD information. And the `batch_size` reference strings are arranged + in order in the same way in the LoDTensor Input(Refs). + + Output(Out) contains the `batch_size` results and each stands for the edit + distance for a pair of strings respectively. If Attr(normalized) is true, + the edit distance will be divided by the length of reference string. + + Args: + + input(Variable): The indices for hypothesis strings. + + label(Variable): The indices for reference strings. + + normalized(bool): Indicated whether to normalize the edit distance by + the length of reference string. + + ignored_tokens(list of int): Tokens that should be removed before + calculating edit distance. + + Returns: + Variable: sequence-to-sequence edit distance in shape [batch_size, 1]. + + Examples: + .. code-block:: python + + x = fluid.layers.data(name='x', shape=[8], dtype='float32') + y = fluid.layers.data(name='y', shape=[7], dtype='float32') + + cost = fluid.layers.edit_distance(input=x,label=y) + """ + helper = LayerHelper("edit_distance", **locals()) + + # remove some tokens from input and labels + if ignored_tokens is not None and len(ignored_tokens) > 0: + erased_input = helper.create_tmp_variable(dtype="int64") + erased_label = helper.create_tmp_variable(dtype="int64") + + helper.append_op( + type="sequence_erase", + inputs={"X": [input]}, + outputs={"Out": [erased_input]}, + attrs={"tokens": ignored_tokens}) + input = erased_input + + helper.append_op( + type="sequence_erase", + inputs={"X": [label]}, + outputs={"Out": [erase_label]}, + attrs={"tokens": ignored_tokens}) + label = erased_label + + # edit distance op + edit_distance_out = helper.create_tmp_variable(dtype="int64") + sequence_num = helper.create_tmp_variable(dtype="int64") + helper.append_op( + type="edit_distance", + inputs={"Hyps": [input], + "Refs": [label]}, + outputs={"Out": [edit_distance_out], + "SequenceNum": [sequence_num]}, + attrs={"normalized": normalized}) + + return edit_distance_out, sequence_num + + +def ctc_greedy_decoder(input, blank, name=None): + """ + This op is used to decode sequences by greedy policy by below steps: + 1. Get the indexes of max value for each row in input. a.k.a. + numpy.argmax(input, axis=0). + 2. For each sequence in result of step1, merge repeated tokens between two + blanks and delete all blanks. + + A simple example as below: + + .. code-block:: text + + Given: + + input.data = [[0.6, 0.1, 0.3, 0.1], + [0.3, 0.2, 0.4, 0.1], + [0.1, 0.5, 0.1, 0.3], + [0.5, 0.1, 0.3, 0.1], + + [0.5, 0.1, 0.3, 0.1], + [0.2, 0.2, 0.2, 0.4], + [0.2, 0.2, 0.1, 0.5], + [0.5, 0.1, 0.3, 0.1]] + + input.lod = [[0, 4, 8]] + + Then: + + output.data = [[2], + [1], + [3]] + + output.lod = [[0, 2, 3]] + + Args: + + input(Variable): (LoDTensor), the probabilities of + variable-length sequences, which is a 2-D Tensor with + LoD information. It's shape is [Lp, num_classes + 1], + where Lp is the sum of all input sequences' length and + num_classes is the true number of classes. (not + including the blank label). + + blank(int): the blank label index of Connectionist Temporal + Classification (CTC) loss, which is in thehalf-opened + interval [0, num_classes + 1). + + Returns: + Variable: CTC greedy decode result. + + Examples: + .. code-block:: python + + x = fluid.layers.data(name='x', shape=[8], dtype='float32') + + cost = fluid.layers.ctc_greedy_decoder(input=x, blank=0) + """ + helper = LayerHelper("ctc_greedy_decoder", **locals()) + # top 1 op + topk_out = helper.create_tmp_variable(dtype=input.dtype) + topk_indices = helper.create_tmp_variable(dtype="int64") + helper.append_op( + type="top_k", + inputs={"X": [input]}, + outputs={"Out": [topk_out], + "Indices": [topk_indices]}, + attrs={"k": 1}) + + # ctc align op + ctc_out = helper.create_tmp_variable(dtype="int64") + helper.append_op( + type="ctc_align", + inputs={"Input": [topk_indices]}, + outputs={"Output": [ctc_out]}, + attrs={"merge_repeated": True, + "blank": blank}) + return ctc_out + + +def warpctc(input, label, blank=0, norm_by_times=False, **kwargs): + """ + An operator integrating the open source Warp-CTC library + (https://github.com/baidu-research/warp-ctc) + to compute Connectionist Temporal Classification (CTC) loss. + It can be aliased as softmax with CTC, since a native softmax activation is + interated to the Warp-CTC library, to to normlize values for each row of the + input tensor. + + Args: + input(Variable): (LodTensor, default: LoDTensor), + the unscaled probabilities of variable-length sequences, + which is a 2-D Tensor with LoD information. + It's shape is [Lp, num_classes + 1], where Lp is the sum of all input + sequences' length and num_classes is the true number of classes. + (not including the blank label). + label(Variable): (LodTensor, default: LoDTensor), the ground truth + of variable-length sequence, which is a 2-D Tensor with LoD + information. It is of the shape [Lg, 1], where Lg is th sum of + all labels' length. + blank: (int, default: 0), the blank label index of Connectionist + Temporal Classification (CTC) loss, which is in the + half-opened interval [0, num_classes + 1). + norm_by_times: (bool, default: false), whether to normalize + the gradients by the number of time-step, which is also the + sequence's length. There is no need to normalize the gradients + if warpctc layer was follewed by a mean_op. + + Returns: + Variable: The Connectionist Temporal Classification (CTC) loss, + which is a 2-D Tensor of the shape [batch_size, 1]. + + Examples: + .. code-block:: python + y = layers.data( + name='y', shape=[11, 8], dtype='float32', lod_level=1) + y_predict = layers.data( + name='y_predict', shape=[11, 1], dtype='float32') + cost = layers.warpctc(input=y_predict, label=y) + + """ + helper = LayerHelper('warpctc', **kwargs) + loss_out = helper.create_tmp_variable(dtype=input.dtype) + grad_out = helper.create_tmp_variable(dtype=input.dtype) + helper.append_op( + type='warpctc', + inputs={'Logits': [input], + 'Label': [label]}, + outputs={'WarpCTCGrad': [grad_out], + 'Loss': [loss_out]}, + attrs={'blank': blank, + 'norm_by_times': norm_by_times}) + return loss_out + + +def sequence_reshape(input, new_dim): + """ + **Sequence Reshape Layer** + + This layer will rearrange the input sequences. The new dimension is set by + user. Length of each sequence is computed according to original length, + original dimension and new dimension. The following example will help to + illustrate the function of this layer: + + .. code-block:: text + + x is a LoDTensor: + x.lod = [[0, 2, 6]] + x.data = [[1, 2], [3, 4], + [5, 6], [7, 8], [9, 10], [11, 12]] + x.dims = [6, 2] + + set new_dim = 4 + + then out is a LoDTensor: + out.lod = [[0, 1, 3]] + out.data = [[1, 2, 3, 4], + [5, 6, 7, 8], [9, 10, 11, 12]] + out.dims = [3, 4] + + Currently, only 1-level LoDTensor is supported and please make sure + (original length * original dimension) can be divided by new dimension with + no remainder for each sequence. + + Args: + input (Variable): (LodTensor, default: LoDTensor), a 2-D LoDTensor + with shape being [N, M] where M for dimension. + new_dim (int): New dimension which the input LoDTensor is reshaped to. + + Returns: + Variable: Reshaped LoDTensor according to new dimension. + + Examples: + .. code-block:: python + + x = fluid.layers.data(name='x', shape=[5, 20], + dtype='float32', lod_level=1) + x_reshaped = layers.sequence_reshape(input=x, new_dim=10) + """ + helper = LayerHelper('sequence_reshape', **locals()) + out = helper.create_tmp_variable(helper.input_dtype()) + helper.append_op( + type='sequence_reshape', + inputs={'X': [input]}, + outputs={'Out': [out]}, + attrs={'new_dim': new_dim}) + return out + + +@autodoc() +def nce(input, + label, + num_total_classes, + sample_weight=None, + param_attr=None, + bias_attr=None, + num_neg_samples=None): + helper = LayerHelper('nce', **locals()) + assert isinstance(input, Variable) + dim = input.shape[1] + assert isinstance(label, Variable) + num_true_class = label.shape[1] + w = helper.create_parameter( + attr=helper.param_attr, + shape=[num_total_classes, dim], + is_bias=False, + dtype=input.dtype) + b = helper.create_parameter( + attr=helper.bias_attr, + shape=[num_total_classes, 1], + is_bias=True, + dtype=input.dtype) + cost = helper.create_tmp_variable(dtype=input.dtype) + sample_logits = helper.create_tmp_variable(dtype=input.dtype) + sample_labels = helper.create_tmp_variable(dtype=label.dtype) + + if num_neg_samples is None: + num_neg_samples = 10 + else: + num_neg_samples = int(num_neg_samples) + + attrs = { + 'num_total_classes': int(num_total_classes), + 'num_neg_samples': num_neg_samples + } + + helper.append_op( + type='nce', + inputs={ + 'Input': input, + 'Label': label, + 'Weight': w, + 'Bias': b, + 'SampleWeight': sample_weight if sample_weight is not None else [] + }, + outputs={ + 'Cost': cost, + 'SampleLogits': sample_logits, + 'SampleLabels': sample_labels + }, + attrs=attrs) + return cost / (num_neg_samples + 1) + + +def transpose(x, perm, name=None): + """ + **transpose Layer** + + Permute the dimensions of `input` according to `perm`. + + The `i`-th dimension of the returned tensor will correspond to the + perm[i]-th dimension of `input`. + + Args: + input (Variable): (Tensor), A Tensor. + perm (list): A permutation of the dimensions of `input`. + + Returns: + Variable: A transposed Tensor. + + Examples: + .. code-block:: python + + x = fluid.layers.data(name='x', shape=[5, 10, 15], dtype='float32') + x_transposed = layers.transpose(x, perm=[1, 0, 2]) + """ + + if len(perm) != len(x.shape): + raise ValueError( + "Input(perm) is the permutation of dimensions of Input(input). " + "It's length shoud be equal to Input(input)'s rank.") + for idx, dim in enumerate(perm): + if dim >= len(x.shape): + raise ValueError( + "Each element in perm should be less than x's rank. " + "%d-th element in perm is %d which accesses x's rank %d." % + (idx, perm[idx], len(x.shape))) + + helper = LayerHelper('transpose', **locals()) + out = helper.create_tmp_variable(x.dtype) + helper.append_op( + type='transpose', + inputs={'X': [x]}, + outputs={'Out': [out]}, + attrs={'axis': perm}) + return out + + +def im2sequence(input, filter_size=1, stride=1, padding=0, name=None): + """ + Extracts image patches from the input tensor to form a tensor of shape + {input.batch_size * output_height * output_width, filter_size_H * + filter_size_W * input.channels} which is similar with im2col. + This op use filter / kernel to scan images and convert these images to + sequences. After expanding, the number of time step are + output_height * output_width for an image, in which output_height and + output_width are calculated by below equation: + + .. math:: + + output\_size = 1 + \ + (2 * padding + img\_size - block\_size + stride - 1) / stride + + And the dimension of each time step is block_y * block_x * input.channels. + + Args: + input (Variable): The input should be a tensor in NCHW format. + + filter_size(int|tuple|None): The filter size. If filter_size is a tuple, + it must contain two integers, (filter_size_H, filter_size_W). + Otherwise, the filter will be a square. + + stride(int|tuple): The stride size. If stride is a tuple, it must + contain two integers, (stride_H, stride_W). Otherwise, the + stride_H = stride_W = stride. Default: stride = 1. + + padding(int|tuple): The padding size. If padding is a tuple, it can + contain two integers like (padding_H, padding_W) which means + padding_up = padding_down = padding_H and + padding_left = padding_right = padding_W. Or it can use + (padding_up, padding_left, padding_down, padding_right) to indicate + paddings of four direction. Otherwise, a scalar padding means + padding_up = padding_down = padding_left = padding_right = padding + Default: padding = 0. + + name (int): The name of this layer. It is optional. + + Returns: + output: The output is a LoDTensor with shape + {input.batch_size * output_height * output_width, + filter_size_H * filter_size_W * input.channels}. + If we regard output as a matrix, each row of this matrix is + a step of a sequence. + + Examples: + + As an example: + + .. code-block:: text + + Given: + + x = [[[[ 6. 2. 1.] + [ 8. 3. 5.] + [ 0. 2. 6.]] + + [[ 2. 4. 4.] + [ 6. 3. 0.] + [ 6. 4. 7.]]] + + [[[ 6. 7. 1.] + [ 5. 7. 9.] + [ 2. 4. 8.]] + + [[ 1. 2. 1.] + [ 1. 3. 5.] + [ 9. 0. 8.]]]] + + x.dims = {2, 2, 3, 3} + + And: + + filter = [2, 2] + stride = [1, 1] + padding = [0, 0] + + Then: + + output.data = [[ 6. 2. 8. 3. 2. 4. 6. 3.] + [ 2. 1. 3. 5. 4. 4. 3. 0.] + [ 8. 3. 0. 2. 6. 3. 6. 4.] + [ 3. 5. 2. 6. 3. 0. 4. 7.] + [ 6. 7. 5. 7. 1. 2. 1. 3.] + [ 7. 1. 7. 9. 2. 1. 3. 5.] + [ 5. 7. 2. 4. 1. 3. 9. 0.] + [ 7. 9. 4. 8. 3. 5. 0. 8.]] + + output.dims = {8, 9} + + output.lod = [[0, 4, 8]] + + The simple usage is: + + .. code-block:: python + + output = fluid.layers.im2sequence( + input=layer, stride=[1, 1], filter_size=[2, 2]) + + """ + + if isinstance(filter_size, int): + filter_size = [filter_size, filter_size] + if isinstance(stride, int): + stride = [stride, stride] + if isinstance(padding, int): + padding = [padding, padding] + if len(padding) == 2: + padding.append(padding[0]) + padding.append(padding[1]) + + helper = LayerHelper('im2sequence', **locals()) + out = helper.create_tmp_variable(dtype=helper.input_dtype()) + helper.append_op( + type='im2sequence', + inputs={'X': input}, + outputs={'Out': out}, + attrs={ + 'kernels': filter_size, + 'strides': stride, + 'paddings': padding, + }) + return out + + +def row_conv(input, future_context_size, param_attr=None, act=None): + """Row Conv Operator. This layer will apply lookahead convolution to + **input**. The input variable should be a 2D LoDTensor with shape [T, D]. + Parameters with shape [future_context_size + 1, D] will be created. The math + equation of row convolution is as follows: + + .. math:: + Out_{i} = \sum_{j = i} ^ {i + \\tau} X_{j} \odot W_{i - j} + + In the above equation: + + * :math:`Out_{i}`: The i-th row of output variable with shape [1, D]. + * :math:`\\tau`: Future context size. + * :math:`X_{j}`: The j-th row of input variable with shape [1, D]. + * :math:`W_{i-j}`: The (i-j)-th row of parameters with shape [1, D]. + + More details about row_conv please refer to the paper \ + (http://www.cs.cmu.edu/~dyogatam/papers/wang+etal.iclrworkshop2016.pdf) and + the design document \ + (https://github.com/PaddlePaddle/Paddle/issues/2228#issuecomment-303903645). + + Args: + input (Variable): Input variable, a 2D LoDTensor with shape [T, D]. + future_context_size (int): Future context size. Please note, the shape + of convolution kernel is [future_context_size + 1, D]. + param_attr (ParamAttr): Attributes of parameters, including + name, initializer etc. + act (str): Non-linear activation to be applied to output variable. + + Returns: + Variable: The output tensor with same shape as input tensor. + + Examples: + .. code-block:: python + + x = fluid.layers.data(name='x', shape=[16], + dtype='float32', lod_level=1) + out = fluid.layers.row_conv(input=x, future_context_size=2) + """ + helper = LayerHelper('row_conv', **locals()) + dtype = helper.input_dtype() + filter_shape = [future_context_size + 1, input.shape[1]] + filter_param = helper.create_parameter( + attr=helper.param_attr, shape=filter_shape, dtype=dtype) + out = helper.create_tmp_variable(dtype) + helper.append_op( + type='row_conv', + inputs={'X': [input], + 'Filter': [filter_param]}, + outputs={'Out': [out]}) + return helper.append_activation(out) + + +def multiplex(inputs, index): + """ + **Multiplex Layer** + + Referring to the given index variable, this layer selects rows from the + input variables to construct a multiplex variable. Assuming that there are + :math:`m` input variables and :math:`I_i` represents the i-th input + variable and :math:`i` is in [0, :math:`m`). All input variables are + tensors with same shape [:math:`d_0`, :math:`d_1`, ..., :math:`d_R`]. + Please note that rank of the input tensor should be at least 2. Each input + variable will be treated as a 2-D matrix with shape [:math:`M`, :math:`N`] + where :math:`M` for :math:`d_0` and :math:`N` for :math:`d_1` * :math:`d_2` + * ... * :math:`d_R`. Let :math:`I_i[j]` be the j-th row of the i-th input + variable. The given index variable should be a 2-D tensor with shape + [:math:`M`, 1]. Let `ID[i]` be the i-th index value of the index variable. + Then the output variable will be a tensor with shape [:math:`d_0`, + :math:`d_1`, ..., :math:`d_R`]. If we treat the output tensor as a 2-D + matrix with shape [:math:`M`, :math:`N`] and let :math:`O[i]` be the i-th + row of the matrix, then `O[i]` is equal to :math:`I_{ID[i]}[i]`. + + Args: + inputs (list): A list of variables to gather from. All variables have the + same shape and the rank is at least 2. + index (Variable): Tensor, index variable which is a 2-D tensor + with shape [M, 1] where M is the batch size. + + Returns: + Variable: Multiplex variable gathered from input variables. + + Examples: + .. code-block:: python + + x1 = fluid.layers.data(name='x1', shape=[4], dtype='float32') + x2 = fluid.layers.data(name='x2', shape=[4], dtype='float32') + index = fluid.layers.data(name='index', shape=[1], dtype='int32') + out = fluid.layers.multiplex(inputs=[x1, x2], index=index) + """ + helper = LayerHelper('multiplex', **locals()) + + if not isinstance(inputs, list) and len(inputs) < 2: + raise ValueError("inputs should be a list object and contains at least " + "2 elements.") + + out = helper.create_tmp_variable(inputs[0].dtype) + helper.append_op( + type='multiplex', + inputs={'X': inputs, + 'Ids': index}, + outputs={'Out': [out]}) + return out diff --git a/python/paddle/v2/fluid/layers/ops.py b/python/paddle/v2/fluid/layers/ops.py index 73d7c895806ef28ffc98db88809e317a86762769..ee3172c7b8dfd65c693e5aee9b55179e654ce7be 100644 --- a/python/paddle/v2/fluid/layers/ops.py +++ b/python/paddle/v2/fluid/layers/ops.py @@ -1,17 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from ..registry import register_layer +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from layer_function_generator import generate_layer_fn __activations__ = [ 'sigmoid', @@ -49,15 +49,18 @@ __all__ = [ 'mul', 'reshape', 'scale', - 'transpose', 'sigmoid_cross_entropy_with_logits', 'elementwise_add', 'elementwise_div', 'elementwise_sub', 'elementwise_mul', + 'elementwise_max', + 'elementwise_min', + 'elementwise_pow', 'clip', + 'clip_by_norm', 'sequence_softmax', ] + __activations__ for _OP in set(__all__): - globals()[_OP] = register_layer(_OP) + globals()[_OP] = generate_layer_fn(_OP) diff --git a/python/paddle/v2/fluid/layers/tensor.py b/python/paddle/v2/fluid/layers/tensor.py index 255b9d467839a05447c8999c047d96769ae25f17..c435c5206d1ef1ef57683a1a47bf089be6526f38 100644 --- a/python/paddle/v2/fluid/layers/tensor.py +++ b/python/paddle/v2/fluid/layers/tensor.py @@ -1,26 +1,29 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from ..layer_helper import LayerHelper from ..param_attr import ParamAttr from ..framework import convert_np_dtype_to_dtype_ from ..framework import Variable +from ..initializer import Constant from ..core import DataType import numpy __all__ = [ 'create_tensor', 'create_parameter', + 'create_global_var', 'cast', 'concat', 'sums', @@ -57,13 +60,22 @@ def create_parameter(shape, Returns: Parameter: the created parameter """ - helper = LayerHelper("create_parameter") + helper = LayerHelper("create_parameter", **locals()) if attr is None: attr = ParamAttr() return helper.create_parameter(attr, shape, dtype, is_bias, default_initializer) +def create_global_var(shape, value, dtype, persistable=False, name=None): + helper = LayerHelper("global_var", **locals()) + var = helper.create_global_variable( + dtype=dtype, shape=shape, persistable=persistable, name=name) + helper.set_variable_initializer( + var, initializer=Constant(value=float(value))) + return var + + def cast(x, dtype): """ This function takes in the input with input_dtype diff --git a/python/paddle/v2/fluid/learning_rate_decay.py b/python/paddle/v2/fluid/learning_rate_decay.py new file mode 100644 index 0000000000000000000000000000000000000000..96b3e9a0d73cede5d6e36308a53ab8927a95a6da --- /dev/null +++ b/python/paddle/v2/fluid/learning_rate_decay.py @@ -0,0 +1,125 @@ +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import layers +from framework import Variable + +__all__ = ['exponential_decay', 'natural_exp_decay', 'inverse_time_decay'] +""" +When training a model, it's often useful to decay the +learning rate during training process, this is called +learning_rate_decay. There are many strategies to do +this, this module will provide some classical method. +User can also implement their own learning_rate_decay +strategy according to this module. +""" + + +def exponential_decay(learning_rate, + global_step, + decay_steps, + decay_rate, + staircase=False): + """Applies exponential decay to the learning rate. + + ```python + decayed_learning_rate = learning_rate * + decay_rate ^ (global_step / decay_steps) + ``` + Args: + learning_rate: A scalar float32 value or a Variable. This + will be the initial learning rate during training + global_step: A Variable that record the training step. + decay_steps: A Python `int32` number. + decay_rate: A Python `float` number. + staircase: Boolean. If set true, decay the learning rate every decay_steps. + + Returns: + The decayed learning rate + """ + if not isinstance(global_step, Variable): + raise ValueError("global_step is required for exponential_decay.") + + # update learning_rate + div_res = global_step / decay_steps + if staircase: + div_res = layers.floor(x=div_res) + return learning_rate * (decay_rate**div_res) + + +def natural_exp_decay(learning_rate, + global_step, + decay_steps, + decay_rate, + staircase=False): + """Applies natural exponential decay to the initial learning rate. + + ```python + if not staircase: + decayed_learning_rate = learning_rate * exp(- decay_rate * (global_step / decay_steps)) + else: + decayed_learning_rate = learning_rate * exp(- decay_rate * (global_step / decay_steps)) + ``` + Args: + learning_rate: A scalar float32 value or a Variable. This + will be the initial learning rate during training + global_step: A Variable that record the training step. + decay_steps: A Python `int32` number. + decay_rate: A Python `float` number. + staircase: Boolean. If set true, decay the learning rate every decay_steps. + + Returns: + The decayed learning rate + """ + if not isinstance(global_step, Variable): + raise ValueError("global_step is required for natural_exp_decay.") + + div_res = global_step / decay_steps + if staircase: + div_res = layers.floor(x=div_res) + return learning_rate * layers.exp(x=(-1 * decay_rate * div_res)) + + +def inverse_time_decay(learning_rate, + global_step, + decay_steps, + decay_rate, + staircase=False): + """Applies inverse time decay to the initial learning rate. + + ```python + if staircase: + decayed_learning_rate = learning_rate / (1 + decay_rate * floor(global_step / decay_step)) + else + decayed_learning_rate = learning_rate / (1 + decay_rate * global_step / decay_step) + ``` + Args: + learning_rate: A scalar float32 value or a Variable. This + will be the initial learning rate during training + global_step: A Variable that record the training step. + decay_steps: A Python `int32` number. + decay_rate: A Python `float` number. + staircase: Boolean. If set true, decay the learning rate every decay_steps. + + Returns: + The decayed learning rate + """ + if not isinstance(global_step, Variable): + raise ValueError("global_step is required for inverse_time_decay.") + + div_res = global_step / decay_steps + if staircase: + div_res = layers.floor(x=div_res) + + return learning_rate / (1 + decay_rate * div_res) diff --git a/python/paddle/v2/fluid/memory_optimization_transpiler.py b/python/paddle/v2/fluid/memory_optimization_transpiler.py index 89ffe26ed1a70942c4994394f9a3635dda13be69..956c5b66da28fd8e74d4fd12f249688daa72d8ac 100644 --- a/python/paddle/v2/fluid/memory_optimization_transpiler.py +++ b/python/paddle/v2/fluid/memory_optimization_transpiler.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from collections import defaultdict import framework from framework import Program, default_main_program, Parameter, Variable @@ -30,10 +31,12 @@ dtype_to_size = { class ControlFlowGraph(object): - def __init__(self, Program): + def __init__(self, Program, ops, forward_num): self._program = Program - self._succesors = defaultdict(set) - self._presucessors = defaultdict(set) + self._ops = ops + self._forward_num = forward_num + self._successors = defaultdict(set) + self._presuccessors = defaultdict(set) self._uses = defaultdict(set) self._defs = defaultdict(set) self._live_in = defaultdict(set) @@ -44,25 +47,16 @@ class ControlFlowGraph(object): self._add(node1, node2) def _add(self, node1, node2): - self._succesors[node1].add(node2) - self._presucessors[node2].add(node1) + self._successors[node1].add(node2) + self._presuccessors[node2].add(node1) def _build_graph(self): - program_desc = self._program.get_desc() - block_size = program_desc.num_blocks() - - # TODO(qijun) handle Program with if/while operators - self.global_block_desc = program_desc.block(0) - self.op_size = self.global_block_desc.op_size() - + self.op_size = len(self._ops) op_node_connections = [(i, i + 1) for i in range(self.op_size - 1)] self._add_connections(op_node_connections) - - self.ops = [self.global_block_desc.op(i) for i in range(self.op_size)] - for i in range(self.op_size): - self._uses[i].update(self.ops[i].input_arg_names()) - self._defs[i].update(self.ops[i].output_arg_names()) + self._uses[i].update(self._ops[i].input_arg_names()) + self._defs[i].update(self._ops[i].output_arg_names()) def _update_graph(self, old_name, new_name, begin_idx=0): for i in range(begin_idx, self.op_size): @@ -102,7 +96,7 @@ class ControlFlowGraph(object): live_out[i] = set(self._live_out[i]) self._live_in[i] = self._uses[i] | ( self._live_out[i] - self._defs[i]) - for s in self._succesors[i]: + for s in self._successors[i]: self._live_out[i] |= self._live_in[s] if self._reach_fixed_point(live_in, live_out): @@ -112,39 +106,76 @@ class ControlFlowGraph(object): u = a & b return a - u, b - u + def _has_var(self, block_desc, var_name, is_forward): + if is_forward: + return block_desc.has_var(str(var_name)) + else: + return block_desc.has_var_recursive(str(var_name)) + + def _find_var(self, block_desc, var_name, is_forward): + if is_forward: + return block_desc.find_var(str(var_name)) + else: + return block_desc.find_var_recursive(str(var_name)) + def memory_optimize(self): + def check_var_validity(block_desc, x, is_forward): + if str(x) == "@EMPTY@": + return False + if not self._has_var(block_desc, x, is_forward): + return False + if self._find_var(block_desc, x, is_forward).persistable(): + return False + if self._find_var( + block_desc, x, + is_forward).type() != core.VarDesc.VarType.LOD_TENSOR: + return False + return True + self._build_graph() self._dataflow_analyze() self.pool = [] for i in range(self.op_size): + op = self._ops[i] + if op.type() == "while" or op.type() == "while_grad": + continue + block_desc = op.block() + is_forward = i < self._forward_num if self.pool: - out_pair = [(x, self.global_block_desc.var(str(x)).shape()) - for x in self._defs[i]] + defs_can_optimize = filter( + lambda x: check_var_validity(block_desc, x, is_forward), + self._defs[i]) + out_pair = [ + (x, self._find_var(block_desc, x, is_forward).shape()) + for x in defs_can_optimize + ] for x, x_shape in out_pair: - if not self.global_block_desc.var(str(x)).persistable(): - for index, cache_pair in enumerate(self.pool): - cache_var = cache_pair[0] - cache_shape = cache_pair[1] - if x_shape == cache_shape: - x_dtype = self.global_block_desc.var(str( - x)).dtype() - cache_dtype = self.global_block_desc.var( - str(cache_var)).dtype() + for index, cache_pair in enumerate(self.pool): + cache_var = cache_pair[0] + cache_shape = cache_pair[1] + if x_shape == cache_shape: + if self._has_var(block_desc, cache_var, is_forward): + x_dtype = self._find_var(block_desc, x, + is_forward).dtype() + cache_dtype = self._find_var( + block_desc, cache_var, is_forward).dtype() # TODO(qijun): actually, we should compare dtype_to_size[x_dtype] # and dtype_to_size[cache_dtype] if x_dtype == cache_dtype: - print( - ("Hit Cache !!!! cache pool index " - "is %d, var name is %s, " - "cached var name is %s, " - "var shape is %s ") % - (index, x, cache_var, str(cache_shape))) + print(("Hit Cache !!!! cache pool index " + "is %d, var name is %s, " + "cached var name is %s, " + "var shape is %s ") % + (index, x, cache_var, + str(cache_shape))) self.pool.pop(index) + if x == cache_var: + break _rename_arg_( - self.ops, x, cache_var, begin_idx=i) - self._program.current_block().var(str( - x)).desc = self.global_block_desc.var( - str(cache_var)) + self._ops, x, cache_var, begin_idx=i) + self._program.block(block_desc.id).var( + str(x)).desc = self._find_var( + block_desc, cache_var, is_forward) self._update_graph( x, cache_var, begin_idx=i) break @@ -152,20 +183,70 @@ class ControlFlowGraph(object): in_diff, out_diff = self._get_diff(self._live_in[i], self._live_out[i]) can_optimize = filter( - lambda x: not self.global_block_desc.var(str(x)).persistable(), + lambda x: check_var_validity(block_desc, x, is_forward), in_diff) if can_optimize: for var_name in can_optimize: - self.pool.append( - (var_name, - self.global_block_desc.var(str(var_name)).shape())) - - def get_program(self): - return self._program + self.pool.append((var_name, self._find_var( + block_desc, var_name, is_forward).shape())) + + +def get_cfgs(input_program): + ops_list = [] + pdesc = input_program.get_desc() + block_desc = pdesc.block(0) + op_size = block_desc.op_size() + # Get global block ops + ops_list.append(([block_desc.op(i) for i in range(op_size)], op_size)) + + while_sub_block_ids = [] + while_grad_sub_block_ids = [] + while_pair = [] + + for i in range(op_size): + op = block_desc.op(i) + if op.type() == "while": + while_sub_block_ids.append(op.attr("sub_block").id) + elif op.type() == "while_grad": + while_grad_sub_block_ids.append(op.attr("sub_block").id) + + # Find while/while_grad block pair + for grad_id in while_grad_sub_block_ids: + parent_id = pdesc.block(grad_id).parent + if parent_id in while_sub_block_ids: + while_pair.append((parent_id, grad_id)) + while_sub_block_ids.remove(parent_id) + + # Get while/while_grad block ops + for parent_id, grad_id in while_pair: + while_block_ops = [] + while_block = pdesc.block(parent_id) + while_block_op_size = while_block.op_size() + for i in range(while_block_op_size): + while_block_ops.append(while_block.op(i)) + + while_grad_block = pdesc.block(grad_id) + while_grad_block_op_size = while_grad_block.op_size() + for i in range(while_grad_block_op_size): + while_block_ops.append(while_grad_block.op(i)) + + ops_list.append((while_block_ops, while_block_op_size)) + + # Process rest while block ops + for parent_id in while_sub_block_ids: + while_block_ops = [] + while_block = pdesc.block(parent_id) + while_block_op_size = while_block.op_size() + for i in range(while_block_op_size): + while_block_ops.append(while_block.op(i)) + + ops_list.append((while_block_ops, while_block_op_size)) + + cfgs = [ControlFlowGraph(input_program, i, j) for i, j in ops_list] + return cfgs def memory_optimize(input_program): - graph = ControlFlowGraph(input_program) - graph.memory_optimize() - result_program = graph.get_program() - return result_program + cfgs = get_cfgs(input_program) + for cfg in cfgs: + cfg.memory_optimize() diff --git a/python/paddle/v2/fluid/net_drawer.py b/python/paddle/v2/fluid/net_drawer.py index 7448975b59ba13bb29cd0a1bea043add251844e6..9b126f51971acabba73db7bd6c33ba39e8876ca3 100644 --- a/python/paddle/v2/fluid/net_drawer.py +++ b/python/paddle/v2/fluid/net_drawer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import argparse import json import logging diff --git a/python/paddle/v2/fluid/nets.py b/python/paddle/v2/fluid/nets.py index b5c26e713db4e5c48d6c0851fd68275aeabeea64..cb63d43709e23ae04c4d23457bbb79e6f7f0ce3c 100644 --- a/python/paddle/v2/fluid/nets.py +++ b/python/paddle/v2/fluid/nets.py @@ -1,22 +1,23 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import layers __all__ = [ "simple_img_conv_pool", "sequence_conv_pool", "glu", + "scaled_dot_product_attention", ] @@ -27,19 +28,22 @@ def simple_img_conv_pool(input, pool_stride, act, param_attr=None, - pool_type='max'): + pool_type='max', + use_cudnn=True): conv_out = layers.conv2d( input=input, num_filters=num_filters, filter_size=filter_size, param_attr=param_attr, - act=act) + act=act, + use_cudnn=use_cudnn) pool_out = layers.pool2d( input=conv_out, pool_size=pool_size, pool_type=pool_type, - pool_stride=pool_stride) + pool_stride=pool_stride, + use_cudnn=use_cudnn) return pool_out @@ -51,9 +55,10 @@ def img_conv_group(input, conv_act=None, param_attr=None, conv_with_batchnorm=False, - conv_batchnorm_drop_rate=None, + conv_batchnorm_drop_rate=0.0, pool_stride=1, - pool_type=None): + pool_type=None, + use_cudnn=True): """ Image Convolution Group, Used for vgg net. """ @@ -84,7 +89,8 @@ def img_conv_group(input, filter_size=conv_filter_size[i], padding=conv_padding[i], param_attr=param_attr[i], - act=local_conv_act) + act=local_conv_act, + use_cudnn=use_cudnn) if conv_with_batchnorm[i]: tmp = layers.batch_norm(input=tmp, act=conv_act) @@ -96,7 +102,8 @@ def img_conv_group(input, input=tmp, pool_size=pool_size, pool_type=pool_type, - pool_stride=pool_stride) + pool_stride=pool_stride, + use_cudnn=use_cudnn) return pool_out @@ -119,21 +126,21 @@ def sequence_conv_pool(input, def glu(input, dim=-1): """ - The gated linear unit composed by split, sigmoid activation and elementwise - multiplication. Specifically, Split the input into two equal sized parts - :math:`a` and :math:`b` along the given dimension and then compute as + The gated linear unit composed by split, sigmoid activation and elementwise + multiplication. Specifically, Split the input into two equal sized parts + :math:`a` and :math:`b` along the given dimension and then compute as following: .. math:: {GLU}(a, b)= a \otimes \sigma(b) - Refer to `Language Modeling with Gated Convolutional Networks + Refer to `Language Modeling with Gated Convolutional Networks `_. - + Args: input (Variable): The input variable which is a Tensor or LoDTensor. - dim (int): The dimension along which to split. If :math:`dim < 0`, the + dim (int): The dimension along which to split. If :math:`dim < 0`, the dimension to split along is :math:`rank(input) + dim`. Returns: @@ -150,3 +157,182 @@ def glu(input, dim=-1): act_b = layers.sigmoid(x=b) out = layers.elementwise_mul(x=a, y=act_b) return out + + +def scaled_dot_product_attention(queries, + keys, + values, + num_heads=1, + dropout_rate=0.): + """ + The dot-product attention. + + Attention mechanism can be seen as mapping a query and a set of key-value + pairs to an output. The output is computed as a weighted sum of the values, + where the weight assigned to each value is computed by a compatibility + function (dot-product here) of the query with the corresponding key. + + The dot-product attention can be implemented through (batch) matrix + multipication as follows: + + .. math:: + + Attention(Q, K, V)= softmax(QK^\mathrm{T})V + + Refer to `Attention Is All You Need + `_. + + Args: + + queries (Variable): The input variable which should be a 3-D Tensor. + keys (Variable): The input variable which should be a 3-D Tensor. + values (Variable): The input variable which should be a 3-D Tensor. + num_heads (int): Head number to compute the scaled dot product + attention. Default value is 1. + dropout_rate (float): The dropout rate to drop the attention weight. + Default value is 0. + + Returns: + + Variable: A 3-D Tensor computed by multi-head scaled dot product + attention. + + Raises: + + ValueError: If input queries, keys, values are not 3-D Tensors. + + NOTE: + 1. When num_heads > 1, three linear projections are learned respectively + to map input queries, keys and values into queries', keys' and values'. + queries', keys' and values' have the same shapes with queries, keys + and values. + + 1. When num_heads == 1, scaled_dot_product_attention has no learnable + parameters. + + Examples: + .. code-block:: python + + # Suppose q, k, v are Tensors with the following shape: + # q: [3, 5, 9], k: [3, 6, 9], v: [3, 6, 10] + + contexts = fluid.nets.scaled_dot_product_attention(q, k, v) + contexts.shape # [3, 5, 10] + """ + if not (len(queries.shape) == len(keys.shape) == len(values.shape) == 3): + raise ValueError( + "Inputs quries, keys and values should all be 3-D tensors.") + + if queries.shape[-1] != keys.shape[-1]: + raise ValueError( + "The hidden size of queries and keys should be the same.") + if keys.shape[-2] != values.shape[-2]: + raise ValueError( + "The max sequence length in query batch and in key batch " + "should be the same.") + if keys.shape[-1] % num_heads != 0: + raise ValueError("The hidden size of keys (%d) must be divisible " + "by the number of attention heads (%d)." % + (keys.shape[-1], num_heads)) + if values.shape[-1] % num_heads != 0: + raise ValueError("The hidden size of values (%d) must be divisible " + "by the number of attention heads (%d)." % + (values.shape[-1], num_heads)) + + def __compute_qkv(queries, keys, values, num_heads): + """ + Add linear projection to queries, keys, and values. + + Args: + queries(Tensor): a 3-D input Tensor. + keys(Tensor): a 3-D input Tensor. + values(Tensor): a 3-D input Tensor. + num_heads(int): The number of heads. Linearly project the inputs + ONLY when num_heads > 1. + + Returns: + Tensor: linearly projected output Tensors: queries', keys' and + values'. They have the same shapes with queries, keys and + values. + """ + + if num_heads == 1: + return queries, keys, values + + q = layers.fc(input=queries, size=queries.shape[-1], num_flatten_dims=2) + k = layers.fc(input=keys, size=keys.shape[-1], num_flatten_dims=2) + v = layers.fc(input=values, size=values.shape[-1], num_flatten_dims=2) + return q, k, v + + def __split_heads(x, num_heads): + """ + Reshape the last dimension of inpunt tensor x so that it becomes two + dimensions. + + Args: + x(Tensor): a 3-D input Tensor. + num_heads(int): The number of heads. + + Returns: + Tensor: a Tensor with shape [..., n, m/num_heads], where m is size + of the last dimension of x. + """ + if num_heads == 1: + return x + + hidden_size = x.shape[-1] + # reshape the 3-D input: [batch_size, max_sequence_length, hidden_dim] + # into a 4-D output: + # [batch_size, max_sequence_length, num_heads, hidden_size_per_head]. + reshaped = layers.reshape( + x=x, + shape=list(x.shape[:-1]) + [num_heads, hidden_size // num_heads]) + + # permuate the dimensions into: + # [batch_size, num_heads, max_sequence_len, hidden_size_per_head] + return layers.transpose(x=reshaped, perm=[0, 2, 1, 3]) + + def __combine_heads(x): + """ + Reshape the last two dimensions of inpunt tensor x so that it becomes + one dimension. + + Args: + x(Tensor): a 4-D input Tensor with shape + [bs, num_heads, max_sequence_length, hidden_dim]. + + Returns: + Tensor: a Tensor with shape + [bs, max_sequence_length, num_heads * hidden_dim]. + """ + + if len(x.shape) == 3: return x + if len(x.shape) != 4: + raise ValueError("Input(x) should be a 4-D Tensor.") + + trans_x = layers.transpose(x, perm=[0, 2, 1, 3]) + return layers.reshape( + x=trans_x, + shape=map(int, [ + trans_x.shape[0], trans_x.shape[1], + trans_x.shape[2] * trans_x.shape[3] + ])) + + q, k, v = __compute_qkv(queries, keys, values, num_heads) + + q = __split_heads(q, num_heads) + k = __split_heads(k, num_heads) + v = __split_heads(v, num_heads) + + key_dim_per_head = keys.shape[-1] // num_heads + scaled_q = layers.scale(x=q, scale=key_dim_per_head**-0.5) + product = layers.matmul(x=k, y=scaled_q, transpose_y=True) + + weights = layers.reshape( + x=layers.reshape( + x=product, shape=[-1, product.shape[-1]], act="softmax"), + shape=product.shape) + if dropout_rate: + weights = layers.dropout(x, dropout_prob=dropout_rate, is_test=False) + ctx_multiheads = layers.matmul(weights, v) + return __combine_heads(ctx_multiheads) diff --git a/python/paddle/v2/fluid/op.py b/python/paddle/v2/fluid/op.py index 4bc0f79c64876829d601ac06b9ca451a15300fe8..f368e0c2d86b233cba49b312febf1293c91f91a2 100644 --- a/python/paddle/v2/fluid/op.py +++ b/python/paddle/v2/fluid/op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core as core import paddle.v2.fluid.proto.framework_pb2 as framework_pb2 diff --git a/python/paddle/v2/fluid/optimizer.py b/python/paddle/v2/fluid/optimizer.py index 8bd62ef0c02b1cae1dca8d6f1414286dc643794b..7844a4e2df1ce3989e48082f6472292560fbf1ee 100644 --- a/python/paddle/v2/fluid/optimizer.py +++ b/python/paddle/v2/fluid/optimizer.py @@ -1,19 +1,21 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from collections import defaultdict import framework +import layers from backward import append_backward from framework import unique_name, program_guard from initializer import Constant @@ -32,9 +34,11 @@ class Optimizer(object): but need to use one of it's implementation. """ - def __init__(self, global_step=None, regularization=None): + def __init__(self, learning_rate, global_step=None, regularization=None): + assert learning_rate is not None self._global_step = global_step self.regularization = regularization + self._global_learning_rate = learning_rate # Dictionary of accumulators. Some optimizer subclasses need to # allocate and manage extra variables associated with the parameters # to train. These variables are called accumulators. @@ -42,6 +46,28 @@ class Optimizer(object): self._accumulators = defaultdict(lambda: dict()) self.helper = None + def _create_global_learning_rate(self): + if isinstance(self._global_learning_rate, float): + self._global_learning_rate = layers.create_global_var( + name=unique_name("learning_rate"), + shape=[1], + value=float(self._global_learning_rate), + dtype='float32', + persistable=True) + + if not isinstance(self._global_learning_rate, framework.Variable): + raise ValueError("learning rate should be a Variable, " + "actual type is %s", + type(self._global_learning_rate)) + + @property + def global_learning_rate(self): + """ + get global decayed learning rate + :return: + """ + return self._global_learning_rate + def _append_optimize_op(self, block, param_and_grad): """ append optimize operator to block and return all the added optimize_op """ @@ -51,17 +77,7 @@ class Optimizer(object): # create learning rate variable for every parameter param = param_and_grad[0] param_lr = param.optimize_attr['learning_rate'] - param_lr_shape = [1] - param_lr_var = self.helper.create_global_variable( - name=unique_name("learning_rate"), - dtype='float32', - shape=param_lr_shape, - lod_level=1, - persistable=True) - param_lr = param_lr * self._learning_rate - self.helper.set_variable_initializer( - var=param_lr_var, initializer=Constant(param_lr)) - return param_lr_var + return self._global_learning_rate * param_lr def _create_accumulators(self, block, parameters): """Create all accumulators needed by the parameters @@ -162,7 +178,7 @@ class Optimizer(object): optimization. This will include parameter update ops, global step update ops and any other custom ops required by subclasses to manage their internal state. - :param startup_program: + :param startup_program: """ # This is a default implementation of create_optimization_pass that # can be shared by most optimizers. This implementation assumes that @@ -177,6 +193,7 @@ class Optimizer(object): self.helper = LayerHelper(self.__class__.__name__) self._create_accumulators(loss.block, [p[0] for p in parameters_and_grads]) + self._create_global_learning_rate() optimize_ops = [] for param_and_grad in parameters_and_grads: @@ -230,9 +247,9 @@ class SGDOptimizer(Optimizer): def __init__(self, learning_rate, **kwargs): assert learning_rate is not None - super(SGDOptimizer, self).__init__(**kwargs) + super(SGDOptimizer, self).__init__( + learning_rate=learning_rate, **kwargs) self.type = "sgd" - self._learning_rate = learning_rate def _append_optimize_op(self, block, param_and_grad): assert isinstance(block, framework.Block) @@ -258,9 +275,9 @@ class MomentumOptimizer(Optimizer): def __init__(self, learning_rate, momentum, use_nesterov=False, **kwargs): assert learning_rate is not None assert momentum is not None - super(MomentumOptimizer, self).__init__(**kwargs) + super(MomentumOptimizer, self).__init__( + learning_rate=learning_rate, **kwargs) self.type = "momentum" - self._learning_rate = learning_rate self._momentum = momentum self._use_nesterov = bool(use_nesterov) @@ -302,9 +319,9 @@ class AdagradOptimizer(Optimizer): def __init__(self, learning_rate, epsilon=1.0e-6, **kwargs): assert learning_rate is not None assert epsilon is not None - super(AdagradOptimizer, self).__init__(**kwargs) + super(AdagradOptimizer, self).__init__( + learning_rate=learning_rate, **kwargs) self.type = "adagrad" - self._learning_rate = learning_rate self._epsilon = epsilon def _create_accumulators(self, block, parameters): @@ -351,9 +368,9 @@ class AdamOptimizer(Optimizer): assert beta1 is not None assert beta2 is not None assert epsilon is not None - super(AdamOptimizer, self).__init__(**kwargs) + super(AdamOptimizer, self).__init__( + learning_rate=learning_rate, **kwargs) self.type = "adam" - self._learning_rate = learning_rate self._beta1 = beta1 self._beta2 = beta2 self._epsilon = epsilon @@ -456,9 +473,9 @@ class AdamaxOptimizer(Optimizer): assert beta1 is not None assert beta2 is not None assert epsilon is not None - super(AdamaxOptimizer, self).__init__(**kwargs) + super(AdamaxOptimizer, self).__init__( + learning_rate=learning_rate, **kwargs) self.type = "adamax" - self._learning_rate = learning_rate self._beta1 = beta1 self._beta2 = beta2 self._epsilon = epsilon @@ -534,9 +551,9 @@ class DecayedAdagradOptimizer(Optimizer): assert decay is not None assert epsilon is not None - super(DecayedAdagradOptimizer, self).__init__(**kwargs) + super(DecayedAdagradOptimizer, self).__init__( + learning_rate=learning_rate, **kwargs) self.type = "decayed_adagrad" - self._learning_rate = learning_rate self._decay = decay self._epsilon = epsilon diff --git a/python/paddle/v2/fluid/param_attr.py b/python/paddle/v2/fluid/param_attr.py index 3af0190590e775d4816410e4dbe0069868ea209c..fc566b8a2480ce9256d610b4731405cd6d89b7e4 100644 --- a/python/paddle/v2/fluid/param_attr.py +++ b/python/paddle/v2/fluid/param_attr.py @@ -1,20 +1,24 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from initializer import Initializer, Xavier, Constant from regularizer import WeightDecayRegularizer -__all__ = ['ParamAttr'] +__all__ = [ + 'ParamAttr', + 'WeightNormParamAttr', +] class ParamAttr(object): @@ -24,13 +28,13 @@ class ParamAttr(object): learning_rate=1.0, regularizer=None, trainable=True, - clip=None): + gradient_clip=None): self.name = name self.initializer = initializer self.learning_rate = learning_rate self.regularizer = regularizer self.trainable = trainable - self.clip = clip + self.gradient_clip = gradient_clip def set_default_initializer(self, initializer): if initializer is None: @@ -76,8 +80,25 @@ class ParamAttr(object): }, 'regularizer': self.regularizer, 'trainable': self.trainable, - 'clip_attr': self.clip + 'gradient_clip_attr': self.gradient_clip } if with_initializer: kwargs['initializer'] = self.initializer return kwargs + + +class WeightNormParamAttr(ParamAttr): + """ + Used for weight normalization. Any field in ParamAttr can also be set here. + Besides, an extra field dim can be set to indicate the dimension except + which to normalize. + """ + # List to record the parameters reparameterized by weight normalization. + # If these parameters are treated as Variable rather than Parameter, + # it can be used to discriminate these parameters and help to serialize + # these paramters for inference. + params_with_weight_norm = [] + + def __init__(self, dim=None, **kwargs): + super(WeightNormParamAttr, self).__init__(**kwargs) + self.dim = dim diff --git a/python/paddle/v2/fluid/profiler.py b/python/paddle/v2/fluid/profiler.py index f049498b9ffbfc915d03673c1d009d0ab5e4f8fc..d4a2cd7eeabecb60699b5be94d89cf7a916749e7 100644 --- a/python/paddle/v2/fluid/profiler.py +++ b/python/paddle/v2/fluid/profiler.py @@ -1,21 +1,22 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -import paddle.v2.fluid.core as core +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import core from contextlib import contextmanager import os -__all__ = ['CudaProfiler'] +__all__ = ['cuda_profiler', 'reset_profiler', 'profiler'] NVPROF_CONFIG = [ "gpustarttimestamp", @@ -62,3 +63,58 @@ def cuda_profiler(output_file, output_mode=None, config=None): # Disables profiler collection. core.nvprof_stop() os.remove(config_file) + + +def reset_profiler(): + """The profiler clear interface. + reset_profiler will clear the previous time record. + """ + core.reset_profiler() + + +@contextmanager +def profiler(state, sorted_key=None): + """The profiler interface. + Different from cuda_profiler, this profiler can be used to profile both CPU + and GPU program. By defalut, it records the CPU and GPU operator kernels, + if you want to profile other program, you can refer the profiling tutorial + to add more records. + + Args: + state (string) : The profiling state, which should be 'CPU' or 'GPU', + telling the profiler to use CPU timer or GPU timer for profiling. + Although users may have already specified the execution place + (CPUPlace/CUDAPlace) in the begining, for flexibility the profiler + would not inherit this place. + sorted_key (string) : If None, the profiling results will be printed + in the order of first end time of events. Otherwise, the profiling + results will be sorted by the this flag. This flag should be one + of 'calls', 'total', 'max', 'min' or 'ave'. + The `calls` means sorting by the number of calls. + The `total` means sorting by the total execution time. + The `max` means sorting by the maximum execution time. + The `min` means sorting by the minimum execution time. + The `ave` means sorting by the average execution time. + """ + + if state not in ['CPU', 'GPU']: + raise ValueError("The state must be 'CPU' or 'GPU'.") + prof_state = core.ProfilerState.kCUDA if state == "GPU" else core.ProfilerState.kCPU + core.enable_profiler(prof_state) + yield + + if sorted_key not in ['calls', 'total', 'max', 'min', 'ave']: + raise ValueError("The state must be in 'calls', 'total', " + "'max', 'min', 'ave'") + sorted_key = 'default' if sorted_key is None else sorted_key + key_map = { + 'default': core.EventSortingKey.kDefault, + 'calls': core.EventSortingKey.kCalls, + 'total': core.EventSortingKey.kTotal, + 'max': core.EventSortingKey.kMax, + 'min': core.EventSortingKey.kMin, + 'ave': core.EventSortingKey.kAve, + } + # TODO(qingqing) : redirect C++ ostream to Python stream. + # with core.ostream_redirect(stdout=True, stderr=True): + core.disable_profiler(key_map[sorted_key]) diff --git a/python/paddle/v2/fluid/regularizer.py b/python/paddle/v2/fluid/regularizer.py index e53dee98fd025df8aa5ff2b74d3bdfd901402965..0273da647afb6e95a136b5ecd0975347d9a378ff 100644 --- a/python/paddle/v2/fluid/regularizer.py +++ b/python/paddle/v2/fluid/regularizer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import framework __all__ = [ @@ -86,6 +87,11 @@ class WeightDecayRegularizer(object): """ raise NotImplementedError() + def __str__(self): + """Debug string + """ + raise NotImplementedError() + class L2DecayRegularizer(WeightDecayRegularizer): """Implements the L2 Weight Decay Regularization @@ -122,6 +128,9 @@ class L2DecayRegularizer(WeightDecayRegularizer): return decay + def __str__(self): + return "L2Decay, regularization_coeff=%f" % self._regularization_coeff + class L1DecayRegularizer(WeightDecayRegularizer): """Implements the L1 Weight Decay Regularization @@ -162,6 +171,9 @@ class L1DecayRegularizer(WeightDecayRegularizer): return decay + def __str__(self): + return "L1Decay, regularization_coeff=%f" % self._regularization_coeff + # We short the class name, since users will use the regulaizer with the package # name. The sample code: diff --git a/python/paddle/v2/fluid/tests/CMakeLists.txt b/python/paddle/v2/fluid/tests/CMakeLists.txt index 9a0240cbf65c7a79e29babc2abcb157ada684c5e..628ce60b406d880d961d705a6abd2b5236fb1c8c 100644 --- a/python/paddle/v2/fluid/tests/CMakeLists.txt +++ b/python/paddle/v2/fluid/tests/CMakeLists.txt @@ -1,8 +1,14 @@ file(GLOB TEST_OPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "test_*.py") string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}") + +if(NOT WITH_DISTRIBUTE) + list(REMOVE_ITEM TEST_OPS test_recv_op) +endif(NOT WITH_DISTRIBUTE) + foreach(src ${TEST_OPS}) py_test(${src} SRCS ${src}.py) endforeach() add_subdirectory(book) add_subdirectory(book_distribute) +add_subdirectory(book_memory_optimization) diff --git a/python/paddle/v2/fluid/tests/__init__.py b/python/paddle/v2/fluid/tests/__init__.py index 2619c1c0e9db17c38ccc6e1dd010bd9c1c5966bd..b94a21a7e406b833797f8f521c62a2351c2bc30a 100644 --- a/python/paddle/v2/fluid/tests/__init__.py +++ b/python/paddle/v2/fluid/tests/__init__.py @@ -1,13 +1,13 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/python/paddle/v2/fluid/tests/book/CMakeLists.txt b/python/paddle/v2/fluid/tests/book/CMakeLists.txt index a35abe3e0c436be4eaed01c9b9183344c6d3b275..dda02c03fd531445c1b33b39a6ded10921991d9c 100644 --- a/python/paddle/v2/fluid/tests/book/CMakeLists.txt +++ b/python/paddle/v2/fluid/tests/book/CMakeLists.txt @@ -1,9 +1,33 @@ file(GLOB TEST_OPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "test_*.py") string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}") -list(REMOVE_ITEM TEST_OPS test_image_classification_train) +list(REMOVE_ITEM TEST_OPS test_image_classification_train test_recognize_digits) py_test(test_image_classification_train_resnet SRCS test_image_classification_train.py ARGS resnet) py_test(test_image_classification_train_vgg SRCS test_image_classification_train.py ARGS vgg) +py_test(test_recognize_digits_mlp_cpu + SRCS test_recognize_digits.py + ARGS mlp) +py_test(test_recognize_digits_mlp_cuda + SRCS test_recognize_digits.py + ARGS mlp --use_cuda) +py_test(test_recognize_digits_conv_cpu + SRCS test_recognize_digits.py + ARGS conv) +py_test(test_recognize_digits_conv_cuda + SRCS test_recognize_digits.py + ARGS conv --use_cuda) +py_test(test_recognize_digits_mlp_cpu_parallel + SRCS test_recognize_digits.py + ARGS mlp --parallel) +py_test(test_recognize_digits_mlp_cuda_parallel + SRCS test_recognize_digits.py + ARGS mlp --use_cuda --parallel) +py_test(test_recognize_digits_conv_cpu_parallel + SRCS test_recognize_digits.py + ARGS conv --parallel) +py_test(test_recognize_digits_conv_cuda_parallel + SRCS test_recognize_digits.py + ARGS conv --use_cuda --parallel) # default test foreach(src ${TEST_OPS}) diff --git a/v1_api_demo/model_zoo/resnet/example/__init__.py b/python/paddle/v2/fluid/tests/book/__init__.py similarity index 89% rename from v1_api_demo/model_zoo/resnet/example/__init__.py rename to python/paddle/v2/fluid/tests/book/__init__.py index f662d6826321eb840739382558f76327d27b5847..b94a21a7e406b833797f8f521c62a2351c2bc30a 100644 --- a/v1_api_demo/model_zoo/resnet/example/__init__.py +++ b/python/paddle/v2/fluid/tests/book/__init__.py @@ -1,4 +1,4 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. diff --git a/python/paddle/v2/fluid/tests/book/test_fit_a_line.py b/python/paddle/v2/fluid/tests/book/test_fit_a_line.py index 904df66dc1869ca4069a3f2e8dbce850c08e4253..0b954c60b6bc2d721c0373243e747056f8f572cf 100644 --- a/python/paddle/v2/fluid/tests/book/test_fit_a_line.py +++ b/python/paddle/v2/fluid/tests/book/test_fit_a_line.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import paddle.v2 as paddle import paddle.v2.fluid as fluid @@ -48,7 +49,7 @@ for pass_id in range(PASS_NUM): avg_loss_value, = exe.run(fluid.default_main_program(), feed=feeder.feed(data), fetch_list=[avg_cost]) - + print(avg_loss_value) if avg_loss_value[0] < 10.0: exit(0) # if avg cost less than 10.0, we think our code is good. exit(1) diff --git a/python/paddle/v2/fluid/tests/book/test_image_classification_train.py b/python/paddle/v2/fluid/tests/book/test_image_classification_train.py index a06486aa08733a589ac9f0c7b65bb8e769eedcb1..30582a21d0a5eeab125f3a2764b45b51aa4f94b6 100644 --- a/python/paddle/v2/fluid/tests/book/test_image_classification_train.py +++ b/python/paddle/v2/fluid/tests/book/test_image_classification_train.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import sys diff --git a/python/paddle/v2/fluid/tests/book/test_label_semantic_roles.py b/python/paddle/v2/fluid/tests/book/test_label_semantic_roles.py index 42971da0f042e970f7f657e4fd3a66ad7ecf0dc1..f85768de99adb8b5005b23278ad807a24c5bff65 100644 --- a/python/paddle/v2/fluid/tests/book/test_label_semantic_roles.py +++ b/python/paddle/v2/fluid/tests/book/test_label_semantic_roles.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import math import numpy as np @@ -33,7 +34,7 @@ mix_hidden_lr = 1e-3 IS_SPARSE = True PASS_NUM = 10 -BATCH_SIZE = 20 +BATCH_SIZE = 10 embedding_name = 'emb' @@ -174,7 +175,7 @@ def main(): paddle.reader.shuffle( paddle.dataset.conll05.test(), buf_size=8192), batch_size=BATCH_SIZE) - #place = fluid.CPUPlace() + # place = fluid.CPUPlace() place = fluid.CUDAPlace(0) feeder = fluid.DataFeeder( feed_list=[ diff --git a/python/paddle/v2/fluid/tests/book/test_machine_translation.py b/python/paddle/v2/fluid/tests/book/test_machine_translation.py index deeb6b1badc9e12caf4c95f949c3c622b04cf8b4..82b760d693560dae1ab1fa39afdc186f60423e65 100644 --- a/python/paddle/v2/fluid/tests/book/test_machine_translation.py +++ b/python/paddle/v2/fluid/tests/book/test_machine_translation.py @@ -1,22 +1,23 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import paddle.v2 as paddle import paddle.v2.fluid as fluid import paddle.v2.fluid.core as core import paddle.v2.fluid.framework as framework -import paddle.v2.fluid.layers as layers +import paddle.v2.fluid.layers as pd from paddle.v2.fluid.executor import Executor dict_size = 30000 @@ -25,53 +26,136 @@ src_dict, trg_dict = paddle.dataset.wmt14.get_dict(dict_size) hidden_dim = 32 word_dim = 16 IS_SPARSE = True -batch_size = 10 -max_length = 50 +batch_size = 2 +max_length = 8 topk_size = 50 trg_dic_size = 10000 +beam_size = 2 decoder_size = hidden_dim +place = core.CPUPlace() + -def encoder_decoder(): +def encoder(): # encoder - src_word_id = layers.data( + src_word_id = pd.data( name="src_word_id", shape=[1], dtype='int64', lod_level=1) - src_embedding = layers.embedding( + src_embedding = pd.embedding( input=src_word_id, size=[dict_size, word_dim], dtype='float32', is_sparse=IS_SPARSE, param_attr=fluid.ParamAttr(name='vemb')) - fc1 = fluid.layers.fc(input=src_embedding, size=hidden_dim * 4, act='tanh') - lstm_hidden0, lstm_0 = layers.dynamic_lstm(input=fc1, size=hidden_dim * 4) - encoder_out = layers.sequence_last_step(input=lstm_hidden0) + fc1 = pd.fc(input=src_embedding, size=hidden_dim * 4, act='tanh') + lstm_hidden0, lstm_0 = pd.dynamic_lstm(input=fc1, size=hidden_dim * 4) + encoder_out = pd.sequence_last_step(input=lstm_hidden0) + return encoder_out + +def decoder_train(context): # decoder - trg_language_word = layers.data( + trg_language_word = pd.data( name="target_language_word", shape=[1], dtype='int64', lod_level=1) - trg_embedding = layers.embedding( + trg_embedding = pd.embedding( input=trg_language_word, size=[dict_size, word_dim], dtype='float32', is_sparse=IS_SPARSE, param_attr=fluid.ParamAttr(name='vemb')) - rnn = fluid.layers.DynamicRNN() + rnn = pd.DynamicRNN() with rnn.block(): current_word = rnn.step_input(trg_embedding) - mem = rnn.memory(init=encoder_out) - fc1 = fluid.layers.fc(input=[current_word, mem], + pre_state = rnn.memory(init=context) + current_state = pd.fc(input=[current_word, pre_state], size=decoder_size, act='tanh') - out = fluid.layers.fc(input=fc1, size=target_dict_dim, act='softmax') - rnn.update_memory(mem, fc1) - rnn.output(out) + + current_score = pd.fc(input=current_state, + size=target_dict_dim, + act='softmax') + rnn.update_memory(pre_state, current_state) + rnn.output(current_score) return rnn() +def decoder_decode(context): + init_state = context + array_len = pd.fill_constant(shape=[1], dtype='int64', value=max_length) + counter = pd.zeros(shape=[1], dtype='int64') + + # fill the first element with init_state + state_array = pd.create_array('float32') + pd.array_write(init_state, array=state_array, i=counter) + + # ids, scores as memory + ids_array = pd.create_array('int64') + scores_array = pd.create_array('float32') + + init_ids = pd.data(name="init_ids", shape=[1], dtype="int64", lod_level=2) + init_scores = pd.data( + name="init_scores", shape=[1], dtype="float32", lod_level=2) + + pd.array_write(init_ids, array=ids_array, i=counter) + pd.array_write(init_scores, array=scores_array, i=counter) + + cond = pd.less_than(x=counter, y=array_len) + + while_op = pd.While(cond=cond) + with while_op.block(): + pre_ids = pd.array_read(array=ids_array, i=counter) + pre_state = pd.array_read(array=state_array, i=counter) + pre_score = pd.array_read(array=scores_array, i=counter) + + # expand the lod of pre_state to be the same with pre_score + pre_state_expanded = pd.sequence_expand(pre_state, pre_score) + + pre_ids_emb = pd.embedding( + input=pre_ids, + size=[dict_size, word_dim], + dtype='float32', + is_sparse=IS_SPARSE) + + # use rnn unit to update rnn + current_state = pd.fc(input=[pre_ids_emb, pre_state_expanded], + size=decoder_size, + act='tanh') + + # use score to do beam search + current_score = pd.fc(input=current_state, + size=target_dict_dim, + act='softmax') + topk_scores, topk_indices = pd.topk(current_score, k=50) + selected_ids, selected_scores = pd.beam_search( + pre_ids, topk_indices, topk_scores, beam_size, end_id=10, level=0) + + pd.increment(x=counter, value=1, in_place=True) + + # update the memories + pd.array_write(current_state, array=state_array, i=counter) + pd.array_write(selected_ids, array=ids_array, i=counter) + pd.array_write(selected_scores, array=scores_array, i=counter) + + pd.less_than(x=counter, y=array_len, cond=cond) + + translation_ids, translation_scores = pd.beam_search_decode( + ids=ids_array, scores=scores_array) + + # return init_ids, init_scores + + return translation_ids, translation_scores + + +def set_init_lod(data, lod, place): + res = core.LoDTensor() + res.set(data, place) + res.set_lod(lod) + return res + + def to_lodtensor(data, place): seq_lens = [len(seq) for seq in data] cur_len = 0 @@ -87,12 +171,13 @@ def to_lodtensor(data, place): return res -def main(): - rnn_out = encoder_decoder() - label = layers.data( +def train_main(): + context = encoder() + rnn_out = decoder_train(context) + label = pd.data( name="target_language_next_word", shape=[1], dtype='int64', lod_level=1) - cost = layers.cross_entropy(input=rnn_out, label=label) - avg_cost = fluid.layers.mean(x=cost) + cost = pd.cross_entropy(input=rnn_out, label=label) + avg_cost = pd.mean(x=cost) optimizer = fluid.optimizer.Adagrad(learning_rate=1e-4) optimizer.minimize(avg_cost) @@ -102,13 +187,12 @@ def main(): paddle.dataset.wmt14.train(dict_size), buf_size=1000), batch_size=batch_size) - place = core.CPUPlace() exe = Executor(place) exe.run(framework.default_startup_program()) batch_id = 0 - for pass_id in xrange(2): + for pass_id in xrange(1): for data in train_data(): word_data = to_lodtensor(map(lambda x: x[0], data), place) trg_word = to_lodtensor(map(lambda x: x[1], data), place) @@ -124,9 +208,48 @@ def main(): print('pass_id=' + str(pass_id) + ' batch=' + str(batch_id) + " avg_cost=" + str(avg_cost_val)) if batch_id > 3: - exit(0) + break batch_id += 1 +def decode_main(): + context = encoder() + translation_ids, translation_scores = decoder_decode(context) + + exe = Executor(place) + exe.run(framework.default_startup_program()) + + init_ids_data = np.array([1 for _ in range(batch_size)], dtype='int64') + init_scores_data = np.array( + [1. for _ in range(batch_size)], dtype='float32') + init_ids_data = init_ids_data.reshape((batch_size, 1)) + init_scores_data = init_scores_data.reshape((batch_size, 1)) + init_lod = [i for i in range(batch_size)] + [batch_size] + init_lod = [init_lod, init_lod] + + train_data = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.wmt14.train(dict_size), buf_size=1000), + batch_size=batch_size) + for _, data in enumerate(train_data()): + init_ids = set_init_lod(init_ids_data, init_lod, place) + init_scores = set_init_lod(init_scores_data, init_lod, place) + + src_word_data = to_lodtensor(map(lambda x: x[0], data), place) + + result_ids, result_scores = exe.run( + framework.default_main_program(), + feed={ + 'src_word_id': src_word_data, + 'init_ids': init_ids, + 'init_scores': init_scores + }, + fetch_list=[translation_ids, translation_scores], + return_numpy=False) + print result_ids.lod() + break + + if __name__ == '__main__': - main() + # train_main() + decode_main() diff --git a/python/paddle/v2/fluid/tests/book/test_recognize_digits.py b/python/paddle/v2/fluid/tests/book/test_recognize_digits.py new file mode 100644 index 0000000000000000000000000000000000000000..b4b6020f58e7538dfe0f98c17d61f3614c3c6fc4 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book/test_recognize_digits.py @@ -0,0 +1,184 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +from __future__ import print_function +import argparse +import paddle.v2.fluid as fluid +import paddle.v2 as paddle +import sys +import numpy + + +def parse_arg(): + parser = argparse.ArgumentParser() + parser.add_argument( + "nn_type", + help="The neural network type, in ['mlp', 'conv']", + type=str, + choices=['mlp', 'conv']) + parser.add_argument( + "--parallel", + help='Run in parallel or not', + default=False, + action="store_true") + parser.add_argument( + "--use_cuda", + help="Run the program by using CUDA", + default=False, + action="store_true") + return parser.parse_args() + + +BATCH_SIZE = 64 + + +def loss_net(hidden, label): + prediction = fluid.layers.fc(input=hidden, size=10, act='softmax') + loss = fluid.layers.cross_entropy(input=prediction, label=label) + avg_loss = fluid.layers.mean(x=loss) + acc = fluid.layers.accuracy(input=prediction, label=label) + return prediction, avg_loss, acc + + +def mlp(img, label): + hidden = fluid.layers.fc(input=img, size=200, act='tanh') + hidden = fluid.layers.fc(input=hidden, size=200, act='tanh') + return loss_net(hidden, label) + + +def conv_net(img, label): + conv_pool_1 = fluid.nets.simple_img_conv_pool( + input=img, + filter_size=5, + num_filters=20, + pool_size=2, + pool_stride=2, + act="relu") + conv_pool_2 = fluid.nets.simple_img_conv_pool( + input=conv_pool_1, + filter_size=5, + num_filters=50, + pool_size=2, + pool_stride=2, + act="relu") + return loss_net(conv_pool_2, label) + + +def train(args, save_dirname=None): + print("recognize digits with args: {0}".format(" ".join(sys.argv[1:]))) + + img = fluid.layers.data(name='img', shape=[1, 28, 28], dtype='float32') + label = fluid.layers.data(name='label', shape=[1], dtype='int64') + + if args.nn_type == 'mlp': + net_conf = mlp + else: + net_conf = conv_net + + if args.parallel: + places = fluid.layers.get_places() + pd = fluid.layers.ParallelDo(places) + with pd.do(): + img_ = pd.read_input(img) + label_ = pd.read_input(label) + prediction, avg_loss, acc = net_conf(img_, label_) + for o in [avg_loss, acc]: + pd.write_output(o) + + avg_loss, acc = pd() + # get mean loss and acc through every devices. + avg_loss = fluid.layers.mean(x=avg_loss) + acc = fluid.layers.mean(x=acc) + else: + prediction, avg_loss, acc = net_conf(img, label) + + test_program = fluid.default_main_program().clone() + + optimizer = fluid.optimizer.Adam(learning_rate=0.001) + optimizer.minimize(avg_loss) + + place = fluid.CUDAPlace(0) if args.use_cuda else fluid.CPUPlace() + + exe = fluid.Executor(place) + exe.run(fluid.default_startup_program()) + + train_reader = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.mnist.train(), buf_size=500), + batch_size=BATCH_SIZE) + test_reader = paddle.batch( + paddle.dataset.mnist.test(), batch_size=BATCH_SIZE) + feeder = fluid.DataFeeder(feed_list=[img, label], place=place) + + PASS_NUM = 100 + for pass_id in range(PASS_NUM): + for batch_id, data in enumerate(train_reader()): + # train a mini-batch, fetch nothing + exe.run(feed=feeder.feed(data)) + if (batch_id + 1) % 10 == 0: + acc_set = [] + avg_loss_set = [] + for test_data in test_reader(): + acc_np, avg_loss_np = exe.run(program=test_program, + feed=feeder.feed(test_data), + fetch_list=[acc, avg_loss]) + acc_set.append(float(acc_np)) + avg_loss_set.append(float(avg_loss_np)) + # get test acc and loss + acc_val = numpy.array(acc_set).mean() + avg_loss_val = numpy.array(avg_loss_set).mean() + if float(acc_val) > 0.85: # test acc > 85% + if save_dirname is not None: + fluid.io.save_inference_model(save_dirname, ["img"], + [prediction], exe) + return + else: + print( + 'PassID {0:1}, BatchID {1:04}, Test Loss {2:2.2}, Acc {3:2.2}'. + format(pass_id, batch_id + 1, + float(avg_loss_val), float(acc_val))) + + +def infer(args, save_dirname=None): + if save_dirname is None: + return + + place = fluid.CUDAPlace(0) if args.use_cuda else fluid.CPUPlace() + exe = fluid.Executor(place) + + # Use fluid.io.load_inference_model to obtain the inference program desc, + # the feed_target_names (the names of variables that will be feeded + # data using feed operators), and the fetch_targets (variables that + # we want to obtain data from using fetch operators). + [inference_program, feed_target_names, + fetch_targets] = fluid.io.load_inference_model(save_dirname, exe) + + # The input's dimension of conv should be 4-D or 5-D. + tensor_img = numpy.random.rand(1, 1, 28, 28).astype("float32") + + # Construct feed as a dictionary of {feed_target_name: feed_target_data} + # and results will contain a list of data corresponding to fetch_targets. + results = exe.run(inference_program, + feed={feed_target_names[0]: tensor_img}, + fetch_list=fetch_targets) + print("infer results: ", results[0]) + + +if __name__ == '__main__': + args = parse_arg() + if not args.use_cuda and not args.parallel: + save_dirname = "recognize_digits_" + args.nn_type + ".inference.model" + else: + save_dirname = None + train(args, save_dirname) + infer(args, save_dirname) diff --git a/python/paddle/v2/fluid/tests/book/test_recognize_digits_conv.py b/python/paddle/v2/fluid/tests/book/test_recognize_digits_conv.py deleted file mode 100644 index 1d5defbed332e8c4c989e6c1f236836bb4b0a3f9..0000000000000000000000000000000000000000 --- a/python/paddle/v2/fluid/tests/book/test_recognize_digits_conv.py +++ /dev/null @@ -1,73 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from __future__ import print_function -import numpy as np -import paddle.v2 as paddle -import paddle.v2.fluid as fluid - -images = fluid.layers.data(name='pixel', shape=[1, 28, 28], dtype='float32') -label = fluid.layers.data(name='label', shape=[1], dtype='int64') -conv_pool_1 = fluid.nets.simple_img_conv_pool( - input=images, - filter_size=5, - num_filters=20, - pool_size=2, - pool_stride=2, - act="relu") -conv_pool_2 = fluid.nets.simple_img_conv_pool( - input=conv_pool_1, - filter_size=5, - num_filters=50, - pool_size=2, - pool_stride=2, - act="relu") - -predict = fluid.layers.fc(input=conv_pool_2, size=10, act="softmax") -cost = fluid.layers.cross_entropy(input=predict, label=label) -avg_cost = fluid.layers.mean(x=cost) -optimizer = fluid.optimizer.Adam(learning_rate=0.01) -optimizer.minimize(avg_cost) - -accuracy = fluid.evaluator.Accuracy(input=predict, label=label) - -BATCH_SIZE = 50 -PASS_NUM = 3 -train_reader = paddle.batch( - paddle.reader.shuffle( - paddle.dataset.mnist.train(), buf_size=500), - batch_size=BATCH_SIZE) - -place = fluid.CPUPlace() -exe = fluid.Executor(place) -feeder = fluid.DataFeeder(feed_list=[images, label], place=place) -exe.run(fluid.default_startup_program()) - -for pass_id in range(PASS_NUM): - accuracy.reset(exe) - for data in train_reader(): - loss, acc = exe.run(fluid.default_main_program(), - feed=feeder.feed(data), - fetch_list=[avg_cost] + accuracy.metrics) - pass_acc = accuracy.eval(exe) - print("pass_id=" + str(pass_id) + " acc=" + str(acc) + " pass_acc=" + - str(pass_acc)) - # print loss, acc - if loss < 10.0 and pass_acc > 0.9: - # if avg cost less than 10.0 and accuracy is larger than 0.9, we think our code is good. - exit(0) - - pass_acc = accuracy.eval(exe) - print("pass_id=" + str(pass_id) + " pass_acc=" + str(pass_acc)) - -exit(1) diff --git a/python/paddle/v2/fluid/tests/book/test_recognize_digits_mlp.py b/python/paddle/v2/fluid/tests/book/test_recognize_digits_mlp.py deleted file mode 100644 index 02da2fcc8544d0f3ccfefa3c88af2f7297d1c76f..0000000000000000000000000000000000000000 --- a/python/paddle/v2/fluid/tests/book/test_recognize_digits_mlp.py +++ /dev/null @@ -1,95 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from __future__ import print_function -import numpy as np -import paddle.v2 as paddle -import paddle.v2.fluid as fluid - -BATCH_SIZE = 128 -image = fluid.layers.data(name='x', shape=[784], dtype='float32') - -regularizer = fluid.regularizer.L2Decay(0.0005 * BATCH_SIZE) - -hidden1 = fluid.layers.fc(input=image, - size=128, - act='relu', - param_attr=fluid.ParamAttr( - regularizer=regularizer, - clip=fluid.clip.ClipByValue(10))) - -hidden2 = fluid.layers.fc(input=hidden1, - size=64, - act='relu', - param_attr=regularizer) - -predict = fluid.layers.fc(input=hidden2, - size=10, - act='softmax', - param_attr=regularizer) - -label = fluid.layers.data(name='y', shape=[1], dtype='int64') - -cost = fluid.layers.cross_entropy(input=predict, label=label) -avg_cost = fluid.layers.mean(x=cost) - -optimizer = fluid.optimizer.Momentum(learning_rate=0.001, momentum=0.9) -opts = optimizer.minimize(avg_cost) - -accuracy = fluid.evaluator.Accuracy(input=predict, label=label) - -inference_program = fluid.default_main_program().clone() -with fluid.program_guard(inference_program): - test_accuracy = fluid.evaluator.Accuracy(input=predict, label=label) - test_target = [avg_cost] + test_accuracy.metrics + test_accuracy.states - inference_program = fluid.io.get_inference_program(test_target) - -train_reader = paddle.batch( - paddle.reader.shuffle( - paddle.dataset.mnist.train(), buf_size=8192), - batch_size=BATCH_SIZE) - -test_reader = paddle.batch(paddle.dataset.mnist.test(), batch_size=128) - -place = fluid.CPUPlace() -exe = fluid.Executor(place) -feeder = fluid.DataFeeder(feed_list=[image, label], place=place) -exe.run(fluid.default_startup_program()) - -PASS_NUM = 100 -for pass_id in range(PASS_NUM): - accuracy.reset(exe) - for data in train_reader(): - out, acc = exe.run(fluid.default_main_program(), - feed=feeder.feed(data), - fetch_list=[avg_cost] + accuracy.metrics) - pass_acc = accuracy.eval(exe) - - test_accuracy.reset(exe) - for data in test_reader(): - out, acc = exe.run(inference_program, - feed=feeder.feed(data), - fetch_list=[avg_cost] + test_accuracy.metrics) - - test_pass_acc = test_accuracy.eval(exe) - print("pass_id=" + str(pass_id) + " train_cost=" + str( - out) + " train_acc=" + str(acc) + " train_pass_acc=" + str(pass_acc) - + " test_acc=" + str(test_pass_acc)) - - if test_pass_acc > 0.7: - fluid.io.save_inference_model( - "./recognize_digits_mlp.inference.model/", ["x"], [predict], - exe) - exit(0) - -exit(1) diff --git a/python/paddle/v2/fluid/tests/book/test_recommender_system.py b/python/paddle/v2/fluid/tests/book/test_recommender_system.py index 47e2afcd83be12c37d95574afff7ccd2e8a781a6..d4a694e5721415fd9c953a83d927b25b80f5fb47 100644 --- a/python/paddle/v2/fluid/tests/book/test_recommender_system.py +++ b/python/paddle/v2/fluid/tests/book/test_recommender_system.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import paddle.v2 as paddle import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/book/test_rnn_encoder_decoder.py b/python/paddle/v2/fluid/tests/book/test_rnn_encoder_decoder.py new file mode 100644 index 0000000000000000000000000000000000000000..fdc60861760163d2ebad3b050e551929321baafd --- /dev/null +++ b/python/paddle/v2/fluid/tests/book/test_rnn_encoder_decoder.py @@ -0,0 +1,204 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import numpy as np +import paddle.v2 as paddle +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import paddle.v2.fluid.framework as framework +import paddle.v2.fluid.layers as layers +from paddle.v2.fluid.executor import Executor + +dict_size = 30000 +source_dict_dim = target_dict_dim = dict_size +src_dict, trg_dict = paddle.dataset.wmt14.get_dict(dict_size) +hidden_dim = 32 +embedding_dim = 16 +batch_size = 10 +max_length = 50 +topk_size = 50 +encoder_size = decoder_size = hidden_dim +IS_SPARSE = True +USE_PEEPHOLES = False + + +def bi_lstm_encoder(input_seq, hidden_size): + input_forward_proj = fluid.layers.fc(input=input_seq, + size=hidden_size * 4, + bias_attr=True) + forward, _ = fluid.layers.dynamic_lstm( + input=input_forward_proj, + size=hidden_size * 4, + use_peepholes=USE_PEEPHOLES) + input_backward_proj = fluid.layers.fc(input=input_seq, + size=hidden_size * 4, + bias_attr=True) + backward, _ = fluid.layers.dynamic_lstm( + input=input_backward_proj, + size=hidden_size * 4, + is_reverse=True, + use_peepholes=USE_PEEPHOLES) + + forward_last = fluid.layers.sequence_last_step(input=forward) + backward_first = fluid.layers.sequence_first_step(input=backward) + + return forward_last, backward_first + + +# FIXME(peterzhang2029): Replace this function with the lstm_unit_op. +def lstm_step(x_t, hidden_t_prev, cell_t_prev, size): + def linear(inputs): + return fluid.layers.fc(input=inputs, size=size, bias_attr=True) + + forget_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t])) + input_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t])) + output_gate = fluid.layers.sigmoid(x=linear([hidden_t_prev, x_t])) + cell_tilde = fluid.layers.tanh(x=linear([hidden_t_prev, x_t])) + + cell_t = fluid.layers.sums(input=[ + fluid.layers.elementwise_mul( + x=forget_gate, y=cell_t_prev), fluid.layers.elementwise_mul( + x=input_gate, y=cell_tilde) + ]) + + hidden_t = fluid.layers.elementwise_mul( + x=output_gate, y=fluid.layers.tanh(x=cell_t)) + + return hidden_t, cell_t + + +def lstm_decoder_without_attention(target_embedding, decoder_boot, context, + decoder_size): + rnn = fluid.layers.DynamicRNN() + + cell_init = fluid.layers.fill_constant_batch_size_like( + input=decoder_boot, + value=0.0, + shape=[-1, decoder_size], + dtype='float32') + cell_init.stop_gradient = False + + with rnn.block(): + current_word = rnn.step_input(target_embedding) + context = rnn.static_input(context) + + hidden_mem = rnn.memory(init=decoder_boot, need_reorder=True) + cell_mem = rnn.memory(init=cell_init) + decoder_inputs = fluid.layers.concat( + input=[context, current_word], axis=1) + h, c = lstm_step(decoder_inputs, hidden_mem, cell_mem, decoder_size) + rnn.update_memory(hidden_mem, h) + rnn.update_memory(cell_mem, c) + out = fluid.layers.fc(input=h, + size=target_dict_dim, + bias_attr=True, + act='softmax') + rnn.output(out) + return rnn() + + +def seq_to_seq_net(): + """Construct a seq2seq network.""" + + src_word_idx = fluid.layers.data( + name='source_sequence', shape=[1], dtype='int64', lod_level=1) + + src_embedding = fluid.layers.embedding( + input=src_word_idx, + size=[source_dict_dim, embedding_dim], + dtype='float32') + + src_forward_last, src_backward_first = bi_lstm_encoder( + input_seq=src_embedding, hidden_size=encoder_size) + + encoded_vector = fluid.layers.concat( + input=[src_forward_last, src_backward_first], axis=1) + + decoder_boot = fluid.layers.fc(input=src_backward_first, + size=decoder_size, + bias_attr=False, + act='tanh') + + trg_word_idx = fluid.layers.data( + name='target_sequence', shape=[1], dtype='int64', lod_level=1) + + trg_embedding = fluid.layers.embedding( + input=trg_word_idx, + size=[target_dict_dim, embedding_dim], + dtype='float32') + + prediction = lstm_decoder_without_attention(trg_embedding, decoder_boot, + encoded_vector, decoder_size) + label = fluid.layers.data( + name='label_sequence', shape=[1], dtype='int64', lod_level=1) + cost = fluid.layers.cross_entropy(input=prediction, label=label) + avg_cost = fluid.layers.mean(x=cost) + + return avg_cost + + +def to_lodtensor(data, place): + seq_lens = [len(seq) for seq in data] + cur_len = 0 + lod = [cur_len] + for l in seq_lens: + cur_len += l + lod.append(cur_len) + flattened_data = np.concatenate(data, axis=0).astype("int64") + flattened_data = flattened_data.reshape([len(flattened_data), 1]) + res = core.LoDTensor() + res.set(flattened_data, place) + res.set_lod([lod]) + return res + + +def main(): + avg_cost = seq_to_seq_net() + + optimizer = fluid.optimizer.Adagrad(learning_rate=1e-4) + optimizer.minimize(avg_cost) + + train_data = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.wmt14.train(dict_size), buf_size=1000), + batch_size=batch_size) + + place = core.CPUPlace() + exe = Executor(place) + + exe.run(framework.default_startup_program()) + + batch_id = 0 + for pass_id in xrange(2): + for data in train_data(): + word_data = to_lodtensor(map(lambda x: x[0], data), place) + trg_word = to_lodtensor(map(lambda x: x[1], data), place) + trg_word_next = to_lodtensor(map(lambda x: x[2], data), place) + outs = exe.run(framework.default_main_program(), + feed={ + 'source_sequence': word_data, + 'target_sequence': trg_word, + 'label_sequence': trg_word_next + }, + fetch_list=[avg_cost]) + avg_cost_val = np.array(outs[0]) + print('pass_id=' + str(pass_id) + ' batch=' + str(batch_id) + + " avg_cost=" + str(avg_cost_val)) + if batch_id > 3: + exit(0) + batch_id += 1 + + +if __name__ == '__main__': + main() diff --git a/python/paddle/v2/fluid/tests/book/test_understand_sentiment.py b/python/paddle/v2/fluid/tests/book/test_understand_sentiment.py new file mode 100644 index 0000000000000000000000000000000000000000..2ba9077a26202b1c16cc480823115f7ad55c2c67 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book/test_understand_sentiment.py @@ -0,0 +1,154 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import paddle.v2.fluid as fluid +import paddle.v2 as paddle +import contextlib + + +def convolution_net(data, label, input_dim, class_dim=2, emb_dim=32, + hid_dim=32): + emb = fluid.layers.embedding(input=data, size=[input_dim, emb_dim]) + conv_3 = fluid.nets.sequence_conv_pool( + input=emb, + num_filters=hid_dim, + filter_size=3, + act="tanh", + pool_type="sqrt") + conv_4 = fluid.nets.sequence_conv_pool( + input=emb, + num_filters=hid_dim, + filter_size=4, + act="tanh", + pool_type="sqrt") + prediction = fluid.layers.fc(input=[conv_3, conv_4], + size=class_dim, + act="softmax") + cost = fluid.layers.cross_entropy(input=prediction, label=label) + avg_cost = fluid.layers.mean(x=cost) + adam_optimizer = fluid.optimizer.Adam(learning_rate=0.002) + adam_optimizer.minimize(avg_cost) + accuracy = fluid.layers.accuracy(input=prediction, label=label) + return avg_cost, accuracy + + +def stacked_lstm_net(data, + label, + input_dim, + class_dim=2, + emb_dim=128, + hid_dim=512, + stacked_num=3): + assert stacked_num % 2 == 1 + + emb = fluid.layers.embedding(input=data, size=[input_dim, emb_dim]) + # add bias attr + + # TODO(qijun) linear act + fc1 = fluid.layers.fc(input=emb, size=hid_dim) + lstm1, cell1 = fluid.layers.dynamic_lstm(input=fc1, size=hid_dim) + + inputs = [fc1, lstm1] + + for i in range(2, stacked_num + 1): + fc = fluid.layers.fc(input=inputs, size=hid_dim) + lstm, cell = fluid.layers.dynamic_lstm( + input=fc, size=hid_dim, is_reverse=(i % 2) == 0) + inputs = [fc, lstm] + + fc_last = fluid.layers.sequence_pool(input=inputs[0], pool_type='max') + lstm_last = fluid.layers.sequence_pool(input=inputs[1], pool_type='max') + + prediction = fluid.layers.fc(input=[fc_last, lstm_last], + size=class_dim, + act='softmax') + cost = fluid.layers.cross_entropy(input=prediction, label=label) + avg_cost = fluid.layers.mean(x=cost) + adam_optimizer = fluid.optimizer.Adam(learning_rate=0.002) + adam_optimizer.minimize(avg_cost) + accuracy = fluid.layers.accuracy(input=prediction, label=label) + return avg_cost, accuracy + + +def main(word_dict, net_method, use_cuda): + if use_cuda and not fluid.core.is_compiled_with_cuda(): + return + + BATCH_SIZE = 128 + PASS_NUM = 5 + dict_dim = len(word_dict) + class_dim = 2 + + data = fluid.layers.data( + name="words", shape=[1], dtype="int64", lod_level=1) + label = fluid.layers.data(name="label", shape=[1], dtype="int64") + cost, acc_out = net_method( + data, label, input_dim=dict_dim, class_dim=class_dim) + + train_data = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.imdb.train(word_dict), buf_size=1000), + batch_size=BATCH_SIZE) + place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace() + exe = fluid.Executor(place) + feeder = fluid.DataFeeder(feed_list=[data, label], place=place) + + exe.run(fluid.default_startup_program()) + + for pass_id in xrange(PASS_NUM): + for data in train_data(): + cost_val, acc_val = exe.run(fluid.default_main_program(), + feed=feeder.feed(data), + fetch_list=[cost, acc_out]) + print("cost=" + str(cost_val) + " acc=" + str(acc_val)) + if cost_val < 0.4 and acc_val > 0.8: + return + raise AssertionError("Cost is too large for {0}".format( + net_method.__name__)) + + +class TestUnderstandSentiment(unittest.TestCase): + @classmethod + def setUpClass(cls): + cls.word_dict = paddle.dataset.imdb.word_dict() + + @contextlib.contextmanager + def new_program_scope(self): + prog = fluid.Program() + startup_prog = fluid.Program() + scope = fluid.core.Scope() + with fluid.scope_guard(scope): + with fluid.program_guard(prog, startup_prog): + yield + + def test_conv_cpu(self): + with self.new_program_scope(): + main(self.word_dict, net_method=convolution_net, use_cuda=False) + + def test_stacked_lstm_cpu(self): + with self.new_program_scope(): + main(self.word_dict, net_method=stacked_lstm_net, use_cuda=False) + + def test_conv_gpu(self): + with self.new_program_scope(): + main(self.word_dict, net_method=convolution_net, use_cuda=True) + + def test_stacked_lstm_gpu(self): + with self.new_program_scope(): + main(self.word_dict, net_method=stacked_lstm_net, use_cuda=True) + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/book/test_understand_sentiment_conv.py b/python/paddle/v2/fluid/tests/book/test_understand_sentiment_conv.py deleted file mode 100644 index b44d2b41e36633ed8feb7dd8c0e3884e48ffe917..0000000000000000000000000000000000000000 --- a/python/paddle/v2/fluid/tests/book/test_understand_sentiment_conv.py +++ /dev/null @@ -1,100 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from __future__ import print_function -import numpy as np -import paddle.v2 as paddle -import paddle.v2.fluid as fluid - - -def convolution_net(data, label, input_dim, class_dim=2, emb_dim=32, - hid_dim=32): - emb = fluid.layers.embedding(input=data, size=[input_dim, emb_dim]) - conv_3 = fluid.nets.sequence_conv_pool( - input=emb, - num_filters=hid_dim, - filter_size=3, - act="tanh", - pool_type="sqrt") - conv_4 = fluid.nets.sequence_conv_pool( - input=emb, - num_filters=hid_dim, - filter_size=4, - act="tanh", - pool_type="sqrt") - prediction = fluid.layers.fc(input=[conv_3, conv_4], - size=class_dim, - act="softmax") - cost = fluid.layers.cross_entropy(input=prediction, label=label) - avg_cost = fluid.layers.mean(x=cost) - adam_optimizer = fluid.optimizer.Adam(learning_rate=0.002) - adam_optimizer.minimize(avg_cost) - accuracy = fluid.evaluator.Accuracy(input=prediction, label=label) - return avg_cost, accuracy, accuracy.metrics[0] - - -def to_lodtensor(data, place): - seq_lens = [len(seq) for seq in data] - cur_len = 0 - lod = [cur_len] - for l in seq_lens: - cur_len += l - lod.append(cur_len) - flattened_data = np.concatenate(data, axis=0).astype("int64") - flattened_data = flattened_data.reshape([len(flattened_data), 1]) - res = fluid.LoDTensor() - res.set(flattened_data, place) - res.set_lod([lod]) - return res - - -def main(): - BATCH_SIZE = 100 - PASS_NUM = 5 - - word_dict = paddle.dataset.imdb.word_dict() - dict_dim = len(word_dict) - class_dim = 2 - - data = fluid.layers.data( - name="words", shape=[1], dtype="int64", lod_level=1) - label = fluid.layers.data(name="label", shape=[1], dtype="int64") - cost, accuracy, acc_out = convolution_net( - data, label, input_dim=dict_dim, class_dim=class_dim) - - train_data = paddle.batch( - paddle.reader.shuffle( - paddle.dataset.imdb.train(word_dict), buf_size=1000), - batch_size=BATCH_SIZE) - place = fluid.CPUPlace() - exe = fluid.Executor(place) - feeder = fluid.DataFeeder(feed_list=[data, label], place=place) - - exe.run(fluid.default_startup_program()) - - for pass_id in xrange(PASS_NUM): - accuracy.reset(exe) - for data in train_data(): - cost_val, acc_val = exe.run(fluid.default_main_program(), - feed=feeder.feed(data), - fetch_list=[cost, acc_out]) - pass_acc = accuracy.eval(exe) - print("cost=" + str(cost_val) + " acc=" + str(acc_val) + - " pass_acc=" + str(pass_acc)) - if cost_val < 1.0 and pass_acc > 0.8: - exit(0) - exit(1) - - -if __name__ == '__main__': - main() diff --git a/python/paddle/v2/fluid/tests/book/test_understand_sentiment_lstm.py b/python/paddle/v2/fluid/tests/book/test_understand_sentiment_lstm.py deleted file mode 100644 index fab8a82f85ddeed7131df3777e978cc7c0a1b86e..0000000000000000000000000000000000000000 --- a/python/paddle/v2/fluid/tests/book/test_understand_sentiment_lstm.py +++ /dev/null @@ -1,159 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -import numpy as np -import paddle.v2 as paddle -import paddle.v2.fluid as fluid -from paddle.v2.fluid.layer_helper import LayerHelper - - -def lstm(x, c_pre_init, hidden_dim, forget_bias=None): - """ - This function helps create an operator for the LSTM (Long Short Term - Memory) cell that can be used inside an RNN. - """ - helper = LayerHelper('lstm_unit', **locals()) - rnn = fluid.layers.StaticRNN() - with rnn.step(): - c_pre = rnn.memory(init=c_pre_init) - x_t = rnn.step_input(x) - - before_fc = fluid.layers.concat(input=[x_t, c_pre], axis=1) - after_fc = fluid.layers.fc(input=before_fc, size=hidden_dim * 4) - - dtype = x.dtype - c = helper.create_tmp_variable(dtype) - h = helper.create_tmp_variable(dtype) - - helper.append_op( - type='lstm_unit', - inputs={"X": after_fc, - "C_prev": c_pre}, - outputs={"C": c, - "H": h}, - attrs={"forget_bias": forget_bias}) - - rnn.update_memory(c_pre, c) - rnn.output(h) - - return rnn() - - -def lstm_net(dict_dim, class_dim=2, emb_dim=32, seq_len=80, batch_size=50): - data = fluid.layers.data( - name="words", - shape=[seq_len * batch_size, 1], - append_batch_size=False, - dtype="int64", - lod_level=1) - label = fluid.layers.data( - name="label", - shape=[batch_size, 1], - append_batch_size=False, - dtype="int64") - - emb = fluid.layers.embedding(input=data, size=[dict_dim, emb_dim]) - emb = fluid.layers.reshape(x=emb, shape=[batch_size, seq_len, emb_dim]) - emb = fluid.layers.transpose(x=emb, axis=[1, 0, 2]) - - c_pre_init = fluid.layers.fill_constant( - dtype=emb.dtype, shape=[batch_size, emb_dim], value=0.0) - c_pre_init.stop_gradient = False - layer_1_out = lstm(emb, c_pre_init=c_pre_init, hidden_dim=emb_dim) - layer_1_out = fluid.layers.transpose(x=layer_1_out, axis=[1, 0, 2]) - - prediction = fluid.layers.fc(input=layer_1_out, - size=class_dim, - act="softmax") - cost = fluid.layers.cross_entropy(input=prediction, label=label) - - avg_cost = fluid.layers.mean(x=cost) - adam_optimizer = fluid.optimizer.Adam(learning_rate=0.002) - adam_optimizer.minimize(avg_cost) - acc = fluid.layers.accuracy(input=prediction, label=label) - - return avg_cost, acc - - -def to_lodtensor(data, place): - seq_lens = [len(seq) for seq in data] - cur_len = 0 - lod = [cur_len] - for l in seq_lens: - cur_len += l - lod.append(cur_len) - flattened_data = np.concatenate(data, axis=0).astype("int64") - flattened_data = flattened_data.reshape([len(flattened_data), 1]) - res = fluid.LoDTensor() - res.set(flattened_data, place) - res.set_lod([lod]) - return res - - -def chop_data(data, chop_len=80, batch_size=50): - data = [(x[0][:chop_len], x[1]) for x in data if len(x[0]) >= chop_len] - - return data[:batch_size] - - -def prepare_feed_data(data, place): - tensor_words = to_lodtensor(map(lambda x: x[0], data), place) - - label = np.array(map(lambda x: x[1], data)).astype("int64") - label = label.reshape([len(label), 1]) - tensor_label = fluid.LoDTensor() - tensor_label.set(label, place) - - return tensor_words, tensor_label - - -def main(): - BATCH_SIZE = 100 - PASS_NUM = 5 - - word_dict = paddle.dataset.imdb.word_dict() - print "load word dict successfully" - dict_dim = len(word_dict) - class_dim = 2 - - cost, acc = lstm_net(dict_dim=dict_dim, class_dim=class_dim) - - train_data = paddle.batch( - paddle.reader.shuffle( - paddle.dataset.imdb.train(word_dict), buf_size=BATCH_SIZE * 10), - batch_size=BATCH_SIZE) - place = fluid.CPUPlace() - exe = fluid.Executor(place) - - exe.run(fluid.default_startup_program()) - - for pass_id in xrange(PASS_NUM): - for data in train_data(): - chopped_data = chop_data(data) - tensor_words, tensor_label = prepare_feed_data(chopped_data, place) - - outs = exe.run(fluid.default_main_program(), - feed={"words": tensor_words, - "label": tensor_label}, - fetch_list=[cost, acc]) - cost_val = np.array(outs[0]) - acc_val = np.array(outs[1]) - - print("cost=" + str(cost_val) + " acc=" + str(acc_val)) - if acc_val > 0.7: - exit(0) - exit(1) - - -if __name__ == '__main__': - main() diff --git a/python/paddle/v2/fluid/tests/book/test_word2vec.py b/python/paddle/v2/fluid/tests/book/test_word2vec.py index 3d4bbccd33d7d9d0f492a76e04b98d8f7efac91d..766ba9681d1bb816170e0458f540b32511c02933 100644 --- a/python/paddle/v2/fluid/tests/book/test_word2vec.py +++ b/python/paddle/v2/fluid/tests/book/test_word2vec.py @@ -1,86 +1,156 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -import numpy as np +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2 as paddle import paddle.v2.fluid as fluid +import unittest +import os + + +def main(use_cuda, is_sparse, parallel): + if use_cuda and not fluid.core.is_compiled_with_cuda(): + return + + PASS_NUM = 100 + EMBED_SIZE = 32 + HIDDEN_SIZE = 256 + N = 5 + BATCH_SIZE = 32 + IS_SPARSE = is_sparse + + def __network__(words): + embed_first = fluid.layers.embedding( + input=words[0], + size=[dict_size, EMBED_SIZE], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr='shared_w') + embed_second = fluid.layers.embedding( + input=words[1], + size=[dict_size, EMBED_SIZE], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr='shared_w') + embed_third = fluid.layers.embedding( + input=words[2], + size=[dict_size, EMBED_SIZE], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr='shared_w') + embed_forth = fluid.layers.embedding( + input=words[3], + size=[dict_size, EMBED_SIZE], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr='shared_w') + + concat_embed = fluid.layers.concat( + input=[embed_first, embed_second, embed_third, embed_forth], axis=1) + hidden1 = fluid.layers.fc(input=concat_embed, + size=HIDDEN_SIZE, + act='sigmoid') + predict_word = fluid.layers.fc(input=hidden1, + size=dict_size, + act='softmax') + cost = fluid.layers.cross_entropy(input=predict_word, label=words[4]) + avg_cost = fluid.layers.mean(x=cost) + return avg_cost + + word_dict = paddle.dataset.imikolov.build_dict() + dict_size = len(word_dict) + + first_word = fluid.layers.data(name='firstw', shape=[1], dtype='int64') + second_word = fluid.layers.data(name='secondw', shape=[1], dtype='int64') + third_word = fluid.layers.data(name='thirdw', shape=[1], dtype='int64') + forth_word = fluid.layers.data(name='forthw', shape=[1], dtype='int64') + next_word = fluid.layers.data(name='nextw', shape=[1], dtype='int64') + + if not parallel: + avg_cost = __network__( + [first_word, second_word, third_word, forth_word, next_word]) + else: + places = fluid.layers.get_places() + pd = fluid.layers.ParallelDo(places) + with pd.do(): + avg_cost = __network__( + map(pd.read_input, [ + first_word, second_word, third_word, forth_word, next_word + ])) + pd.write_output(avg_cost) + + avg_cost = fluid.layers.mean(x=pd()) + + sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001) + sgd_optimizer.minimize(avg_cost) + + train_reader = paddle.batch( + paddle.dataset.imikolov.train(word_dict, N), BATCH_SIZE) + + place = fluid.CUDAPlace(0) if use_cuda else fluid.CPUPlace() + exe = fluid.Executor(place) + feeder = fluid.DataFeeder( + feed_list=[first_word, second_word, third_word, forth_word, next_word], + place=place) + + exe.run(fluid.default_startup_program()) + + for pass_id in range(PASS_NUM): + for data in train_reader(): + avg_cost_np = exe.run(fluid.default_main_program(), + feed=feeder.feed(data), + fetch_list=[avg_cost]) + if avg_cost_np[0] < 5.0: + return + raise AssertionError("Cost is too large {0:2.2}".format(avg_cost_np[0])) + + +FULL_TEST = os.getenv('FULL_TEST', + '0').lower() in ['true', '1', 't', 'y', 'yes', 'on'] +SKIP_REASON = "Only run minimum number of tests in CI server, to make CI faster" + + +class W2VTest(unittest.TestCase): + pass + + +def inject_test_method(use_cuda, is_sparse, parallel): + fn_name = "test_{0}_{1}_{2}".format("cuda" if use_cuda else "cpu", "sparse" + if is_sparse else "dense", "parallel" + if parallel else "normal") + + def __impl__(*args, **kwargs): + prog = fluid.Program() + startup_prog = fluid.Program() + scope = fluid.core.Scope() + with fluid.scope_guard(scope): + with fluid.program_guard(prog, startup_prog): + main(use_cuda=use_cuda, is_sparse=is_sparse, parallel=parallel) + + if use_cuda and is_sparse and parallel: + fn = __impl__ + else: + # skip the other test when on CI server + fn = unittest.skipUnless( + condition=FULL_TEST, reason=SKIP_REASON)(__impl__) + + setattr(W2VTest, fn_name, fn) + + +for use_cuda in (False, True): + for is_sparse in (False, True): + for parallel in (False, True): + inject_test_method(use_cuda, is_sparse, parallel) -PASS_NUM = 100 -EMBED_SIZE = 32 -HIDDEN_SIZE = 256 -N = 5 -BATCH_SIZE = 32 -IS_SPARSE = True - -word_dict = paddle.dataset.imikolov.build_dict() -dict_size = len(word_dict) - -first_word = fluid.layers.data(name='firstw', shape=[1], dtype='int64') -second_word = fluid.layers.data(name='secondw', shape=[1], dtype='int64') -third_word = fluid.layers.data(name='thirdw', shape=[1], dtype='int64') -forth_word = fluid.layers.data(name='forthw', shape=[1], dtype='int64') -next_word = fluid.layers.data(name='nextw', shape=[1], dtype='int64') - -embed_first = fluid.layers.embedding( - input=first_word, - size=[dict_size, EMBED_SIZE], - dtype='float32', - is_sparse=IS_SPARSE, - param_attr='shared_w') -embed_second = fluid.layers.embedding( - input=second_word, - size=[dict_size, EMBED_SIZE], - dtype='float32', - is_sparse=IS_SPARSE, - param_attr='shared_w') -embed_third = fluid.layers.embedding( - input=third_word, - size=[dict_size, EMBED_SIZE], - dtype='float32', - is_sparse=IS_SPARSE, - param_attr='shared_w') -embed_forth = fluid.layers.embedding( - input=forth_word, - size=[dict_size, EMBED_SIZE], - dtype='float32', - is_sparse=IS_SPARSE, - param_attr='shared_w') - -concat_embed = fluid.layers.concat( - input=[embed_first, embed_second, embed_third, embed_forth], axis=1) -hidden1 = fluid.layers.fc(input=concat_embed, size=HIDDEN_SIZE, act='sigmoid') -predict_word = fluid.layers.fc(input=hidden1, size=dict_size, act='softmax') -cost = fluid.layers.cross_entropy(input=predict_word, label=next_word) -avg_cost = fluid.layers.mean(x=cost) -sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.001) -sgd_optimizer.minimize(avg_cost) - -train_reader = paddle.batch( - paddle.dataset.imikolov.train(word_dict, N), BATCH_SIZE) - -place = fluid.CPUPlace() -exe = fluid.Executor(place) -feeder = fluid.DataFeeder( - feed_list=[first_word, second_word, third_word, forth_word, next_word], - place=place) - -exe.run(fluid.default_startup_program()) - -for pass_id in range(PASS_NUM): - for data in train_reader(): - avg_cost_np = exe.run(fluid.default_main_program(), - feed=feeder.feed(data), - fetch_list=[avg_cost]) - if avg_cost_np[0] < 5.0: - exit(0) # if avg cost less than 10.0, we think our code is good. -exit(1) +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_fit_a_line.py b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_fit_a_line.py index b886071f94768d3373bbf0e0b7655c924b218645..9774edebfb1de0ae73970d582c620f8a984a4ebf 100644 --- a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_fit_a_line.py +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_fit_a_line.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import paddle.v2 as paddle import paddle.v2.fluid as fluid @@ -53,8 +54,9 @@ if training_role == "PSERVER": if not current_endpoint: print("need env SERVER_ENDPOINT") exit(1) - pserver_prog = t.get_pserver_program(current_endpoint, optimize_ops) - exe.run(fluid.default_startup_program()) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) exe.run(pserver_prog) else: trainer_prog = t.get_trainer_program() @@ -66,10 +68,10 @@ else: fluid.io.save_persistables(exe, "./fit_a_line.model/") fluid.io.load_persistables(exe, "./fit_a_line.model/") for data in train_reader(): - avg_loss_value, = exe.run(trainer_prog, - feed=feeder.feed(data), - fetch_list=[avg_cost]) - + avg_loss_value = exe.run(trainer_prog, + feed=feeder.feed(data), + fetch_list=[avg_cost]) + print("loss:" + str(avg_loss_value)) if avg_loss_value[0] < 10.0: exit(0) exit(1) diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_image_classification.py b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_image_classification.py new file mode 100644 index 0000000000000000000000000000000000000000..298ecfc386b3ae093cf714a41f5072759cb2cf2e --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_image_classification.py @@ -0,0 +1,170 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from __future__ import print_function + +import paddle.v2 as paddle +import paddle.v2.fluid as fluid +import os +import sys + +TRAINERS = 5 +BATCH_SIZE = 128 +PASS_NUM = 100 + + +def resnet_cifar10(input, depth=32): + def conv_bn_layer(input, ch_out, filter_size, stride, padding, act='relu'): + tmp = fluid.layers.conv2d( + input=input, + filter_size=filter_size, + num_filters=ch_out, + stride=stride, + padding=padding, + act=None, + bias_attr=False) + return fluid.layers.batch_norm(input=tmp, act=act) + + def shortcut(input, ch_in, ch_out, stride): + if ch_in != ch_out: + return conv_bn_layer(input, ch_out, 1, stride, 0, None) + else: + return input + + def basicblock(input, ch_in, ch_out, stride): + tmp = conv_bn_layer(input, ch_out, 3, stride, 1) + tmp = conv_bn_layer(tmp, ch_out, 3, 1, 1, act=None) + short = shortcut(input, ch_in, ch_out, stride) + return fluid.layers.elementwise_add(x=tmp, y=short, act='relu') + + def layer_warp(block_func, input, ch_in, ch_out, count, stride): + tmp = block_func(input, ch_in, ch_out, stride) + for i in range(1, count): + tmp = block_func(tmp, ch_out, ch_out, 1) + return tmp + + assert (depth - 2) % 6 == 0 + n = (depth - 2) / 6 + conv1 = conv_bn_layer( + input=input, ch_out=16, filter_size=3, stride=1, padding=1) + res1 = layer_warp(basicblock, conv1, 16, 16, n, 1) + res2 = layer_warp(basicblock, res1, 16, 32, n, 2) + res3 = layer_warp(basicblock, res2, 32, 64, n, 2) + pool = fluid.layers.pool2d( + input=res3, pool_size=8, pool_type='avg', pool_stride=1) + return pool + + +def vgg16_bn_drop(input): + def conv_block(input, num_filter, groups, dropouts): + return fluid.nets.img_conv_group( + input=input, + pool_size=2, + pool_stride=2, + conv_num_filter=[num_filter] * groups, + conv_filter_size=3, + conv_act='relu', + conv_with_batchnorm=True, + conv_batchnorm_drop_rate=dropouts, + pool_type='max') + + conv1 = conv_block(input, 64, 2, [0.3, 0]) + conv2 = conv_block(conv1, 128, 2, [0.4, 0]) + conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0]) + conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0]) + conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0]) + + drop = fluid.layers.dropout(x=conv5, dropout_prob=0.5) + fc1 = fluid.layers.fc(input=drop, size=512, act=None) + bn = fluid.layers.batch_norm(input=fc1, act='relu') + drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5) + fc2 = fluid.layers.fc(input=drop2, size=512, act=None) + return fc2 + + +classdim = 10 +data_shape = [3, 32, 32] + +images = fluid.layers.data(name='pixel', shape=data_shape, dtype='float32') +label = fluid.layers.data(name='label', shape=[1], dtype='int64') + +net_type = "vgg" +if len(sys.argv) >= 2: + net_type = sys.argv[1] + +if net_type == "vgg": + print("training vgg net") + net = vgg16_bn_drop(images) +elif net_type == "resnet": + print("training resnet") + net = resnet_cifar10(images, 32) +else: + raise ValueError("%s network is not supported" % net_type) + +predict = fluid.layers.fc(input=net, size=classdim, act='softmax') +cost = fluid.layers.cross_entropy(input=predict, label=label) +avg_cost = fluid.layers.mean(x=cost) + +optimizer = fluid.optimizer.Adam(learning_rate=0.001) +optimize_ops, params_grads = optimizer.minimize(avg_cost) + +accuracy = fluid.evaluator.Accuracy(input=predict, label=label) + +train_reader = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.cifar.train10(), buf_size=128 * 10), + batch_size=BATCH_SIZE) + +place = fluid.CPUPlace() +feeder = fluid.DataFeeder(place=place, feed_list=[images, label]) +exe = fluid.Executor(place) + +t = fluid.DistributeTranspiler() +# all parameter server endpoints list for spliting parameters +pserver_endpoints = os.getenv("PSERVERS") +# server endpoint for current node +current_endpoint = os.getenv("SERVER_ENDPOINT") +# run as trainer or parameter server +training_role = os.getenv("TRAINING_ROLE", + "TRAINER") # get the training role: trainer/pserver +t.transpile( + optimize_ops, params_grads, pservers=pserver_endpoints, trainers=TRAINERS) + +if training_role == "PSERVER": + if not current_endpoint: + print("need env SERVER_ENDPOINT") + exit(1) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) + exe.run(pserver_prog) +elif training_role == "TRAINER": + trainer_prog = t.get_trainer_program() + exe.run(fluid.default_startup_program()) + + for pass_id in range(PASS_NUM): + accuracy.reset(exe) + for data in train_reader(): + loss, acc = exe.run(trainer_prog, + feed=feeder.feed(data), + fetch_list=[avg_cost] + accuracy.metrics) + pass_acc = accuracy.eval(exe) + print("pass_id:" + str(pass_id) + "loss:" + str(loss) + " pass_acc:" + + str(pass_acc)) + # this model is slow, so if we can train two mini batches, + # we think it works properly. + print("trainer run end") +else: + print("environment var TRAINER_ROLE should be TRAINER os PSERVER") +exit(1) diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_label_semantic_roles.py b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_label_semantic_roles.py index 2b5a098ff253b8a96afba7cd03d7f9998ff400af..08bb67b0a1f53c73b713238ab45ec8055726cf9c 100644 --- a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_label_semantic_roles.py +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_label_semantic_roles.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import math import numpy as np @@ -197,8 +198,9 @@ def main(): if not current_endpoint: print("need env SERVER_ENDPOINT") exit(1) - pserver_prog = t.get_pserver_program(current_endpoint, optimize_ops) - exe.run(fluid.default_startup_program()) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) exe.run(pserver_prog) elif training_role == "TRAINER": trainer_prog = t.get_trainer_program() diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_word2vec.py b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_word2vec.py index dc04af5b7b6ee143847685d6cf4da91747afd3ec..04b3113690fde072ab74893508298b920ab9599e 100644 --- a/python/paddle/v2/fluid/tests/book_distribute/notest_dist_word2vec.py +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_dist_word2vec.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import numpy as np import paddle.v2 as paddle @@ -87,8 +88,9 @@ if training_role == "PSERVER": if not current_endpoint: print("need env SERVER_ENDPOINT") exit(1) - pserver_prog = t.get_pserver_program(current_endpoint, optimize_ops) - exe.run(fluid.default_startup_program()) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) exe.run(pserver_prog) elif training_role == "TRAINER": feeder = fluid.DataFeeder( diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_machine_translation.py b/python/paddle/v2/fluid/tests/book_distribute/notest_machine_translation.py new file mode 100644 index 0000000000000000000000000000000000000000..adeacd4adf2150e0302965d80457e26d07c6b96d --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_machine_translation.py @@ -0,0 +1,157 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import numpy as np +import paddle.v2 as paddle +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import paddle.v2.fluid.framework as framework +import paddle.v2.fluid.layers as layers +from paddle.v2.fluid.executor import Executor +import os + +dict_size = 30000 +source_dict_dim = target_dict_dim = dict_size +src_dict, trg_dict = paddle.dataset.wmt14.get_dict(dict_size) +hidden_dim = 32 +word_dim = 16 +IS_SPARSE = True +batch_size = 10 +max_length = 50 +topk_size = 50 +trg_dic_size = 10000 + +decoder_size = hidden_dim + + +def encoder_decoder(): + # encoder + src_word_id = layers.data( + name="src_word_id", shape=[1], dtype='int64', lod_level=1) + src_embedding = layers.embedding( + input=src_word_id, + size=[dict_size, word_dim], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr=fluid.ParamAttr(name='vemb')) + + fc1 = fluid.layers.fc(input=src_embedding, size=hidden_dim * 4, act='tanh') + lstm_hidden0, lstm_0 = layers.dynamic_lstm(input=fc1, size=hidden_dim * 4) + encoder_out = layers.sequence_last_step(input=lstm_hidden0) + + # decoder + trg_language_word = layers.data( + name="target_language_word", shape=[1], dtype='int64', lod_level=1) + trg_embedding = layers.embedding( + input=trg_language_word, + size=[dict_size, word_dim], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr=fluid.ParamAttr(name='vemb')) + + rnn = fluid.layers.DynamicRNN() + with rnn.block(): + current_word = rnn.step_input(trg_embedding) + mem = rnn.memory(init=encoder_out) + fc1 = fluid.layers.fc(input=[current_word, mem], + size=decoder_size, + act='tanh') + out = fluid.layers.fc(input=fc1, size=target_dict_dim, act='softmax') + rnn.update_memory(mem, fc1) + rnn.output(out) + + return rnn() + + +def to_lodtensor(data, place): + seq_lens = [len(seq) for seq in data] + cur_len = 0 + lod = [cur_len] + for l in seq_lens: + cur_len += l + lod.append(cur_len) + flattened_data = np.concatenate(data, axis=0).astype("int64") + flattened_data = flattened_data.reshape([len(flattened_data), 1]) + res = core.LoDTensor() + res.set(flattened_data, place) + res.set_lod([lod]) + return res + + +def main(): + rnn_out = encoder_decoder() + label = layers.data( + name="target_language_next_word", shape=[1], dtype='int64', lod_level=1) + cost = layers.cross_entropy(input=rnn_out, label=label) + avg_cost = fluid.layers.mean(x=cost) + + optimizer = fluid.optimizer.Adagrad(learning_rate=1e-4) + optimize_ops, params_grads = optimizer.minimize(avg_cost) + + train_data = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.wmt14.train(dict_size), buf_size=1000), + batch_size=batch_size) + + place = core.CPUPlace() + exe = Executor(place) + + t = fluid.DistributeTranspiler() + # all parameter server endpoints list for spliting parameters + pserver_endpoints = os.getenv("PSERVERS") + # server endpoint for current node + current_endpoint = os.getenv("SERVER_ENDPOINT") + # run as trainer or parameter server + training_role = os.getenv( + "TRAINING_ROLE", "TRAINER") # get the training role: trainer/pserver + t.transpile( + optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) + + if training_role == "PSERVER": + if not current_endpoint: + print("need env SERVER_ENDPOINT") + exit(1) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) + exe.run(pserver_prog) + elif training_role == "TRAINER": + trainer_prog = t.get_trainer_program() + exe.run(framework.default_startup_program()) + + batch_id = 0 + for pass_id in xrange(2): + for data in train_data(): + word_data = to_lodtensor(map(lambda x: x[0], data), place) + trg_word = to_lodtensor(map(lambda x: x[1], data), place) + trg_word_next = to_lodtensor(map(lambda x: x[2], data), place) + outs = exe.run(trainer_prog, + feed={ + 'src_word_id': word_data, + 'target_language_word': trg_word, + 'target_language_next_word': trg_word_next + }, + fetch_list=[avg_cost]) + avg_cost_val = np.array(outs[0]) + print('pass_id=' + str(pass_id) + ' batch=' + str(batch_id) + + " avg_cost=" + str(avg_cost_val)) + if batch_id > 3: + exit(0) + batch_id += 1 + else: + print("environment var TRAINER_ROLE should be TRAINER os PSERVER") + + +if __name__ == '__main__': + main() diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_conv_dist.py b/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_conv_dist.py index 27512c4f7812b6b55d5dc6d1a12c3b83df8b3e6f..f18ca05c78093ff4ca22bf8b30d59240ee55ee8b 100644 --- a/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_conv_dist.py +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_conv_dist.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import numpy as np import paddle.v2 as paddle @@ -52,26 +53,27 @@ train_reader = paddle.batch( place = fluid.CPUPlace() exe = fluid.Executor(place) -t = fluid.DistributeTranspiler() -# all parameter server endpoints list for spliting parameters -pserver_endpoints = os.getenv("PSERVERS") -# server endpoint for current node -current_endpoint = os.getenv("SERVER_ENDPOINT") -# run as trainer or parameter server +pserver_endpoints = os.getenv("PSERVERS") # all pserver endpoints +trainers = int(os.getenv("TRAINERS")) # total trainer count +current_endpoint = os.getenv("SERVER_ENDPOINT") # current pserver endpoint training_role = os.getenv("TRAINING_ROLE", "TRAINER") # get the training role: trainer/pserver -t.transpile(optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) +t = fluid.DistributeTranspiler() +t.transpile( + optimize_ops, params_grads, pservers=pserver_endpoints, trainers=trainers) if training_role == "PSERVER": if not current_endpoint: print("need env SERVER_ENDPOINT") exit(1) - pserver_prog = t.get_pserver_program(current_endpoint, optimize_ops) - exe.run(fluid.default_startup_program()) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) exe.run(pserver_prog) elif training_role == "TRAINER": trainer_prog = t.get_trainer_program() feeder = fluid.DataFeeder(feed_list=[images, label], place=place) + # TODO(typhoonzero): change trainer startup program to fetch parameters from pserver exe.run(fluid.default_startup_program()) for pass_id in range(PASS_NUM): diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_mlp_dist.py b/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_mlp_dist.py new file mode 100644 index 0000000000000000000000000000000000000000..7733248cb447ed953d8b05945d51370c4c293489 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_recognize_digits_mlp_dist.py @@ -0,0 +1,89 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from __future__ import print_function +import numpy as np +import paddle.v2 as paddle +import paddle.v2.fluid as fluid +import os + +BATCH_SIZE = 128 +PASS_NUM = 100 + +images = fluid.layers.data(name='x', shape=[784], dtype='float32') + +# TODO(aroraabhinav) Add regularization and error clipping after +# Issue 7432(https://github.com/PaddlePaddle/Paddle/issues/7432) is resolved. +hidden1 = fluid.layers.fc(input=images, size=128, act='relu') +hidden2 = fluid.layers.fc(input=hidden1, size=64, act='relu') +predict = fluid.layers.fc(input=hidden2, size=10, act='softmax') + +label = fluid.layers.data(name='y', shape=[1], dtype='int64') + +cost = fluid.layers.cross_entropy(input=predict, label=label) +avg_cost = fluid.layers.mean(x=cost) + +optimizer = fluid.optimizer.Momentum(learning_rate=0.001, momentum=0.9) +optimize_ops, params_grads = optimizer.minimize(avg_cost) + +accuracy = fluid.evaluator.Accuracy(input=predict, label=label) + +train_reader = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.mnist.train(), buf_size=8192), + batch_size=BATCH_SIZE) + +place = fluid.CPUPlace() +exe = fluid.Executor(place) + +t = fluid.DistributeTranspiler() +# all parameter server endpoints list for spliting parameters +pserver_endpoints = os.getenv("PSERVERS") +# server endpoint for current node +current_endpoint = os.getenv("SERVER_ENDPOINT") +# run as trainer or parameter server +training_role = os.getenv("TRAINING_ROLE", + "TRAINER") # get the training role: trainer/pserver +t.transpile(optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) + +if training_role == "PSERVER": + if not current_endpoint: + print("need env SERVER_ENDPOINT") + exit(1) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) + exe.run(pserver_prog) +elif training_role == "TRAINER": + trainer_prog = t.get_trainer_program() + feeder = fluid.DataFeeder(feed_list=[images, label], place=place) + exe.run(fluid.default_startup_program()) + + for pass_id in range(PASS_NUM): + accuracy.reset(exe) + batch_id = 0 + for data in train_reader(): + loss, acc = exe.run(trainer_prog, + feed=feeder.feed(data), + fetch_list=[avg_cost] + accuracy.metrics) + pass_acc = accuracy.eval(exe) + if batch_id % 100 == 0: + print("batch_id %d, loss: %f, acc: %f" % + (batch_id, loss, pass_acc)) + batch_id += 1 + + pass_acc = accuracy.eval(exe) + print("pass_id=" + str(pass_id) + " pass_acc=" + str(pass_acc)) +else: + print("environment var TRAINER_ROLE should be TRAINER os PSERVER") diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_recommender_system_dist.py b/python/paddle/v2/fluid/tests/book_distribute/notest_recommender_system_dist.py new file mode 100644 index 0000000000000000000000000000000000000000..2d8885e377b0a10d8b5bad4e8fcecb9cc6fc8b64 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_recommender_system_dist.py @@ -0,0 +1,216 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import numpy as np +import os +import paddle.v2 as paddle +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import paddle.v2.fluid.layers as layers +import paddle.v2.fluid.nets as nets +from paddle.v2.fluid.optimizer import SGDOptimizer + +IS_SPARSE = True +BATCH_SIZE = 256 +PASS_NUM = 100 + + +def get_usr_combined_features(): + USR_DICT_SIZE = paddle.dataset.movielens.max_user_id() + 1 + uid = layers.data(name='user_id', shape=[1], dtype='int64') + usr_emb = layers.embedding( + input=uid, + dtype='float32', + size=[USR_DICT_SIZE, 32], + param_attr='user_table', + is_sparse=IS_SPARSE) + usr_fc = layers.fc(input=usr_emb, size=32) + USR_GENDER_DICT_SIZE = 2 + + usr_gender_id = layers.data(name='gender_id', shape=[1], dtype='int64') + usr_gender_emb = layers.embedding( + input=usr_gender_id, + size=[USR_GENDER_DICT_SIZE, 16], + param_attr='gender_table', + is_sparse=IS_SPARSE) + usr_gender_fc = layers.fc(input=usr_gender_emb, size=16) + + USR_AGE_DICT_SIZE = len(paddle.dataset.movielens.age_table) + usr_age_id = layers.data(name='age_id', shape=[1], dtype="int64") + usr_age_emb = layers.embedding( + input=usr_age_id, + size=[USR_AGE_DICT_SIZE, 16], + is_sparse=IS_SPARSE, + param_attr='age_table') + usr_age_fc = layers.fc(input=usr_age_emb, size=16) + + USR_JOB_DICT_SIZE = paddle.dataset.movielens.max_job_id() + 1 + usr_job_id = layers.data(name='job_id', shape=[1], dtype="int64") + usr_job_emb = layers.embedding( + input=usr_job_id, + size=[USR_JOB_DICT_SIZE, 16], + param_attr='job_table', + is_sparse=IS_SPARSE) + usr_job_fc = layers.fc(input=usr_job_emb, size=16) + + concat_embed = layers.concat( + input=[usr_fc, usr_gender_fc, usr_age_fc, usr_job_fc], axis=1) + + usr_combined_features = layers.fc(input=concat_embed, size=200, act="tanh") + return usr_combined_features + + +def get_mov_combined_features(): + MOV_DICT_SIZE = paddle.dataset.movielens.max_movie_id() + 1 + mov_id = layers.data(name='movie_id', shape=[1], dtype='int64') + mov_emb = layers.embedding( + input=mov_id, + dtype='float32', + size=[MOV_DICT_SIZE, 32], + param_attr='movie_table', + is_sparse=IS_SPARSE) + mov_fc = layers.fc(input=mov_emb, size=32) + + CATEGORY_DICT_SIZE = len(paddle.dataset.movielens.movie_categories()) + category_id = layers.data(name='category_id', shape=[1], dtype='int64') + mov_categories_emb = layers.embedding( + input=category_id, size=[CATEGORY_DICT_SIZE, 32], is_sparse=IS_SPARSE) + mov_categories_hidden = layers.sequence_pool( + input=mov_categories_emb, pool_type="sum") + + MOV_TITLE_DICT_SIZE = len(paddle.dataset.movielens.get_movie_title_dict()) + mov_title_id = layers.data(name='movie_title', shape=[1], dtype='int64') + mov_title_emb = layers.embedding( + input=mov_title_id, size=[MOV_TITLE_DICT_SIZE, 32], is_sparse=IS_SPARSE) + mov_title_conv = nets.sequence_conv_pool( + input=mov_title_emb, + num_filters=32, + filter_size=3, + act="tanh", + pool_type="sum") + + concat_embed = layers.concat( + input=[mov_fc, mov_categories_hidden, mov_title_conv], axis=1) + + mov_combined_features = layers.fc(input=concat_embed, size=200, act="tanh") + return mov_combined_features + + +def model(): + usr_combined_features = get_usr_combined_features() + mov_combined_features = get_mov_combined_features() + + # need cos sim + inference = layers.cos_sim(X=usr_combined_features, Y=mov_combined_features) + scale_infer = layers.scale(x=inference, scale=5.0) + + label = layers.data(name='score', shape=[1], dtype='float32') + square_cost = layers.square_error_cost(input=scale_infer, label=label) + avg_cost = layers.mean(x=square_cost) + + return avg_cost + + +def func_feed(feeding, data, place): + feed_tensors = {} + for (key, idx) in feeding.iteritems(): + tensor = core.LoDTensor() + if key != "category_id" and key != "movie_title": + if key == "score": + numpy_data = np.array(map(lambda x: x[idx], data)).astype( + "float32") + else: + numpy_data = np.array(map(lambda x: x[idx], data)).astype( + "int64") + else: + numpy_data = map(lambda x: np.array(x[idx]).astype("int64"), data) + lod_info = [len(item) for item in numpy_data] + offset = 0 + lod = [offset] + for item in lod_info: + offset += item + lod.append(offset) + numpy_data = np.concatenate(numpy_data, axis=0) + tensor.set_lod([lod]) + + numpy_data = numpy_data.reshape([numpy_data.shape[0], 1]) + tensor.set(numpy_data, place) + feed_tensors[key] = tensor + return feed_tensors + + +def main(): + cost = model() + optimizer = SGDOptimizer(learning_rate=0.2) + optimize_ops, params_grads = optimizer.minimize(cost) + + train_reader = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.movielens.train(), buf_size=8192), + batch_size=BATCH_SIZE) + + place = fluid.CPUPlace() + exe = fluid.Executor(place) + + t = fluid.DistributeTranspiler() + + # all parameter server endpoints list for spliting parameters + pserver_endpoints = os.getenv("PSERVERS") + # server endpoint for current node + current_endpoint = os.getenv("SERVER_ENDPOINT") + # run as trainer or parameter server + training_role = os.getenv("TRAINING_ROLE", "TRAINER") + t.transpile( + optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) + + if training_role == "PSERVER": + if not current_endpoint: + print("need env SERVER_ENDPOINT") + exit(1) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) + exe.run(pserver_prog) + elif training_role == "TRAINER": + exe.run(fluid.default_startup_program()) + trainer_prog = t.get_trainer_program() + + feeding = { + 'user_id': 0, + 'gender_id': 1, + 'age_id': 2, + 'job_id': 3, + 'movie_id': 4, + 'category_id': 5, + 'movie_title': 6, + 'score': 7 + } + + for pass_id in range(PASS_NUM): + for data in train_reader(): + outs = exe.run(trainer_prog, + feed=func_feed(feeding, data, place), + fetch_list=[cost]) + out = np.array(outs[0]) + print("cost=" + str(out[0])) + if out[0] < 6.0: + print("Training complete. Average cost is less than 6.0.") + # if avg cost less than 6.0, we think our code is good. + exit(0) + else: + print("environment var TRAINER_ROLE should be TRAINER os PSERVER") + + +if __name__ == '__main__': + main() diff --git a/python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_conv_dist.py b/python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_conv_dist.py index 74f20f3f4cc8c81a38c1ad8ab33df6a07fbcad44..49f26d6b69a836cdc44244eb8938884637acf720 100644 --- a/python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_conv_dist.py +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_conv_dist.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import os import numpy as np @@ -92,15 +93,16 @@ def main(): t.transpile( optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) - exe.run(fluid.default_startup_program()) - if training_role == "PSERVER": if not current_endpoint: print("need env SERVER_ENDPOINT") exit(1) - pserver_prog = t.get_pserver_program(current_endpoint, optimize_ops) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) exe.run(pserver_prog) elif training_role == "TRAINER": + exe.run(fluid.default_startup_program()) trainer_prog = t.get_trainer_program() feeder = fluid.DataFeeder(feed_list=[data, label], place=place) diff --git a/python/paddle/v2/fluid/tests/book/test_understand_sentiment_dynamic_lstm.py b/python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_dynamic_lstm.py similarity index 50% rename from python/paddle/v2/fluid/tests/book/test_understand_sentiment_dynamic_lstm.py rename to python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_dynamic_lstm.py index 5a139c1dcd41305aa5aece96f6b2aabde0235b95..bff376a0e2ee0fbb0d869e0dddf4460ed5dc4ac6 100644 --- a/python/paddle/v2/fluid/tests/book/test_understand_sentiment_dynamic_lstm.py +++ b/python/paddle/v2/fluid/tests/book_distribute/notest_understand_sentiment_dynamic_lstm.py @@ -1,17 +1,19 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np +import os import paddle.v2 as paddle import paddle.v2.fluid as fluid @@ -49,9 +51,9 @@ def stacked_lstm_net(data, cost = fluid.layers.cross_entropy(input=prediction, label=label) avg_cost = fluid.layers.mean(x=cost) adam_optimizer = fluid.optimizer.Adam(learning_rate=0.002) - adam_optimizer.minimize(avg_cost) + optimize_ops, params_grads = adam_optimizer.minimize(avg_cost) accuracy = fluid.evaluator.Accuracy(input=prediction, label=label) - return avg_cost, accuracy, accuracy.metrics[0] + return avg_cost, accuracy, accuracy.metrics[0], optimize_ops, params_grads def to_lodtensor(data, place): @@ -74,14 +76,14 @@ def main(): PASS_NUM = 5 word_dict = paddle.dataset.imdb.word_dict() - print "load word dict successfully" + print "loaded word dict successfully" dict_dim = len(word_dict) class_dim = 2 data = fluid.layers.data( name="words", shape=[1], dtype="int64", lod_level=1) label = fluid.layers.data(name="label", shape=[1], dtype="int64") - cost, accuracy, acc_out = stacked_lstm_net( + cost, accuracy, acc_out, optimize_ops, params_grads = stacked_lstm_net( data, label, input_dim=dict_dim, class_dim=class_dim) train_data = paddle.batch( @@ -92,20 +94,41 @@ def main(): exe = fluid.Executor(place) feeder = fluid.DataFeeder(feed_list=[data, label], place=place) - exe.run(fluid.default_startup_program()) - - for pass_id in xrange(PASS_NUM): - accuracy.reset(exe) - for data in train_data(): - cost_val, acc_val = exe.run(fluid.default_main_program(), - feed=feeder.feed(data), - fetch_list=[cost, acc_out]) - pass_acc = accuracy.eval(exe) - print("cost=" + str(cost_val) + " acc=" + str(acc_val) + - " pass_acc=" + str(pass_acc)) - if cost_val < 1.0 and acc_val > 0.8: - exit(0) - exit(1) + t = fluid.DistributeTranspiler() + # all parameter server endpoints list for spliting parameters + pserver_endpoints = os.getenv("PSERVERS") + # server endpoint for current node + current_endpoint = os.getenv("SERVER_ENDPOINT") + # run as trainer or parameter server + training_role = os.getenv( + "TRAINING_ROLE", "TRAINER") # get the training role: trainer/pserver + t.transpile( + optimize_ops, params_grads, pservers=pserver_endpoints, trainers=2) + + if training_role == "PSERVER": + if not current_endpoint: + print("need env SERVER_ENDPOINT") + exit(1) + pserver_prog = t.get_pserver_program(current_endpoint) + pserver_startup = t.get_startup_program(current_endpoint, pserver_prog) + exe.run(pserver_startup) + exe.run(pserver_prog) + elif training_role == "TRAINER": + exe.run(fluid.default_startup_program()) + trainer_prog = t.get_trainer_program() + for pass_id in xrange(PASS_NUM): + accuracy.reset(exe) + for data in train_data(): + cost_val, acc_val = exe.run(trainer_prog, + feed=feeder.feed(data), + fetch_list=[cost, acc_out]) + pass_acc = accuracy.eval(exe) + print("cost=" + str(cost_val) + " acc=" + str(acc_val) + + " pass_acc=" + str(pass_acc)) + if cost_val < 1.0 and acc_val > 0.8: + exit(0) + else: + print("environment var TRAINER_ROLE should be TRAINER os PSERVER") if __name__ == '__main__': diff --git a/python/paddle/v2/fluid/tests/book_distribute/test_split_var.py b/python/paddle/v2/fluid/tests/book_distribute/test_split_var.py index f979f642d8f8cf5869cd74d6f89d1d01f5860504..4a50049bf2644f237de9feadc284ead05fa2f36c 100644 --- a/python/paddle/v2/fluid/tests/book_distribute/test_split_var.py +++ b/python/paddle/v2/fluid/tests/book_distribute/test_split_var.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import math import unittest from paddle.v2.fluid.distribute_transpiler import split_dense_variable diff --git a/python/paddle/v2/fluid/tests/book_memory_optimization/CMakeLists.txt b/python/paddle/v2/fluid/tests/book_memory_optimization/CMakeLists.txt new file mode 100644 index 0000000000000000000000000000000000000000..213af5d27f711214feda3d200ced57bf71fbf6c2 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_memory_optimization/CMakeLists.txt @@ -0,0 +1,11 @@ +file(GLOB TEST_OPS RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "test_*.py") +string(REPLACE ".py" "" TEST_OPS "${TEST_OPS}") + +list(REMOVE_ITEM TEST_OPS test_memopt_image_classification_train) +py_test(test_memopt_image_classification_train_resnet SRCS test_memopt_image_classification_train.py ARGS resnet) +py_test(test_memopt_image_classification_train_vgg SRCS test_memopt_image_classification_train.py ARGS vgg) + +# default test +foreach(src ${TEST_OPS}) + py_test(${src} SRCS ${src}.py) +endforeach() diff --git a/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py b/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py new file mode 100644 index 0000000000000000000000000000000000000000..7ad5e2c594f24999e298533b6c05ba688a935f0b --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_fit_a_line.py @@ -0,0 +1,66 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import numpy as np +import paddle.v2 as paddle +import paddle.v2.fluid as fluid + +# need to fix random seed and training data to compare the loss +# value accurately calculated by the default and the memory optimization +# version. +fluid.default_startup_program().random_seed = 111 + +x = fluid.layers.data(name='x', shape=[13], dtype='float32') + +y_predict = fluid.layers.fc(input=x, size=1, act=None) + +y = fluid.layers.data(name='y', shape=[1], dtype='float32') + +cost = fluid.layers.square_error_cost(input=y_predict, label=y) +avg_cost = fluid.layers.mean(x=cost) + +sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.1) +sgd_optimizer.minimize(avg_cost) + +fluid.memory_optimize(fluid.default_main_program()) + +BATCH_SIZE = 200 + +# fix the order of training data +train_reader = paddle.batch( + paddle.dataset.uci_housing.train(), batch_size=BATCH_SIZE) + +# train_reader = paddle.batch( +# paddle.reader.shuffle( +# paddle.dataset.uci_housing.train(), buf_size=500), +# batch_size=BATCH_SIZE) + +place = fluid.CPUPlace() +feeder = fluid.DataFeeder(place=place, feed_list=[x, y]) +exe = fluid.Executor(place) + +exe.run(fluid.default_startup_program()) + +PASS_NUM = 100 +for pass_id in range(PASS_NUM): + fluid.io.save_persistables(exe, "./fit_a_line.model/") + fluid.io.load_persistables(exe, "./fit_a_line.model/") + for data in train_reader(): + avg_loss_value, = exe.run(fluid.default_main_program(), + feed=feeder.feed(data), + fetch_list=[avg_cost]) + + if avg_loss_value[0] < 10.0: + exit(0) # if avg cost less than 10.0, we think our code is good. +exit(1) diff --git a/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_image_classification_train.py b/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_image_classification_train.py new file mode 100644 index 0000000000000000000000000000000000000000..26673afd83c48328c3f354e82bfa3725aa4805b5 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_image_classification_train.py @@ -0,0 +1,158 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from __future__ import print_function + +import sys + +import paddle.v2 as paddle +import paddle.v2.fluid as fluid + +# need to fix random seed and training data to compare the loss +# value accurately calculated by the default and the memory optimization +# version. +fluid.default_startup_program().random_seed = 111 + + +def resnet_cifar10(input, depth=32): + def conv_bn_layer(input, ch_out, filter_size, stride, padding, act='relu'): + tmp = fluid.layers.conv2d( + input=input, + filter_size=filter_size, + num_filters=ch_out, + stride=stride, + padding=padding, + act=None, + bias_attr=False) + return fluid.layers.batch_norm(input=tmp, act=act) + + def shortcut(input, ch_in, ch_out, stride): + if ch_in != ch_out: + return conv_bn_layer(input, ch_out, 1, stride, 0, None) + else: + return input + + def basicblock(input, ch_in, ch_out, stride): + tmp = conv_bn_layer(input, ch_out, 3, stride, 1) + tmp = conv_bn_layer(tmp, ch_out, 3, 1, 1, act=None) + short = shortcut(input, ch_in, ch_out, stride) + return fluid.layers.elementwise_add(x=tmp, y=short, act='relu') + + def layer_warp(block_func, input, ch_in, ch_out, count, stride): + tmp = block_func(input, ch_in, ch_out, stride) + for i in range(1, count): + tmp = block_func(tmp, ch_out, ch_out, 1) + return tmp + + assert (depth - 2) % 6 == 0 + n = (depth - 2) / 6 + conv1 = conv_bn_layer( + input=input, ch_out=16, filter_size=3, stride=1, padding=1) + res1 = layer_warp(basicblock, conv1, 16, 16, n, 1) + res2 = layer_warp(basicblock, res1, 16, 32, n, 2) + res3 = layer_warp(basicblock, res2, 32, 64, n, 2) + pool = fluid.layers.pool2d( + input=res3, pool_size=8, pool_type='avg', pool_stride=1) + return pool + + +def vgg16_bn_drop(input): + def conv_block(input, num_filter, groups, dropouts): + return fluid.nets.img_conv_group( + input=input, + pool_size=2, + pool_stride=2, + conv_num_filter=[num_filter] * groups, + conv_filter_size=3, + conv_act='relu', + conv_with_batchnorm=True, + conv_batchnorm_drop_rate=dropouts, + pool_type='max') + + conv1 = conv_block(input, 64, 2, [0.3, 0]) + conv2 = conv_block(conv1, 128, 2, [0.4, 0]) + conv3 = conv_block(conv2, 256, 3, [0.4, 0.4, 0]) + conv4 = conv_block(conv3, 512, 3, [0.4, 0.4, 0]) + conv5 = conv_block(conv4, 512, 3, [0.4, 0.4, 0]) + + drop = fluid.layers.dropout(x=conv5, dropout_prob=0.5) + fc1 = fluid.layers.fc(input=drop, size=512, act=None) + bn = fluid.layers.batch_norm(input=fc1, act='relu') + drop2 = fluid.layers.dropout(x=bn, dropout_prob=0.5) + fc2 = fluid.layers.fc(input=drop2, size=512, act=None) + return fc2 + + +classdim = 10 +data_shape = [3, 32, 32] + +images = fluid.layers.data(name='pixel', shape=data_shape, dtype='float32') +label = fluid.layers.data(name='label', shape=[1], dtype='int64') + +net_type = "vgg" +if len(sys.argv) >= 2: + net_type = sys.argv[1] + +if net_type == "vgg": + print("train vgg net") + net = vgg16_bn_drop(images) +elif net_type == "resnet": + print("train resnet") + net = resnet_cifar10(images, 32) +else: + raise ValueError("%s network is not supported" % net_type) + +predict = fluid.layers.fc(input=net, size=classdim, act='softmax') +cost = fluid.layers.cross_entropy(input=predict, label=label) +avg_cost = fluid.layers.mean(x=cost) + +optimizer = fluid.optimizer.Adam(learning_rate=0.001) +opts = optimizer.minimize(avg_cost) + +accuracy = fluid.evaluator.Accuracy(input=predict, label=label) + +fluid.memory_optimize(fluid.default_main_program()) + +BATCH_SIZE = 128 +PASS_NUM = 1 + +# fix the order of training data +train_reader = paddle.batch( + paddle.dataset.cifar.train10(), batch_size=BATCH_SIZE) + +# train_reader = paddle.batch( +# paddle.reader.shuffle( +# paddle.dataset.cifar.train10(), buf_size=128 * 10), +# batch_size=BATCH_SIZE) + +place = fluid.CPUPlace() +exe = fluid.Executor(place) +feeder = fluid.DataFeeder(place=place, feed_list=[images, label]) +exe.run(fluid.default_startup_program()) + +i = 0 +for pass_id in range(PASS_NUM): + accuracy.reset(exe) + for data in train_reader(): + loss, acc = exe.run(fluid.default_main_program(), + feed=feeder.feed(data), + fetch_list=[avg_cost] + accuracy.metrics) + pass_acc = accuracy.eval(exe) + print("loss:" + str(loss) + " acc:" + str(acc) + " pass_acc:" + str( + pass_acc)) + # this model is slow, so if we can train two mini batch, we think it works properly. + if i > 2: + exit(0) + i += 1 +exit(1) diff --git a/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_machine_translation.py b/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_machine_translation.py new file mode 100644 index 0000000000000000000000000000000000000000..ffd53e7a78142162317a677de49c1821635a65b5 --- /dev/null +++ b/python/paddle/v2/fluid/tests/book_memory_optimization/test_memopt_machine_translation.py @@ -0,0 +1,144 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import numpy as np +import paddle.v2 as paddle +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import paddle.v2.fluid.framework as framework +import paddle.v2.fluid.layers as layers +from paddle.v2.fluid.executor import Executor + +dict_size = 30000 +source_dict_dim = target_dict_dim = dict_size +src_dict, trg_dict = paddle.dataset.wmt14.get_dict(dict_size) +hidden_dim = 32 +word_dim = 16 +IS_SPARSE = True +batch_size = 10 +max_length = 50 +topk_size = 50 +trg_dic_size = 10000 + +decoder_size = hidden_dim + +# need to fix random seed and training data to compare the loss +# value accurately calculated by the default and the memory optimization +# version. +fluid.default_startup_program().random_seed = 111 + + +def encoder_decoder(): + # encoder + src_word_id = layers.data( + name="src_word_id", shape=[1], dtype='int64', lod_level=1) + src_embedding = layers.embedding( + input=src_word_id, + size=[dict_size, word_dim], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr=fluid.ParamAttr(name='vemb')) + + fc1 = fluid.layers.fc(input=src_embedding, size=hidden_dim * 4, act='tanh') + lstm_hidden0, lstm_0 = layers.dynamic_lstm(input=fc1, size=hidden_dim * 4) + encoder_out = layers.sequence_last_step(input=lstm_hidden0) + + # decoder + trg_language_word = layers.data( + name="target_language_word", shape=[1], dtype='int64', lod_level=1) + trg_embedding = layers.embedding( + input=trg_language_word, + size=[dict_size, word_dim], + dtype='float32', + is_sparse=IS_SPARSE, + param_attr=fluid.ParamAttr(name='vemb')) + + rnn = fluid.layers.DynamicRNN() + with rnn.block(): + current_word = rnn.step_input(trg_embedding) + mem = rnn.memory(init=encoder_out) + fc1 = fluid.layers.fc(input=[current_word, mem], + size=decoder_size, + act='tanh') + out = fluid.layers.fc(input=fc1, size=target_dict_dim, act='softmax') + rnn.update_memory(mem, fc1) + rnn.output(out) + + return rnn() + + +def to_lodtensor(data, place): + seq_lens = [len(seq) for seq in data] + cur_len = 0 + lod = [cur_len] + for l in seq_lens: + cur_len += l + lod.append(cur_len) + flattened_data = np.concatenate(data, axis=0).astype("int64") + flattened_data = flattened_data.reshape([len(flattened_data), 1]) + res = core.LoDTensor() + res.set(flattened_data, place) + res.set_lod([lod]) + return res + + +def main(): + rnn_out = encoder_decoder() + label = layers.data( + name="target_language_next_word", shape=[1], dtype='int64', lod_level=1) + cost = layers.cross_entropy(input=rnn_out, label=label) + avg_cost = fluid.layers.mean(x=cost) + + optimizer = fluid.optimizer.Adagrad(learning_rate=1e-4) + optimizer.minimize(avg_cost) + + fluid.memory_optimize(fluid.default_main_program()) + + # fix the order of training data + train_data = paddle.batch( + paddle.dataset.wmt14.train(dict_size), batch_size=batch_size) + + # train_data = paddle.batch( + # paddle.reader.shuffle( + # paddle.dataset.wmt14.train(dict_size), buf_size=1000), + # batch_size=batch_size) + + place = core.CPUPlace() + exe = Executor(place) + + exe.run(framework.default_startup_program()) + + batch_id = 0 + for pass_id in xrange(10): + for data in train_data(): + word_data = to_lodtensor(map(lambda x: x[0], data), place) + trg_word = to_lodtensor(map(lambda x: x[1], data), place) + trg_word_next = to_lodtensor(map(lambda x: x[2], data), place) + outs = exe.run(fluid.default_main_program(), + feed={ + 'src_word_id': word_data, + 'target_language_word': trg_word, + 'target_language_next_word': trg_word_next + }, + fetch_list=[avg_cost]) + avg_cost_val = np.array(outs[0]) + print('pass_id=' + str(pass_id) + ' batch=' + str(batch_id) + + " avg_cost=" + str(avg_cost_val)) + if batch_id > 2: + exit(0) + batch_id += 1 + + +if __name__ == '__main__': + main() diff --git a/python/paddle/v2/fluid/tests/decorators.py b/python/paddle/v2/fluid/tests/decorators.py index 3b314a15e1b054ece50ad5d697c5fac3bbfedbdc..0a8a2ccc4dc2bdfbda1a502651559647ddd8f422 100644 --- a/python/paddle/v2/fluid/tests/decorators.py +++ b/python/paddle/v2/fluid/tests/decorators.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid as fluid __all__ = ['many_times', 'prog_scope'] diff --git a/python/paddle/v2/fluid/tests/demo/fc_gan.py b/python/paddle/v2/fluid/tests/demo/fc_gan.py index 5f9e8f950779be214a5fd18c4b9d3a0c3f74282b..0652c8134d58bcb6e5bba469ae16ff1ab4fdae4b 100644 --- a/python/paddle/v2/fluid/tests/demo/fc_gan.py +++ b/python/paddle/v2/fluid/tests/demo/fc_gan.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import errno import math import os diff --git a/python/paddle/v2/fluid/tests/op_test.py b/python/paddle/v2/fluid/tests/op_test.py index c3b2220e6e2ca6285e4b1193620e7c560b1f7bfa..3f6d7070c2987d0557c60db84a2c679cd2cfe36b 100644 --- a/python/paddle/v2/fluid/tests/op_test.py +++ b/python/paddle/v2/fluid/tests/op_test.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import random @@ -333,7 +334,7 @@ class OpTest(unittest.TestCase): def check_output(self, atol=1e-5): places = [core.CPUPlace()] - if core.is_compile_gpu() and core.op_support_gpu(self.op_type): + if core.is_compiled_with_cuda() and core.op_support_gpu(self.op_type): places.append(core.CUDAPlace(0)) for place in places: self.check_output_with_place(place, atol) @@ -366,7 +367,7 @@ class OpTest(unittest.TestCase): max_relative_error=0.005, user_defined_grads=None): places = [core.CPUPlace()] - if core.is_compile_gpu() and core.op_support_gpu(self.op_type): + if core.is_compiled_with_cuda() and core.op_support_gpu(self.op_type): places.append(core.CUDAPlace(0)) for place in places: self.check_grad_with_place(place, inputs_to_check, output_names, diff --git a/python/paddle/v2/fluid/tests/test_accuracy_op.py b/python/paddle/v2/fluid/tests/test_accuracy_op.py index a20abac8a0ce689e1c49f7f7e082137d2cb3fbb4..ac3f3bdff44a870cb68d317d5a57e7a25270e6c3 100644 --- a/python/paddle/v2/fluid/tests/test_accuracy_op.py +++ b/python/paddle/v2/fluid/tests/test_accuracy_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_activation_op.py b/python/paddle/v2/fluid/tests/test_activation_op.py index a6a6eb9d635ed3ece0a0f22d629955b06098e321..1de5d446b8eaf57d3718dde7540c929996ee3432 100644 --- a/python/paddle/v2/fluid/tests/test_activation_op.py +++ b/python/paddle/v2/fluid/tests/test_activation_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -185,8 +186,7 @@ class TestFloor(OpTest): self.op_type = "floor" x = np.random.uniform(-1, 1, [4, 4]).astype("float32") self.inputs = {'X': x} - # numpy floor need +1 - self.outputs = {'Out': np.floor(self.inputs['X']) + 1.0} + self.outputs = {'Out': np.floor(self.inputs['X'])} def test_check_output(self): self.check_output() diff --git a/python/paddle/v2/fluid/tests/test_adadelta_op.py b/python/paddle/v2/fluid/tests/test_adadelta_op.py index 8de6a1f9a9da4834d7b52baade12bb15ef128cad..949318d00776712ad08d335f7afdb9b7d9140c42 100644 --- a/python/paddle/v2/fluid/tests/test_adadelta_op.py +++ b/python/paddle/v2/fluid/tests/test_adadelta_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_adagrad_op.py b/python/paddle/v2/fluid/tests/test_adagrad_op.py index 30ed092d4894f5c28775748a71f474c46a1cb2d3..3556bcf8ba0d7f16b1d9bf50e46aebde83de2e25 100644 --- a/python/paddle/v2/fluid/tests/test_adagrad_op.py +++ b/python/paddle/v2/fluid/tests/test_adagrad_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import paddle.v2.fluid.core as core @@ -179,7 +180,7 @@ class TestSparseAdagradOp(unittest.TestCase): def test_sparse_adagrad(self): places = [core.CPUPlace()] - if core.is_compile_gpu(): + if core.is_compiled_with_cuda(): places.append(core.CUDAPlace(0)) for place in places: self.check_with_place(place) diff --git a/python/paddle/v2/fluid/tests/test_adam_op.py b/python/paddle/v2/fluid/tests/test_adam_op.py index 32d00cf702e7fdca2b9f975dcaf93d721fe6cce1..df1fa8983c1984a9bb9f204aded148c17d3d609d 100644 --- a/python/paddle/v2/fluid/tests/test_adam_op.py +++ b/python/paddle/v2/fluid/tests/test_adam_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -304,7 +305,7 @@ class TestSparseAdamOp(unittest.TestCase): def test_sparse_sgd(self): places = [core.CPUPlace()] - if core.is_compile_gpu(): + if core.is_compiled_with_cuda(): places.append(core.CUDAPlace(0)) for place in places: self.check_with_place(place) diff --git a/python/paddle/v2/fluid/tests/test_adamax_op.py b/python/paddle/v2/fluid/tests/test_adamax_op.py index 35b2bc47ed62cb21b6e58a172e5b7e4d34f52eb4..e285c454f035936c9dd28bc41b9174f780201ba0 100644 --- a/python/paddle/v2/fluid/tests/test_adamax_op.py +++ b/python/paddle/v2/fluid/tests/test_adamax_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_array_read_write_op.py b/python/paddle/v2/fluid/tests/test_array_read_write_op.py index 8775cd4f9fb93e439920e29df4b525485254754c..a32c24486e1f4340d913da3ea42e7c9ff4a48d90 100644 --- a/python/paddle/v2/fluid/tests/test_array_read_write_op.py +++ b/python/paddle/v2/fluid/tests/test_array_read_write_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core import paddle.v2.fluid.layers as layers diff --git a/python/paddle/v2/fluid/tests/test_assign_op.py b/python/paddle/v2/fluid/tests/test_assign_op.py index 4ac173c96bd2c02fd0704d8e2c22faa96b65714e..fbbfe0d02c4036ea3e971a9c0a9a2469ca62ad53 100644 --- a/python/paddle/v2/fluid/tests/test_assign_op.py +++ b/python/paddle/v2/fluid/tests/test_assign_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import op_test import numpy import unittest diff --git a/python/paddle/v2/fluid/tests/test_assign_value_op.py b/python/paddle/v2/fluid/tests/test_assign_value_op.py index f4e2ff9bdeb88cc1fa19055a8e1ff4e6156f0477..93970f863b165c44c554a40e2fa60388800ce300 100644 --- a/python/paddle/v2/fluid/tests/test_assign_value_op.py +++ b/python/paddle/v2/fluid/tests/test_assign_value_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid as fluid import paddle.v2.fluid.layers as layers import op_test diff --git a/python/paddle/v2/fluid/tests/test_auc_op.py b/python/paddle/v2/fluid/tests/test_auc_op.py index aa74d224d5dcdcce4c478df8f85fb45d05d87f32..5e4caedf5d612ea9d1a4bcf17beee69316f9266d 100644 --- a/python/paddle/v2/fluid/tests/test_auc_op.py +++ b/python/paddle/v2/fluid/tests/test_auc_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_batch_norm_op.py b/python/paddle/v2/fluid/tests/test_batch_norm_op.py index fe82b7d7f31aa58d38bddba72e64d471c0a2eec0..cf13166f255c782bdcec622d58d073a0943c8e1e 100644 --- a/python/paddle/v2/fluid/tests/test_batch_norm_op.py +++ b/python/paddle/v2/fluid/tests/test_batch_norm_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -351,7 +352,7 @@ class TestBatchNormOp(OpTest): print "op test backward passed: ", str(place), data_layout places = [core.CPUPlace()] - if core.is_compile_gpu() and core.op_support_gpu("batch_norm"): + if core.is_compiled_with_cuda() and core.op_support_gpu("batch_norm"): places.append(core.CUDAPlace(0)) for place in places: diff --git a/python/paddle/v2/fluid/tests/test_beam_search_decode_op.py b/python/paddle/v2/fluid/tests/test_beam_search_decode_op.py index 9ef6e08cc189035c55d52ecaf209d5d607de0ed0..36747849859fd54b34b5f5c25e9f5b4c779774fb 100644 --- a/python/paddle/v2/fluid/tests/test_beam_search_decode_op.py +++ b/python/paddle/v2/fluid/tests/test_beam_search_decode_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_beam_search_op.py b/python/paddle/v2/fluid/tests/test_beam_search_op.py index f31c737ba6cb7d17405e96506834627c9c5761b4..4da463df260efe48498207e9758f91d0bf95e7fe 100644 --- a/python/paddle/v2/fluid/tests/test_beam_search_op.py +++ b/python/paddle/v2/fluid/tests/test_beam_search_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import logging from paddle.v2.fluid.op import Operator, DynamicRecurrentOp import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_bilinear_tensor_product_op.py b/python/paddle/v2/fluid/tests/test_bilinear_tensor_product_op.py index aed1bf4d3ae867dafdefe67d36bb350453e3ede0..4b03f512c2f0d073fad3ba04b5e72adec13af6ed 100644 --- a/python/paddle/v2/fluid/tests/test_bilinear_tensor_product_op.py +++ b/python/paddle/v2/fluid/tests/test_bilinear_tensor_product_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_bipartite_match_op.py b/python/paddle/v2/fluid/tests/test_bipartite_match_op.py new file mode 100644 index 0000000000000000000000000000000000000000..74138298978c7c18936f53761b313887f07aea81 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_bipartite_match_op.py @@ -0,0 +1,100 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. +import unittest +import numpy as np +from op_test import OpTest + + +def bipartite_match(distance, match_indices, match_dist): + """Bipartite Matching algorithm. + Arg: + distance (numpy.array) : The distance of two entries with shape [M, N]. + match_indices (numpy.array): the matched indices from column to row + with shape [1, N], it must be initialized to -1. + match_dist (numpy.array): The matched distance from column to row + with shape [1, N], it must be initialized to 0. + """ + match_pair = [] + row, col = distance.shape + for i in range(row): + for j in range(col): + match_pair.append((i, j, distance[i][j])) + + match_sorted = sorted(match_pair, key=lambda tup: tup[2], reverse=True) + + row_indices = -1 * np.ones((row, ), dtype=np.int) + + idx = 0 + for i, j, dist in match_sorted: + if idx >= row: + break + if match_indices[j] == -1 and row_indices[i] == -1 and dist > 0: + match_indices[j] = i + row_indices[i] = j + match_dist[j] = dist + idx += 1 + + +def batch_bipartite_match(distance, lod): + """Bipartite Matching algorithm for batch input. + Arg: + distance (numpy.array) : The distance of two entries with shape [M, N]. + lod (list of int): The offsets of each input in this batch. + """ + n = len(lod) - 1 + m = distance.shape[1] + match_indices = -1 * np.ones((n, m), dtype=np.int) + match_dist = np.zeros((n, m), dtype=np.float32) + for i in range(len(lod) - 1): + bipartite_match(distance[lod[i]:lod[i + 1], :], match_indices[i, :], + match_dist[i, :]) + return match_indices, match_dist + + +class TestBipartiteMatchOpForWithLoD(OpTest): + def setUp(self): + self.op_type = 'bipartite_match' + lod = [[0, 5, 11, 23]] + dist = np.random.random((23, 217)).astype('float32') + match_indices, match_dist = batch_bipartite_match(dist, lod[0]) + + self.inputs = {'DistMat': (dist, lod)} + self.outputs = { + 'ColToRowMatchIndices': (match_indices), + 'ColToRowMatchDis': (match_dist), + } + + def test_check_output(self): + self.check_output() + + +class TestBipartiteMatchOpWithoutLoD(OpTest): + def setUp(self): + self.op_type = 'bipartite_match' + lod = [[0, 8]] + dist = np.random.random((8, 17)).astype('float32') + match_indices, match_dist = batch_bipartite_match(dist, lod[0]) + + self.inputs = {'DistMat': dist} + self.outputs = { + 'ColToRowMatchIndices': match_indices, + 'ColToRowMatchDis': match_dist, + } + + def test_check_output(self): + self.check_output() + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_calc_gradient.py b/python/paddle/v2/fluid/tests/test_calc_gradient.py index b99eeb09cdbb7c2174ea5f29ba254ce342517f9c..c773e81768f50c6bd3865f8dd527f4d955a95229 100644 --- a/python/paddle/v2/fluid/tests/test_calc_gradient.py +++ b/python/paddle/v2/fluid/tests/test_calc_gradient.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid as fluid diff --git a/python/paddle/v2/fluid/tests/test_cast_op.py b/python/paddle/v2/fluid/tests/test_cast_op.py index 3795b96dbf0f8f50ede2aeda7262ba61c095d6af..327b246ed80596c10877734cdea083264a5b9309 100644 --- a/python/paddle/v2/fluid/tests/test_cast_op.py +++ b/python/paddle/v2/fluid/tests/test_cast_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import op_test import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_chunk_eval_op.py b/python/paddle/v2/fluid/tests/test_chunk_eval_op.py index 59ef2bbb2feebc5930ff9513598e0ac55376635f..5c3efe9baa7da6161e22b274fb5116ffaed68f1a 100644 --- a/python/paddle/v2/fluid/tests/test_chunk_eval_op.py +++ b/python/paddle/v2/fluid/tests/test_chunk_eval_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_clip_by_norm_op.py b/python/paddle/v2/fluid/tests/test_clip_by_norm_op.py index 5147e75046294cb8fc7673f2e58a4fb8e50e0e53..b30f321c79f3d3a4061ee2fc2fb55fb5fab95f27 100644 --- a/python/paddle/v2/fluid/tests/test_clip_by_norm_op.py +++ b/python/paddle/v2/fluid/tests/test_clip_by_norm_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_clip_op.py b/python/paddle/v2/fluid/tests/test_clip_op.py index 3338dc61b38c7e46a048458ae81d7592e30bbffb..ef0b75e286797436d730278585e946d1d897edc2 100644 --- a/python/paddle/v2/fluid/tests/test_clip_op.py +++ b/python/paddle/v2/fluid/tests/test_clip_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_compare_op.py b/python/paddle/v2/fluid/tests/test_compare_op.py index fbf8921e40563cca28ade38465f68eb133b263e4..c9be80fc45cd3428937998357b9dd9cbde1547cc 100644 --- a/python/paddle/v2/fluid/tests/test_compare_op.py +++ b/python/paddle/v2/fluid/tests/test_compare_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import op_test import unittest import numpy @@ -37,8 +38,6 @@ def create_test_class(op_type, typename, callback): for _type_name in {'float32', 'float64', 'int32', 'int64'}: create_test_class('less_than', _type_name, lambda _a, _b: _a < _b) create_test_class('less_equal', _type_name, lambda _a, _b: _a <= _b) - create_test_class('greater_than', _type_name, lambda _a, _b: _a > _b) - create_test_class('greater_equal', _type_name, lambda _a, _b: _a >= _b) create_test_class('equal', _type_name, lambda _a, _b: _a == _b) if __name__ == '__main__': diff --git a/python/paddle/v2/fluid/tests/test_concat_op.py b/python/paddle/v2/fluid/tests/test_concat_op.py index 3e413e15404f64755fdf1e1db936ca34d61d2b03..ea0a95ebec24477797c1a17096c61132248587e5 100644 --- a/python/paddle/v2/fluid/tests/test_concat_op.py +++ b/python/paddle/v2/fluid/tests/test_concat_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_cond_op.py b/python/paddle/v2/fluid/tests/test_cond_op.py index 5312fa51a253a7151107d508120cb590aa822364..4b7ca0963e92ead0b4168ae732d70ff8497d38c0 100644 --- a/python/paddle/v2/fluid/tests/test_cond_op.py +++ b/python/paddle/v2/fluid/tests/test_cond_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import logging import paddle.v2.fluid.core as core import unittest diff --git a/python/paddle/v2/fluid/tests/test_conditional_block.py b/python/paddle/v2/fluid/tests/test_conditional_block.py index 965e7d39c807265d5884dc10cce228c5d7f5823f..5ee729cfee6f432091c07f360f2d0c97c3801b99 100644 --- a/python/paddle/v2/fluid/tests/test_conditional_block.py +++ b/python/paddle/v2/fluid/tests/test_conditional_block.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.layers as layers import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_const_value.py b/python/paddle/v2/fluid/tests/test_const_value.py index 190bfa779b4a2fac4d425ba9e45223d2aa04173b..d5b7cfded1a943ad84493afe367bf33c5304db42 100644 --- a/python/paddle/v2/fluid/tests/test_const_value.py +++ b/python/paddle/v2/fluid/tests/test_const_value.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.framework as framework diff --git a/python/paddle/v2/fluid/tests/test_conv2d_op.py b/python/paddle/v2/fluid/tests/test_conv2d_op.py index 8b03a3ae16592888119e1c9ea797e7bbe8acd324..24de74d730eedbccb4837598bd6d2eb92da59e0d 100644 --- a/python/paddle/v2/fluid/tests/test_conv2d_op.py +++ b/python/paddle/v2/fluid/tests/test_conv2d_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_conv2d_transpose_op.py b/python/paddle/v2/fluid/tests/test_conv2d_transpose_op.py index b7b86c58fb81d44b153c3b6724d5ebd524db55e3..0c76e222c90c7a61c08240e6b3d25fbb5b979252 100644 --- a/python/paddle/v2/fluid/tests/test_conv2d_transpose_op.py +++ b/python/paddle/v2/fluid/tests/test_conv2d_transpose_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_conv3d_op.py b/python/paddle/v2/fluid/tests/test_conv3d_op.py index 5b0397cc690caccdc78317071a10eb451afda979..8121e3286597e5842138eac1801f4466db24f799 100644 --- a/python/paddle/v2/fluid/tests/test_conv3d_op.py +++ b/python/paddle/v2/fluid/tests/test_conv3d_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_conv3d_transpose_op.py b/python/paddle/v2/fluid/tests/test_conv3d_transpose_op.py index b08969062a8ccfec063d7bc1ae3af38f5776fce7..4934c5a34e519578183a577adf66b48e6c1047d5 100644 --- a/python/paddle/v2/fluid/tests/test_conv3d_transpose_op.py +++ b/python/paddle/v2/fluid/tests/test_conv3d_transpose_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_conv_shift_op.py b/python/paddle/v2/fluid/tests/test_conv_shift_op.py index 14b2640e24cb8bd51111f14c187be73e423d20d2..7029d5a2eb433707ed7e7dd8acddbd0fef80e40c 100644 --- a/python/paddle/v2/fluid/tests/test_conv_shift_op.py +++ b/python/paddle/v2/fluid/tests/test_conv_shift_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_cos_sim_op.py b/python/paddle/v2/fluid/tests/test_cos_sim_op.py index f6e5e2cbe9ed54d197b6ee640b4ffd1b3c101f99..33db12ba9c7b119259fc918a70c029a99aea0c68 100644 --- a/python/paddle/v2/fluid/tests/test_cos_sim_op.py +++ b/python/paddle/v2/fluid/tests/test_cos_sim_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_create_op_doc_string.py b/python/paddle/v2/fluid/tests/test_create_op_doc_string.py index 6c922642210fd66604a66d1dd520fa5783c522c3..2b7951ecea791b43ffe9123fda77ec80d626e065 100644 --- a/python/paddle/v2/fluid/tests/test_create_op_doc_string.py +++ b/python/paddle/v2/fluid/tests/test_create_op_doc_string.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.layers as layers diff --git a/python/paddle/v2/fluid/tests/test_crf_decoding_op.py b/python/paddle/v2/fluid/tests/test_crf_decoding_op.py index 40e80a824a2fdd4ec2a6f2d96a2a6bf14ab74be0..f819387cdc4476187d108e7cffca7851baea5933 100644 --- a/python/paddle/v2/fluid/tests/test_crf_decoding_op.py +++ b/python/paddle/v2/fluid/tests/test_crf_decoding_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import random import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_crop_op.py b/python/paddle/v2/fluid/tests/test_crop_op.py index a0b2fc954dda10ad264fecaf8827732b8820c71b..36bf1761689c32928ea8f9e2996038ae94c92bdd 100644 --- a/python/paddle/v2/fluid/tests/test_crop_op.py +++ b/python/paddle/v2/fluid/tests/test_crop_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_cross_entropy_op.py b/python/paddle/v2/fluid/tests/test_cross_entropy_op.py index f05e6b235656c75cb75c0ae6a8758ceb48a54352..ae8e9be6de453e78f8e941641919a4c8eaae7e30 100644 --- a/python/paddle/v2/fluid/tests/test_cross_entropy_op.py +++ b/python/paddle/v2/fluid/tests/test_cross_entropy_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest, randomize_probability diff --git a/python/paddle/v2/fluid/tests/test_ctc_align.py b/python/paddle/v2/fluid/tests/test_ctc_align.py new file mode 100644 index 0000000000000000000000000000000000000000..773c69d1ad0794d2e4edfb1f6f8140cbcd64bee6 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_ctc_align.py @@ -0,0 +1,76 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import sys +import unittest +import numpy as np +from op_test import OpTest +from test_softmax_op import stable_softmax + + +def CTCAlign(input, lod, blank, merge_repeated): + lod0 = lod[0] + result = [] + for i in range(len(lod0) - 1): + prev_token = -1 + for j in range(lod0[i], lod0[i + 1]): + token = input[j][0] + if (token != blank) and not (merge_repeated and + token == prev_token): + result.append(token) + prev_token = token + result = np.array(result).reshape([len(result), 1]).astype("int32") + return result + + +class TestCTCAlignOp(OpTest): + def config(self): + self.op_type = "ctc_align" + self.input_lod = [[0, 11, 18]] + self.blank = 0 + self.merge_repeated = False + self.input = np.array( + [0, 1, 2, 2, 0, 4, 0, 4, 5, 0, 6, 6, 0, 0, 7, 7, 7, 0]).reshape( + [18, 1]).astype("int32") + + def setUp(self): + self.config() + output = CTCAlign(self.input, self.input_lod, self.blank, + self.merge_repeated) + + self.inputs = {"Input": (self.input, self.input_lod), } + self.outputs = {"Output": output} + self.attrs = { + "blank": self.blank, + "merge_repeated": self.merge_repeated + } + + def test_check_output(self): + self.check_output() + pass + + +class TestCTCAlignOpCase1(TestCTCAlignOp): + def config(self): + self.op_type = "ctc_align" + self.input_lod = [[0, 11, 19]] + self.blank = 0 + self.merge_repeated = True + self.input = np.array( + [0, 1, 2, 2, 0, 4, 0, 4, 5, 0, 6, 6, 0, 0, 7, 7, 7, 0, 0]).reshape( + [19, 1]).astype("int32") + + +if __name__ == "__main__": + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_data_feeder.py b/python/paddle/v2/fluid/tests/test_data_feeder.py index 5574766f8fad2272721f515311190c3dc41ace85..f967221015821e409592aeec93ffda8e4a4f3252 100644 --- a/python/paddle/v2/fluid/tests/test_data_feeder.py +++ b/python/paddle/v2/fluid/tests/test_data_feeder.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid as fluid diff --git a/python/paddle/v2/fluid/tests/test_decayed_adagrad_op.py b/python/paddle/v2/fluid/tests/test_decayed_adagrad_op.py index 5e745a284316e9be512178f909399ca4c7708901..78d4e3608e6f9fef247551a9ec8a70899bf8f86c 100644 --- a/python/paddle/v2/fluid/tests/test_decayed_adagrad_op.py +++ b/python/paddle/v2/fluid/tests/test_decayed_adagrad_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_default_scope_funcs.py b/python/paddle/v2/fluid/tests/test_default_scope_funcs.py index 7a62168be90514b933f0a8a58e754609ecaad646..5ff52f6d6b46f18a12fc508d8ce1a9fb20e6d6ca 100644 --- a/python/paddle/v2/fluid/tests/test_default_scope_funcs.py +++ b/python/paddle/v2/fluid/tests/test_default_scope_funcs.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.v2.fluid.default_scope_funcs import * import unittest diff --git a/python/paddle/v2/fluid/tests/test_detection_output_op.py b/python/paddle/v2/fluid/tests/test_detection_output_op.py index 147a43628c671658a5d4338da99423eda851b195..8a5e06b38f5ed5336ef02bac7876610758b44258 100644 --- a/python/paddle/v2/fluid/tests/test_detection_output_op.py +++ b/python/paddle/v2/fluid/tests/test_detection_output_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -67,4 +68,6 @@ class TestUnpoolOp(OpTest): if __name__ == '__main__': - unittest.main() + # FIXME: detection_output_op will be rewritten. This unittest should be + # enabled after rewriting. + exit(0) # temporary disable this unittest diff --git a/python/paddle/v2/fluid/tests/test_dropout_op.py b/python/paddle/v2/fluid/tests/test_dropout_op.py index f401050dcc39d4bf786e33ef3e8a4c33e0250044..b0c55df9f58834688846c5362113464996eb286a 100644 --- a/python/paddle/v2/fluid/tests/test_dropout_op.py +++ b/python/paddle/v2/fluid/tests/test_dropout_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -20,7 +21,7 @@ class TestDropoutOp(OpTest): def setUp(self): self.op_type = "dropout" self.inputs = {'X': np.random.random((32, 64)).astype("float32")} - self.attrs = {'dropout_prob': 0.0, 'is_test': False} + self.attrs = {'dropout_prob': 0.0, 'fix_seed': True, 'is_test': False} self.outputs = { 'Out': self.inputs['X'], 'Mask': np.ones((32, 64)).astype('float32') @@ -37,7 +38,7 @@ class TestDropoutOp2(TestDropoutOp): def setUp(self): self.op_type = "dropout" self.inputs = {'X': np.random.random((32, 64)).astype("float32")} - self.attrs = {'dropout_prob': 1.0, 'is_test': False} + self.attrs = {'dropout_prob': 1.0, 'fix_seed': True, 'is_test': False} self.outputs = { 'Out': np.zeros((32, 64)).astype('float32'), 'Mask': np.zeros((32, 64)).astype('float32') @@ -48,7 +49,7 @@ class TestDropoutOp3(TestDropoutOp): def setUp(self): self.op_type = "dropout" self.inputs = {'X': np.random.random((32, 64, 2)).astype("float32")} - self.attrs = {'dropout_prob': 0.0, 'is_test': False} + self.attrs = {'dropout_prob': 0.0, 'fix_seed': True, 'is_test': False} self.outputs = { 'Out': self.inputs['X'], 'Mask': np.ones((32, 64, 2)).astype('float32') @@ -59,7 +60,7 @@ class TestDropoutOp4(OpTest): def setUp(self): self.op_type = "dropout" self.inputs = {'X': np.random.random((32, 64)).astype("float32")} - self.attrs = {'dropout_prob': 0.35, 'is_test': True} + self.attrs = {'dropout_prob': 0.35, 'fix_seed': True, 'is_test': True} self.outputs = { 'Out': self.inputs['X'] * (1.0 - self.attrs['dropout_prob']) } diff --git a/python/paddle/v2/fluid/tests/test_dyn_rnn.py b/python/paddle/v2/fluid/tests/test_dyn_rnn.py index a946fea58d67887cac40bd35b14e23752d6a1619..2ac926c63c906767cb08c561b043b8a6cc6b36bd 100644 --- a/python/paddle/v2/fluid/tests/test_dyn_rnn.py +++ b/python/paddle/v2/fluid/tests/test_dyn_rnn.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid as fluid import paddle.v2 as paddle import unittest diff --git a/python/paddle/v2/fluid/tests/test_dynrnn_gradient_check.py b/python/paddle/v2/fluid/tests/test_dynrnn_gradient_check.py index 95cc80739d6940612d303a0033c74aa91a7465cc..dd608432df411a670210148af81ff4a6d3151a85 100644 --- a/python/paddle/v2/fluid/tests/test_dynrnn_gradient_check.py +++ b/python/paddle/v2/fluid/tests/test_dynrnn_gradient_check.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy import random import collections diff --git a/python/paddle/v2/fluid/tests/test_dynrnn_static_input.py b/python/paddle/v2/fluid/tests/test_dynrnn_static_input.py index d6878f0b6d07b74612dd47794e254e4d7d98a124..d14923b6b30d80e89f27221030f60edf947ae63d 100644 --- a/python/paddle/v2/fluid/tests/test_dynrnn_static_input.py +++ b/python/paddle/v2/fluid/tests/test_dynrnn_static_input.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2 as paddle import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_edit_distance_op.py b/python/paddle/v2/fluid/tests/test_edit_distance_op.py index 38e87728b387bb70a8921a2fe73a4e69701aabe9..bebdc5cba36fc96d31162d0d7d43e52064ca8e2d 100644 --- a/python/paddle/v2/fluid/tests/test_edit_distance_op.py +++ b/python/paddle/v2/fluid/tests/test_edit_distance_op.py @@ -1,3 +1,17 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -38,8 +52,8 @@ class TestEditDistanceOp(OpTest): def setUp(self): self.op_type = "edit_distance" normalized = False - x1 = np.array([[0, 12, 3, 5, 8, 2]]).astype("int32") - x2 = np.array([[0, 12, 4, 7, 8]]).astype("int32") + x1 = np.array([[0, 12, 3, 5, 8, 2]]).astype("int64") + x2 = np.array([[0, 12, 4, 7, 8]]).astype("int64") x1 = np.transpose(x1) x2 = np.transpose(x2) x1_lod = [0, 1, 5] @@ -47,6 +61,7 @@ class TestEditDistanceOp(OpTest): num_strs = len(x1_lod) - 1 distance = np.zeros((num_strs, 1)).astype("float32") + sequence_num = np.array(2).astype("int64") for i in range(0, num_strs): distance[i] = Levenshtein( hyp=x1[x1_lod[i]:x1_lod[i + 1]], @@ -56,7 +71,7 @@ class TestEditDistanceOp(OpTest): distance[i] = distance[i] / len_ref self.attrs = {'normalized': normalized} self.inputs = {'Hyps': (x1, [x1_lod]), 'Refs': (x2, [x2_lod])} - self.outputs = {'Out': distance} + self.outputs = {'Out': distance, 'SequenceNum': sequence_num} def test_check_output(self): self.check_output() @@ -66,8 +81,8 @@ class TestEditDistanceOpNormalized(OpTest): def setUp(self): self.op_type = "edit_distance" normalized = True - x1 = np.array([[0, 10, 3, 6, 5, 8, 2]]).astype("int32") - x2 = np.array([[0, 10, 4, 6, 7, 8]]).astype("int32") + x1 = np.array([[0, 10, 3, 6, 5, 8, 2]]).astype("int64") + x2 = np.array([[0, 10, 4, 6, 7, 8]]).astype("int64") x1 = np.transpose(x1) x2 = np.transpose(x2) x1_lod = [0, 1, 3, 6] @@ -75,6 +90,7 @@ class TestEditDistanceOpNormalized(OpTest): num_strs = len(x1_lod) - 1 distance = np.zeros((num_strs, 1)).astype("float32") + sequence_num = np.array(3).astype("int64") for i in range(0, num_strs): distance[i] = Levenshtein( hyp=x1[x1_lod[i]:x1_lod[i + 1]], @@ -84,7 +100,7 @@ class TestEditDistanceOpNormalized(OpTest): distance[i] = distance[i] / len_ref self.attrs = {'normalized': normalized} self.inputs = {'Hyps': (x1, [x1_lod]), 'Refs': (x2, [x2_lod])} - self.outputs = {'Out': distance} + self.outputs = {'Out': distance, 'SequenceNum': sequence_num} def test_check_output(self): self.check_output() diff --git a/python/paddle/v2/fluid/tests/test_elementwise_add_op.py b/python/paddle/v2/fluid/tests/test_elementwise_add_op.py index 1e88231877f869052d859a30fbbd8f7690b64095..3564772fb52882e9e58ea88caeb12c5e91137525 100644 --- a/python/paddle/v2/fluid/tests/test_elementwise_add_op.py +++ b/python/paddle/v2/fluid/tests/test_elementwise_add_op.py @@ -1,16 +1,16 @@ # Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import unittest import numpy as np from op_test import OpTest @@ -40,6 +40,16 @@ class TestElementwiseOp(OpTest): ['X'], 'Out', max_relative_error=0.005, no_grad_set=set('Y')) +class TestElementwiseAddOp_scalar(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_add" + self.inputs = { + 'X': np.random.rand(2, 3, 4).astype(np.float32), + 'Y': np.random.rand(1).astype(np.float32) + } + self.outputs = {'Out': self.inputs['X'] + self.inputs['Y']} + + class TestElementwiseAddOp_Vector(TestElementwiseOp): def setUp(self): self.op_type = "elementwise_add" diff --git a/python/paddle/v2/fluid/tests/test_elementwise_div_op.py b/python/paddle/v2/fluid/tests/test_elementwise_div_op.py index fbabc79be2406efa916d2acafe9ce3d69112b160..77b113af7693c4a71a5a13c791cfb3e0420f4ff8 100644 --- a/python/paddle/v2/fluid/tests/test_elementwise_div_op.py +++ b/python/paddle/v2/fluid/tests/test_elementwise_div_op.py @@ -1,16 +1,16 @@ # Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import unittest import numpy as np from op_test import OpTest @@ -45,6 +45,16 @@ class ElementwiseDivOp(OpTest): ['X'], 'Out', max_relative_error=0.05, no_grad_set=set('Y')) +class TestElementwiseDivOp_scalar(ElementwiseDivOp): + def setUp(self): + self.op_type = "elementwise_div" + self.inputs = { + 'X': np.random.uniform(0.1, 1, [2, 3, 4]).astype(np.float32), + 'Y': np.random.uniform(0.1, 1, [1]).astype(np.float32) + } + self.outputs = {'Out': self.inputs['X'] / self.inputs['Y']} + + class TestElementwiseDivOp_Vector(ElementwiseDivOp): def setUp(self): self.op_type = "elementwise_div" diff --git a/python/paddle/v2/fluid/tests/test_elementwise_max_op.py b/python/paddle/v2/fluid/tests/test_elementwise_max_op.py new file mode 100644 index 0000000000000000000000000000000000000000..0fc15693b1d2fbd89a1659dfbe5de0fff8d15762 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_elementwise_max_op.py @@ -0,0 +1,130 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +from op_test import OpTest + + +class TestElementwiseOp(OpTest): + def setUp(self): + self.op_type = "elementwise_max" + # If x and y have the same value, the max() is not differentiable. + # So we generate test data by the following method + # to avoid them being too close to each other. + x = np.random.uniform(0.1, 1, [13, 17]).astype("float32") + sgn = np.random.choice([-1, 1], [13, 17]).astype("float32") + y = x + sgn * np.random.uniform(0.1, 1, [13, 17]).astype("float32") + self.inputs = {'X': x, 'Y': y} + self.outputs = {'Out': np.maximum(self.inputs['X'], self.inputs['Y'])} + + def test_check_output(self): + self.check_output() + + def test_check_grad_normal(self): + self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.005) + + def test_check_grad_ingore_x(self): + self.check_grad( + ['Y'], 'Out', max_relative_error=0.005, no_grad_set=set("X")) + + def test_check_grad_ingore_y(self): + self.check_grad( + ['X'], 'Out', max_relative_error=0.005, no_grad_set=set('Y')) + + +class TestElementwiseMaxOp_scalar(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_max" + x = np.random.random_integers(-5, 5, [2, 3, 4]).astype("float32") + y = np.array([0.5]).astype("float32") + self.inputs = {'X': x, 'Y': y} + self.outputs = {'Out': np.maximum(self.inputs['X'], self.inputs['Y'])} + + +class TestElementwiseMaxOp_Vector(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_max" + x = np.random.random((32, )).astype("float32") + sgn = np.random.choice([-1, 1], (32, )).astype("float32") + y = x + sgn * np.random.uniform(0.1, 1, (32, )).astype("float32") + self.inputs = {'X': x, 'Y': y} + self.outputs = {'Out': np.maximum(self.inputs['X'], self.inputs['Y'])} + + +class TestElementwiseMaxOp_broadcast_0(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_max" + x = np.random.uniform(0.5, 1, (2, 3, 4)).astype(np.float32) + sgn = np.random.choice([-1, 1], (2, )).astype(np.float32) + y = x[:, 0, 0] + sgn * \ + np.random.uniform(1, 2, (2, )).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.attrs = {'axis': 0} + self.outputs = { + 'Out': + np.maximum(self.inputs['X'], self.inputs['Y'].reshape(2, 1, 1)) + } + + +class TestElementwiseMaxOp_broadcast_1(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_max" + x = np.random.uniform(0.5, 1, (2, 3, 4)).astype(np.float32) + sgn = np.random.choice([-1, 1], (3, )).astype(np.float32) + y = x[0, :, 0] + sgn * \ + np.random.uniform(1, 2, (3, )).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.attrs = {'axis': 1} + self.outputs = { + 'Out': + np.maximum(self.inputs['X'], self.inputs['Y'].reshape(1, 3, 1)) + } + + +class TestElementwiseMaxOp_broadcast_2(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_max" + x = np.random.uniform(0.5, 1, (2, 3, 4)).astype(np.float32) + sgn = np.random.choice([-1, 1], (4, )).astype(np.float32) + y = x[0, 0, :] + sgn * \ + np.random.uniform(1, 2, (4, )).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.outputs = { + 'Out': + np.maximum(self.inputs['X'], self.inputs['Y'].reshape(1, 1, 4)) + } + + +class TestElementwiseMaxOp_broadcast_3(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_max" + x = np.random.uniform(0.5, 1, (2, 3, 4, 5)).astype(np.float32) + sgn = np.random.choice([-1, 1], (3, 4)).astype(np.float32) + y = x[0, :, :, 0] + sgn * \ + np.random.uniform(1, 2, (3, 4)).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.attrs = {'axis': 1} + self.outputs = { + 'Out': + np.maximum(self.inputs['X'], self.inputs['Y'].reshape(1, 3, 4, 1)) + } + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_elementwise_min_op.py b/python/paddle/v2/fluid/tests/test_elementwise_min_op.py new file mode 100644 index 0000000000000000000000000000000000000000..51584d6980924f2f6dcaedf4eec7bc75de33564b --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_elementwise_min_op.py @@ -0,0 +1,130 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +from op_test import OpTest + + +class TestElementwiseOp(OpTest): + def setUp(self): + self.op_type = "elementwise_min" + # If x and y have the same value, the min() is not differentiable. + # So we generate test data by the following method + # to avoid them being too close to each other. + x = np.random.uniform(0.1, 1, [13, 17]).astype("float32") + sgn = np.random.choice([-1, 1], [13, 17]).astype("float32") + y = x + sgn * np.random.uniform(0.1, 1, [13, 17]).astype("float32") + self.inputs = {'X': x, 'Y': y} + self.outputs = {'Out': np.minimum(self.inputs['X'], self.inputs['Y'])} + + def test_check_output(self): + self.check_output() + + def test_check_grad_normal(self): + self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.005) + + def test_check_grad_ingore_x(self): + self.check_grad( + ['Y'], 'Out', max_relative_error=0.005, no_grad_set=set("X")) + + def test_check_grad_ingore_y(self): + self.check_grad( + ['X'], 'Out', max_relative_error=0.005, no_grad_set=set('Y')) + + +class TestElementwiseMinOp_scalar(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_min" + x = np.random.random_integers(-5, 5, [2, 3, 4]).astype("float32") + y = np.array([0.5]).astype("float32") + self.inputs = {'X': x, 'Y': y} + self.outputs = {'Out': np.minimum(self.inputs['X'], self.inputs['Y'])} + + +class TestElementwiseMaxOp_Vector(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_min" + x = np.random.random((32, )).astype("float32") + sgn = np.random.choice([-1, 1], (32, )).astype("float32") + y = x + sgn * np.random.uniform(0.1, 1, (32, )).astype("float32") + self.inputs = {'X': x, 'Y': y} + self.outputs = {'Out': np.minimum(self.inputs['X'], self.inputs['Y'])} + + +class TestElementwiseMaxOp_broadcast_0(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_min" + x = np.random.uniform(0.5, 1, (2, 3, 4)).astype(np.float32) + sgn = np.random.choice([-1, 1], (2, )).astype(np.float32) + y = x[:, 0, 0] + sgn * \ + np.random.uniform(1, 2, (2, )).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.attrs = {'axis': 0} + self.outputs = { + 'Out': + np.minimum(self.inputs['X'], self.inputs['Y'].reshape(2, 1, 1)) + } + + +class TestElementwiseMaxOp_broadcast_1(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_min" + x = np.random.uniform(0.5, 1, (2, 3, 4)).astype(np.float32) + sgn = np.random.choice([-1, 1], (3, )).astype(np.float32) + y = x[0, :, 0] + sgn * \ + np.random.uniform(1, 2, (3, )).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.attrs = {'axis': 1} + self.outputs = { + 'Out': + np.minimum(self.inputs['X'], self.inputs['Y'].reshape(1, 3, 1)) + } + + +class TestElementwiseMaxOp_broadcast_2(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_min" + x = np.random.uniform(0.5, 1, (2, 3, 4)).astype(np.float32) + sgn = np.random.choice([-1, 1], (4, )).astype(np.float32) + y = x[0, 0, :] + sgn * \ + np.random.uniform(1, 2, (4, )).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.outputs = { + 'Out': + np.minimum(self.inputs['X'], self.inputs['Y'].reshape(1, 1, 4)) + } + + +class TestElementwiseMaxOp_broadcast_3(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_min" + x = np.random.uniform(0.5, 1, (2, 3, 4, 5)).astype(np.float32) + sgn = np.random.choice([-1, 1], (3, 4)).astype(np.float32) + y = x[0, :, :, 0] + sgn * \ + np.random.uniform(1, 2, (3, 4)).astype(np.float32) + self.inputs = {'X': x, 'Y': y} + + self.attrs = {'axis': 1} + self.outputs = { + 'Out': + np.minimum(self.inputs['X'], self.inputs['Y'].reshape(1, 3, 4, 1)) + } + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_elementwise_mul_op.py b/python/paddle/v2/fluid/tests/test_elementwise_mul_op.py index ef3a829abc65f3af60b899ba50927424fd563607..12dfa6599cd634e1d806980f89e7c013b8eb8754 100644 --- a/python/paddle/v2/fluid/tests/test_elementwise_mul_op.py +++ b/python/paddle/v2/fluid/tests/test_elementwise_mul_op.py @@ -1,16 +1,16 @@ # Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import unittest import numpy as np from op_test import OpTest @@ -38,6 +38,16 @@ class ElementwiseMulOp(OpTest): self.check_grad(['X'], 'Out', no_grad_set=set('Y')) +class TestElementwiseMulOp_scalar(ElementwiseMulOp): + def setUp(self): + self.op_type = "elementwise_mul" + self.inputs = { + 'X': np.random.rand(2, 3, 4).astype(np.float32), + 'Y': np.random.rand(1).astype(np.float32) + } + self.outputs = {'Out': self.inputs['X'] * self.inputs['Y']} + + class TestElementwiseMulOp_Vector(ElementwiseMulOp): def setUp(self): self.op_type = "elementwise_mul" diff --git a/python/paddle/v2/fluid/tests/test_elementwise_pow_op.py b/python/paddle/v2/fluid/tests/test_elementwise_pow_op.py new file mode 100644 index 0000000000000000000000000000000000000000..e31749df9baf10215fcd0cca3c1097f00c163ec7 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_elementwise_pow_op.py @@ -0,0 +1,43 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +import unittest +import numpy as np +from op_test import OpTest + + +class TestElementwisePowOp(OpTest): + def setUp(self): + self.op_type = "elementwise_pow" + self.inputs = { + 'X': np.random.uniform(0.1, 1, [13, 17]).astype("float32"), + 'Y': np.random.uniform(0.1, 1, [13, 17]).astype("float32") + } + self.outputs = {'Out': np.power(self.inputs['X'], self.inputs['Y'])} + + def test_check_output(self): + self.check_output() + + +class TestElementwisePowOp_scalar(TestElementwisePowOp): + def setUp(self): + self.op_type = "elementwise_pow" + self.inputs = { + 'X': np.random.rand(2, 3, 4).astype('float32'), + 'Y': np.random.rand(1).astype('float32') + } + self.outputs = {'Out': np.power(self.inputs['X'], self.inputs['Y'])} + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_elementwise_sub_op.py b/python/paddle/v2/fluid/tests/test_elementwise_sub_op.py index db24db7b304e2ef8b946e128b22ed62d71260ddc..cf53d85bbad81f393a6263f8742fc942d357135f 100644 --- a/python/paddle/v2/fluid/tests/test_elementwise_sub_op.py +++ b/python/paddle/v2/fluid/tests/test_elementwise_sub_op.py @@ -1,16 +1,16 @@ # Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import unittest import numpy as np from op_test import OpTest @@ -40,6 +40,16 @@ class TestElementwiseOp(OpTest): ['X'], 'Out', max_relative_error=0.005, no_grad_set=set('Y')) +class TestElementwiseSubOp_scalar(TestElementwiseOp): + def setUp(self): + self.op_type = "elementwise_sub" + self.inputs = { + 'X': np.random.rand(2, 3, 4).astype(np.float32), + 'Y': np.random.rand(1).astype(np.float32) + } + self.outputs = {'Out': self.inputs['X'] - self.inputs['Y']} + + class TestElementwiseSubOp_Vector(TestElementwiseOp): def setUp(self): self.op_type = "elementwise_sub" diff --git a/python/paddle/v2/fluid/tests/test_clip.py b/python/paddle/v2/fluid/tests/test_error_clip.py similarity index 77% rename from python/paddle/v2/fluid/tests/test_clip.py rename to python/paddle/v2/fluid/tests/test_error_clip.py index 63353a10963532ec5b35eff22644adf4823243aa..6f7718f4d8751fc0514bafd342d9d4309d39b86b 100644 --- a/python/paddle/v2/fluid/tests/test_clip.py +++ b/python/paddle/v2/fluid/tests/test_error_clip.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import numpy as np import paddle.v2 as paddle diff --git a/python/paddle/v2/fluid/tests/test_exception.py b/python/paddle/v2/fluid/tests/test_exception.py index 98c4cbe3f2f2ec3a2036a3fdb84aa5ee4600695c..cd57ca586bfbc0773a2fd02d0c6d28182df2366b 100644 --- a/python/paddle/v2/fluid/tests/test_exception.py +++ b/python/paddle/v2/fluid/tests/test_exception.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core as core import unittest diff --git a/python/paddle/v2/fluid/tests/test_executor_and_mul.py b/python/paddle/v2/fluid/tests/test_executor_and_mul.py index e8baf631e52e4acf645b90b0a9d319d296354944..44f93be6cb3778822b849118396dc51084e66f6b 100644 --- a/python/paddle/v2/fluid/tests/test_executor_and_mul.py +++ b/python/paddle/v2/fluid/tests/test_executor_and_mul.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy diff --git a/python/paddle/v2/fluid/tests/test_expand_op.py b/python/paddle/v2/fluid/tests/test_expand_op.py index 0524f2041fb1b40e5214039fe72d672c1a3691d6..b1a1cbc0fae15b3e6159c9ec850ebde7cf19c228 100644 --- a/python/paddle/v2/fluid/tests/test_expand_op.py +++ b/python/paddle/v2/fluid/tests/test_expand_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_feed_fetch_method.py b/python/paddle/v2/fluid/tests/test_feed_fetch_method.py index 718311517dfee51f4e3c724074cd0018a3fa757c..827a7590ff3ca6b32fb37dda67468f3b1f95ee1b 100644 --- a/python/paddle/v2/fluid/tests/test_feed_fetch_method.py +++ b/python/paddle/v2/fluid/tests/test_feed_fetch_method.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core as core import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_fill_constant_batch_size_like_op.py b/python/paddle/v2/fluid/tests/test_fill_constant_batch_size_like_op.py index 0adc487c04ae4af7ab052530b1d3ca75ec1eddbd..f34a1ceb230208d6842f473636348e654154c2f4 100644 --- a/python/paddle/v2/fluid/tests/test_fill_constant_batch_size_like_op.py +++ b/python/paddle/v2/fluid/tests/test_fill_constant_batch_size_like_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_fill_constant_op.py b/python/paddle/v2/fluid/tests/test_fill_constant_op.py index 50d4ccb3bdd115f8fa5ea53d5751426cac44cedb..a05fa39729d4cd91c93cab935ba6f8cc68a57c52 100644 --- a/python/paddle/v2/fluid/tests/test_fill_constant_op.py +++ b/python/paddle/v2/fluid/tests/test_fill_constant_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_fill_op.py b/python/paddle/v2/fluid/tests/test_fill_op.py index 42b06ec87c2b1811aa678f0563522f3d90405478..901546f6f8965e3ffb55c6686c229061e15eae8d 100644 --- a/python/paddle/v2/fluid/tests/test_fill_op.py +++ b/python/paddle/v2/fluid/tests/test_fill_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_fill_zeros_like_op.py b/python/paddle/v2/fluid/tests/test_fill_zeros_like_op.py index a28bed9697309da29dce1220dc8080e0ad2083fd..b7f0b96647d5fe645157498f15a8a5d0e4430c67 100644 --- a/python/paddle/v2/fluid/tests/test_fill_zeros_like_op.py +++ b/python/paddle/v2/fluid/tests/test_fill_zeros_like_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_framework_debug_str.py b/python/paddle/v2/fluid/tests/test_framework_debug_str.py index 6c82e67220f36217d88d2b7f73ede8f85e11d632..f8fcfb2249bc635ce297381359db954bfd62df26 100644 --- a/python/paddle/v2/fluid/tests/test_framework_debug_str.py +++ b/python/paddle/v2/fluid/tests/test_framework_debug_str.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest from paddle.v2.fluid.framework import Program diff --git a/python/paddle/v2/fluid/tests/test_ftrl_op.py b/python/paddle/v2/fluid/tests/test_ftrl_op.py index 599233efd93d91514171c46f70994bc45c9c6722..895337de0fb9885b3d78d0af23c03fd82a360497 100644 --- a/python/paddle/v2/fluid/tests/test_ftrl_op.py +++ b/python/paddle/v2/fluid/tests/test_ftrl_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_gather_op.py b/python/paddle/v2/fluid/tests/test_gather_op.py index 95093f9b846b7bfef4ad3b3dd777c5f7980fe83a..76756367976c7056904d6d757ef9635261dc8985 100644 --- a/python/paddle/v2/fluid/tests/test_gather_op.py +++ b/python/paddle/v2/fluid/tests/test_gather_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_gaussian_random_op.py b/python/paddle/v2/fluid/tests/test_gaussian_random_op.py index bf4785211e8e6d4af97f49d67847df7ac72dfa71..79beb8b1fcef610bc2f3e8d18da4345baa9b99c3 100644 --- a/python/paddle/v2/fluid/tests/test_gaussian_random_op.py +++ b/python/paddle/v2/fluid/tests/test_gaussian_random_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy @@ -32,7 +33,7 @@ class TestGaussianRandomOp(unittest.TestCase): self.gaussian_random_test(place=fluid.CPUPlace()) def test_gpu(self): - if core.is_compile_gpu(): + if core.is_compiled_with_cuda(): self.gaussian_random_test(place=fluid.CUDAPlace(0)) def gaussian_random_test(self, place): diff --git a/python/paddle/v2/fluid/tests/test_get_places_op.py b/python/paddle/v2/fluid/tests/test_get_places_op.py index b44011fb76be712d11bbd72ce95027a439a4d2c1..68698c5f4a2354595dcc0cb271ad9f57a35386e1 100644 --- a/python/paddle/v2/fluid/tests/test_get_places_op.py +++ b/python/paddle/v2/fluid/tests/test_get_places_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid as fluid import decorators import unittest diff --git a/python/paddle/v2/fluid/tests/test_gradient_clip.py b/python/paddle/v2/fluid/tests/test_gradient_clip.py new file mode 100644 index 0000000000000000000000000000000000000000..9337791c21183fd7c2e5d6b9d47c99d762c93d46 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_gradient_clip.py @@ -0,0 +1,82 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import numpy as np +import paddle.v2 as paddle +import paddle.v2.fluid as fluid + +BATCH_SIZE = 128 +CLIP = 1 + +prog = fluid.framework.Program() +with fluid.program_guard(main_program=prog): + image = fluid.layers.data(name='x', shape=[784], dtype='float32') + + hidden1 = fluid.layers.fc(input=image, size=128, act='relu') + hidden2 = fluid.layers.fc(input=hidden1, size=64, act='relu') + predict = fluid.layers.fc(input=hidden2, size=10, act='softmax') + + label = fluid.layers.data(name='y', shape=[1], dtype='int64') + + cost = fluid.layers.cross_entropy(input=predict, label=label) + avg_cost = fluid.layers.mean(x=cost) + +prog_clip = prog.clone() + +avg_cost_clip = prog_clip.block(0).var(avg_cost.name) + +p_g = fluid.backward.append_backward(loss=avg_cost) +p_g_clip = fluid.backward.append_backward(loss=avg_cost_clip) + +with fluid.program_guard(main_program=prog_clip): + fluid.clip.set_gradient_clip( + fluid.clip.GradientClipByGlobalNorm(clip_norm=CLIP)) + p_g_clip = fluid.clip.append_gradient_clip_ops(p_g_clip) + +grad_list = [elem[1] for elem in p_g] +grad_clip_list = [elem[1] for elem in p_g_clip] + +train_reader = paddle.batch( + paddle.reader.shuffle( + paddle.dataset.mnist.train(), buf_size=8192), + batch_size=BATCH_SIZE) + +place = fluid.CPUPlace() +exe = fluid.Executor(place) +feeder = fluid.DataFeeder(feed_list=[image, label], place=place) +exe.run(fluid.default_startup_program()) + +count = 0 +for data in train_reader(): + count += 1 + if count > 5: + break + out = exe.run(prog, feed=feeder.feed(data), fetch_list=grad_list) + out_clip = exe.run(prog_clip, + feed=feeder.feed(data), + fetch_list=grad_clip_list) + global_norm = 0 + for v in out[1:]: + global_norm += np.sum(np.power(v, 2)) + global_norm = np.sqrt(global_norm) + + global_norm_clip = 0 + for v in out_clip[1:]: + global_norm_clip += np.sum(np.power(v, 2)) + global_norm_clip = np.sqrt(global_norm_clip) + + if not np.isclose( + a=global_norm_clip, b=np.minimum(global_norm, CLIP), rtol=5e-3): + exit(1) +exit(0) diff --git a/python/paddle/v2/fluid/tests/test_gru_op.py b/python/paddle/v2/fluid/tests/test_gru_op.py index a6647d1bf28b8b4412d40f795d59eb526e0b7781..69cfd6c481cba30148b0f367711c7ee8c25acd3a 100644 --- a/python/paddle/v2/fluid/tests/test_gru_op.py +++ b/python/paddle/v2/fluid/tests/test_gru_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import math diff --git a/python/paddle/v2/fluid/tests/test_gru_unit_op.py b/python/paddle/v2/fluid/tests/test_gru_unit_op.py index 53f10c32c7cd143b89da77e24555a7f59473f3a1..71f13c4513622579b5a8a718954e6b493b929ce1 100644 --- a/python/paddle/v2/fluid/tests/test_gru_unit_op.py +++ b/python/paddle/v2/fluid/tests/test_gru_unit_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import math import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_hinge_loss_op.py b/python/paddle/v2/fluid/tests/test_hinge_loss_op.py index dc7774d01c0a6d6a80d0e914cd327e1a4f9919e9..71ff47316eccd9a9f0a34942feaea6722a4b28f5 100644 --- a/python/paddle/v2/fluid/tests/test_hinge_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_hinge_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_huber_loss_op.py b/python/paddle/v2/fluid/tests/test_huber_loss_op.py index 18a48bb18ced83d79e602e0103a3e8d0a87c66ef..e4560af77826a7368b305eb58515e4e44b450a78 100644 --- a/python/paddle/v2/fluid/tests/test_huber_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_huber_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_im2sequence_op.py b/python/paddle/v2/fluid/tests/test_im2sequence_op.py new file mode 100644 index 0000000000000000000000000000000000000000..2cab3e31a50034e3b1b362b59690e425aef1c399 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_im2sequence_op.py @@ -0,0 +1,167 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. +import unittest +import numpy as np +from op_test import OpTest + + +def get_output_shape(attrs, in_shape): + img_height = in_shape[2] + img_width = in_shape[3] + + paddings = attrs['paddings'] + kernels = attrs['kernels'] + strides = attrs['strides'] + + output_height = \ + 1 + \ + (img_height + paddings[0] + paddings[2] - kernels[0] + strides[0] - 1) / \ + strides[0] + + output_width = \ + 1 + \ + (img_width + paddings[1] + paddings[3] - kernels[1] + strides[1] - 1) / \ + strides[1] + + return output_height, output_width + + +def im2col(attrs, im, col): + """ + im: {CHW} + col: + {outputHeight, outputWidth, inputChannels, filterHeight, filterWidth} + """ + input_channels, input_height, input_width = im.shape + output_height, output_width, _, filter_height, filter_width = col.shape + + stride_height, stride_width = attrs['strides'] + padding_height, padding_width = attrs['paddings'][0:2] + + for col_row_idx in range(0, output_height): + for col_col_idx in range(0, output_width): + for channel in range(0, input_channels): + for filter_row_idx in range(0, filter_height): + for filter_col_idx in range(0, filter_width): + im_row_offset = col_row_idx * stride_height \ + + filter_row_idx - padding_height + + im_col_offset = col_col_idx * stride_width \ + + filter_col_idx - padding_width + + if (im_row_offset < 0 or + im_row_offset >= input_height or + im_col_offset < 0 or + im_col_offset >= input_width): + col[col_row_idx][col_col_idx][channel][\ + filter_row_idx][filter_col_idx] = 0.0 + else: + im_offset = (channel * input_height + im_row_offset \ + ) * input_width + im_col_offset + + col[col_row_idx][col_col_idx][channel][\ + filter_row_idx][filter_col_idx] = im[channel][ \ + im_row_offset][im_col_offset] + + +def Im2Sequence(inputs, attrs): + output_height, output_width = get_output_shape(attrs, inputs.shape) + img_channels = inputs.shape[1] + batch_size = inputs.shape[0] + out = np.zeros([ + batch_size, output_height, output_width, img_channels, + attrs['kernels'][0], attrs['kernels'][1] + ]).astype("float32") + + for i in range(len(inputs)): + im2col(attrs, inputs[i], out[i]) + + out = out.reshape([ + batch_size * output_height * output_width, + img_channels * attrs['kernels'][0] * attrs['kernels'][1] + ]) + return out + + +class TestBlockExpandOp(OpTest): + def config(self): + self.batch_size = 1 + self.img_channels = 3 + self.img_height = 4 + self.img_width = 4 + self.attrs = { + 'kernels': [2, 2], + 'strides': [1, 1], + 'paddings': [1, 1, 1, 1] + } + + def setUp(self): + self.config() + self.op_type = "im2sequence" + x = np.random.uniform(0.1, 1, [ + self.batch_size, self.img_channels, self.img_height, self.img_width + ]).astype("float32") + + out = Im2Sequence(x, self.attrs) + self.inputs = {'X': x} + self.outputs = {'Out': out} + + def test_check_output(self): + self.check_output() + + def test_check_grad_normal(self): + self.check_grad(['X'], 'Out') + + +class TestBlockExpandOpCase2(TestBlockExpandOp): + def config(self): + self.batch_size = 2 + self.img_channels = 3 + self.img_height = 4 + self.img_width = 5 + self.attrs = { + 'kernels': [2, 1], + 'strides': [2, 1], + 'paddings': [2, 1, 2, 1] + } + + +class TestBlockExpandOpCase3(TestBlockExpandOp): + def config(self): + self.batch_size = 3 + self.img_channels = 1 + self.img_height = 4 + self.img_width = 5 + self.attrs = { + 'kernels': [2, 1], + 'strides': [2, 1], + 'paddings': [2, 0, 2, 0] + } + + +class TestBlockExpandOpCase4(TestBlockExpandOp): + def config(self): + self.batch_size = 2 + self.img_channels = 2 + self.img_height = 3 + self.img_width = 3 + self.attrs = { + 'kernels': [2, 2], + 'strides': [1, 1], + 'paddings': [0, 0, 0, 0] + } + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_image_classification_layer.py b/python/paddle/v2/fluid/tests/test_image_classification_layer.py index 9d676e87594dfad7ede3ac237effb234f44e8369..c64cfed5f583fb48588e479f862917a7b4d37c2c 100644 --- a/python/paddle/v2/fluid/tests/test_image_classification_layer.py +++ b/python/paddle/v2/fluid/tests/test_image_classification_layer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid as fluid diff --git a/python/paddle/v2/fluid/tests/test_infer_shape.py b/python/paddle/v2/fluid/tests/test_infer_shape.py index 0c2a6f1423c45b3a5c8cb9f0d9b3d7004f997b36..521096388a338ceda0c71f6b7963c06f85263c69 100644 --- a/python/paddle/v2/fluid/tests/test_infer_shape.py +++ b/python/paddle/v2/fluid/tests/test_infer_shape.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_inference_model_io.py b/python/paddle/v2/fluid/tests/test_inference_model_io.py index c5cad2166bd17241c6c86e5cd0614df36ad2961c..adf428aa848268704c1749b360a4b499be2383e8 100644 --- a/python/paddle/v2/fluid/tests/test_inference_model_io.py +++ b/python/paddle/v2/fluid/tests/test_inference_model_io.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_initializer.py b/python/paddle/v2/fluid/tests/test_initializer.py index fa3c2afeedca4247379a34dea957bcceb9f134e1..67746b4d7d96a6bb039e05795f11f3cb117cf85c 100644 --- a/python/paddle/v2/fluid/tests/test_initializer.py +++ b/python/paddle/v2/fluid/tests/test_initializer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import unittest diff --git a/python/paddle/v2/fluid/tests/test_iou_similarity_op.py b/python/paddle/v2/fluid/tests/test_iou_similarity_op.py new file mode 100644 index 0000000000000000000000000000000000000000..128f2e4977195a563efcd26364cc6261da2dd685 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_iou_similarity_op.py @@ -0,0 +1,55 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +import sys +import math +from op_test import OpTest + + +class TestIOUSimilarityOp(OpTest): + def test_check_output(self): + self.check_output() + + def setUp(self): + self.op_type = "iou_similarity" + self.boxes1 = np.array( + [[4.0, 3.0, 7.0, 5.0], [5.0, 6.0, 10.0, 7.0]]).astype('float32') + self.boxes2 = np.array([[3.0, 4.0, 6.0, 8.0], [14.0, 14.0, 15.0, 15.0], + [0.0, 0.0, 20.0, 20.0]]).astype('float32') + self.output = np.array( + [[2.0 / 16.0, 0, 6.0 / 400.0], + [1.0 / 16.0, 0.0, 5.0 / 400.0]]).astype('float32') + + self.inputs = {'X': self.boxes1, 'Y': self.boxes2} + + self.outputs = {'Out': self.output} + + +class TestIOUSimilarityOpWithLoD(TestIOUSimilarityOp): + def test_check_output(self): + self.check_output() + + def setUp(self): + super(TestIOUSimilarityOpWithLoD, self).setUp() + self.boxes1_lod = [[0, 1, 2]] + self.output_lod = [[0, 1, 2]] + + self.inputs = {'X': (self.boxes1, self.boxes1_lod), 'Y': self.boxes2} + self.outputs = {'Out': (self.output, self.output_lod)} + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_is_empty_op.py b/python/paddle/v2/fluid/tests/test_is_empty_op.py index d6876a885f53ae2dc574871013d570a801b73c7f..7c17e3d57aa3eb93d0bc82ca81a96a7a1866f15c 100644 --- a/python/paddle/v2/fluid/tests/test_is_empty_op.py +++ b/python/paddle/v2/fluid/tests/test_is_empty_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from paddle.v2.fluid.op import Operator diff --git a/python/paddle/v2/fluid/tests/test_l1_norm_op.py b/python/paddle/v2/fluid/tests/test_l1_norm_op.py index 92484c49f03ebad6f07ff2613c346545d15f1d72..bbc20878468371a547d2b9e275157dab51a79408 100644 --- a/python/paddle/v2/fluid/tests/test_l1_norm_op.py +++ b/python/paddle/v2/fluid/tests/test_l1_norm_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import unittest from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_label_smooth_op.py b/python/paddle/v2/fluid/tests/test_label_smooth_op.py new file mode 100644 index 0000000000000000000000000000000000000000..19a4df57446c0c83b415909df3e0246bf2716881 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_label_smooth_op.py @@ -0,0 +1,55 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +from op_test import OpTest + + +class TestLabelSmoothOp(OpTest): + def config(self): + self.op_type = "label_smooth" + self.epsilon = 0.1 + batch_size, self.label_dim = 5, 10 + self.label = np.zeros((batch_size, self.label_dim)).astype("float64") + nonzero_index = np.random.randint(self.label_dim, size=(batch_size)) + self.label[np.arange(batch_size), nonzero_index] = 1 + + def setUp(self): + self.config() + smoothed_label = (1 - self.epsilon + ) * self.label + self.epsilon / self.label_dim + self.inputs = {'X': self.label} + self.attrs = {'epsilon': self.epsilon} + self.outputs = {'Out': smoothed_label} + + def test_check_output(self): + self.check_output() + + def test_check_grad(self): + self.check_grad(["X"], "Out") + + +class TestLabelSmoothOpWithPriorDist(TestLabelSmoothOp): + def setUp(self): + self.config() + dist = np.random.random((1, self.label_dim)) + smoothed_label = (1 - self.epsilon) * self.label + self.epsilon * dist + self.inputs = {'X': self.label, 'PriorDist': dist} + self.attrs = {'epsilon': self.epsilon} + self.outputs = {'Out': smoothed_label} + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_layer_norm_op.py b/python/paddle/v2/fluid/tests/test_layer_norm_op.py new file mode 100644 index 0000000000000000000000000000000000000000..68cf8673cd46677065588f652482cd0df08b3450 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_layer_norm_op.py @@ -0,0 +1,252 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +import unittest +import numpy as np + +from operator import mul +from op_test import OpTest +import paddle.v2.fluid.core as core +from paddle.v2.fluid.op import Operator +from paddle.v2.fluid.framework import grad_var_name + + +def _reference_layer_norm_naive(x, scale, beta, epsilon, begin_norm_axis=1): + x_shape = x.shape + N = reduce(mul, x_shape[0:begin_norm_axis], 1) + D = reduce(mul, x_shape[begin_norm_axis:len(x_shape)], 1) + x.shape = [N, D] + + mean = np.mean(x, axis=1) + var = np.var(x, axis=1) + epsilon + output = scale.reshape([1, D]) * np.divide( + (x - mean.reshape([N, 1])), + (np.sqrt(var)).reshape([N, 1])) + beta.reshape([1, D]) + + x.shape, output.shape = x_shape, x_shape + return output, mean, var + + +def _reference_layer_norm_grad(x, grad_y, scale, mean, var, begin_norm_axis=1): + x_shape = x.shape + scale_shape = scale.shape + N = reduce(mul, x_shape[0:begin_norm_axis], 1) + D = reduce(mul, x_shape[begin_norm_axis:len(x_shape)], 1) + x.shape, grad_y.shape = [N, D], [N, D] + var.shape, mean.shape = [N, 1], [N, 1] + scale.shape = [1, D] + + # d_bias + d_bias = np.sum(grad_y, axis=0).reshape([1, D]) + # d_scale + d_scale = np.sum(((x - mean) * np.sqrt(1 / var)) * grad_y, + axis=0).reshape([1, D]) + # dx + dx_end = scale * np.sqrt(1.0 / var) * grad_y + d_mean_0 = np.sum(-np.sqrt(1.0 / var) * grad_y * scale, axis=1).reshape( + [N, 1]) # the second part equals to zero. + d_mean = 1.0 / D * d_mean_0 + d_std = np.sum( + -(1.0 / var) * (x - mean) * grad_y * scale, axis=1).reshape([N, 1]) * ( + 1.0 / D * np.sqrt(1.0 / var).reshape([N, 1]) * (x - mean)) + + grad_x = dx_end + d_mean + d_std + + grad_y.shape = x_shape + x.shape = x_shape + scale.shape = scale_shape + return grad_x, d_scale, d_bias + + +def get_backward_op(scope, op, no_grad_set): + backward_op = core.Operator.backward(op, no_grad_set) + for input in backward_op.input_vars(): + var = scope.var(input) + var.get_tensor() + for output in backward_op.output_vars(): + var = scope.var(output) + var.get_tensor() + return backward_op + + +def create_or_get_tensor(scope, var_name, var, place): + tensor = scope.var(var_name).get_tensor() + if var is not None: + assert isinstance(var, np.ndarray) + tensor.set_lod([[]]) + tensor.set_dims(var.shape) + tensor.set(var, place) + return tensor + + +def set_output_grad(scope, outputs, place, feed_dict=None): + def __set_tensor__(name, data=None): + out_tensor = scope.find_var(name).get_tensor() + grad_tensor = scope.var(grad_var_name(name)).get_tensor() + out_dtype = out_tensor.dtype() + if data is None: + if out_dtype == core.DataType.FP64: + data = np.ones(out_tensor.shape(), dtype=np.float64) + elif out_dtype == core.DataType.FP32: + data = np.ones(out_tensor.shape(), dtype=np.float32) + else: + raise ValueError("Not supported data type " + str(out_dtype)) + grad_tensor.set(data, place) + + for output in outputs: + data = None + if output in feed_dict: + data = feed_dict[output] + __set_tensor__(output, data) + + +class TestLayerNormdOp(OpTest): + def __assert_close(self, tensor, np_array, msg, atol=1e-4): + self.assertTrue( + np.allclose( + np.array(tensor).reshape(np_array.shape), np_array, atol=atol), + msg) + + def __assert_grad_close(self, + tensor, + np_array, + name, + place, + max_relative_error=0.02): + a = np.array(tensor).reshape(np_array.shape) + b = np_array + abs_a = np.abs(a) + abs_a[abs_a < 1e-5] = 1 + + diff_mat = np.abs(a - b) / abs_a + max_diff = np.max(diff_mat) + + def err_msg(): + offset = np.argmax(diff_mat > max_relative_error) + return ("%s Variable %s max gradient diff %f over limit %f, " + "the first error element is %d, %f, %f") % ( + "Gradient Check On %s" % str(place), name, max_diff, + max_relative_error, offset, a.flatten()[offset], + b.flatten()[offset]) + + self.assertLessEqual(max_diff, max_relative_error, err_msg()) + + def check_forward_backward(self, shape, begin_norm_axis): + def test_with_place(place, shape, begin_norm_axis=1): + # setUp + assert begin_norm_axis > 0 and begin_norm_axis < len( + shape), 'begin_norm_axis must be between 0 and len(shape)-1.' + # attr + epsilon = 0.00001 + x_shape = shape + D = reduce(mul, x_shape[begin_norm_axis:len(x_shape)], 1) + scale_shape = [D] + np.random.random(123) + x_val = np.random.random_sample(x_shape).astype(np.float32) + scale_val = np.random.random_sample(scale_shape).astype(np.float32) + bias_val = np.random.random_sample(scale_shape).astype(np.float32) + y_grad = np.random.random_sample(x_shape).astype(np.float32) + + # run forward + y_out, saved_mean, var_ref = _reference_layer_norm_naive( + x_val, scale_val, bias_val, epsilon, begin_norm_axis) + naive_fw = {"Y": y_out, "Mean": saved_mean, "Variance": var_ref} + + # get gradient + x_grad_ref, scale_grad_ref, bias_grad_ref = _reference_layer_norm_grad( + x_val, y_grad, scale_val, saved_mean, var_ref, begin_norm_axis) + naive_grad = { + "X": x_grad_ref, + "Scale": scale_grad_ref, + "Bias": bias_grad_ref + } + + scope = core.Scope() + + # create input + input_map = {"X": x_val, "Scale": scale_val, "Bias": bias_val} + for i_name in input_map: + create_or_get_tensor(scope, i_name, input_map[i_name], place) + + # create output + output_map = {"Y": None, "Mean": None, "Variance": None} + output_tensor = {} + for o_name in output_map: + output_tensor[o_name] = create_or_get_tensor( + scope, o_name, output_map[o_name], place) + + layer_norm_op = Operator( + "layer_norm", + # inputs + X="X", + Scale="Scale", + Bias="Bias", + # outputs + Y="Y", + Mean="Mean", + Variance="Variance", + # attrs + epsilon=epsilon, + begin_norm_axis=begin_norm_axis) + + layer_norm_op.run(scope, place) + + # check forward result + atol = 5e-2 if isinstance(place, core.CUDAPlace) else 1e-4 + for o_tensor in output_tensor: + self.__assert_close(output_tensor[o_tensor], naive_fw[o_tensor], + o_tensor, atol) + + # run backward + layer_norm_op_grad = get_backward_op(scope, layer_norm_op, set()) + set_output_grad( + scope, ["Y", "Mean", "Variance"], + place, + feed_dict={"Y": y_grad}) + layer_norm_op_grad.run(scope, place) + + # get output + grad_tensor = {} + for o_name in naive_grad: + grad_tensor[o_name] = x_ = create_or_get_tensor( + scope, grad_var_name(o_name), None, place) + + # check gradient output + for o_grad in naive_grad: + self.__assert_grad_close(grad_tensor[o_grad], + naive_grad[o_grad], o_grad + "@GRAD", + place) + + places = [core.CPUPlace()] + if core.is_compiled_with_cuda() and core.op_support_gpu("layer_norm"): + places.append(core.CUDAPlace(0)) + + for place in places: + test_with_place(place, shape, begin_norm_axis) + + def test_check_forward_backward_with_scale_and_bias(self): + self.check_forward_backward(shape=[2, 3, 4, 5], begin_norm_axis=1) + self.check_forward_backward(shape=[2, 3, 4, 5], begin_norm_axis=3) + + def test_check_forward_backward_with_scale(self): + pass # TODO(zcd) + + def test_check_forward_backward_with_bias(self): + pass # TODO(zcd) + + def test_check_forward_backward(self): + pass # TODO(zcd) + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_layers.py b/python/paddle/v2/fluid/tests/test_layers.py index a4e155b534a41e385167e6a6f01e32cfedf580e2..3f54e28defb76d3430a82e791578e20b84833f16 100644 --- a/python/paddle/v2/fluid/tests/test_layers.py +++ b/python/paddle/v2/fluid/tests/test_layers.py @@ -1,23 +1,25 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import unittest import paddle.v2.fluid.layers as layers import paddle.v2.fluid.nets as nets -from paddle.v2.fluid.framework import Program, program_guard +from paddle.v2.fluid.framework import Program, program_guard, default_main_program from paddle.v2.fluid.param_attr import ParamAttr +import decorators class TestBook(unittest.TestCase): @@ -200,6 +202,18 @@ class TestBook(unittest.TestCase): x_t=x_t, hidden_t_prev=prev_hidden, cell_t_prev=prev_cell)) print(str(program)) + def test_dynamic_lstmp(self): + program = Program() + with program_guard(program): + hidden_dim, proj_dim = 16, 8 + seq_data = layers.data( + name='seq_data', shape=[10, 10], dtype='float32', lod_level=1) + fc_out = layers.fc(input=seq_data, size=4 * hidden_dim) + self.assertIsNotNone( + layers.dynamic_lstmp( + input=fc_out, size=4 * hidden_dim, proj_size=proj_dim)) + print(str(program)) + def test_sequence_softmax(self): program = Program() with program_guard(program): @@ -216,6 +230,77 @@ class TestBook(unittest.TestCase): self.assertIsNotNone(x) print(str(program)) + def test_sequence_reshape(self): + program = Program() + with program_guard(program): + x = layers.data(name='x', shape=[8], dtype='float32', lod_level=1) + out = layers.sequence_reshape(input=x, new_dim=16) + self.assertIsNotNone(out) + print(str(program)) + + def test_im2sequence(self): + print("test_im2sequence") + program = Program() + with program_guard(program): + x = layers.data(name='x', shape=[3, 128, 128], dtype='float32') + output = layers.im2sequence( + input=x, stride=[1, 1], filter_size=[2, 2]) + self.assertIsNotNone(output) + print(str(program)) + + @decorators.prog_scope() + def test_nce(self): + window_size = 5 + words = [] + for i in xrange(window_size): + words.append( + layers.data( + name='word_{0}'.format(i), shape=[1], dtype='int64')) + + dict_size = 10000 + label_word = int(window_size / 2) + 1 + + embs = [] + for i in xrange(window_size): + if i == label_word: + continue + + emb = layers.embedding( + input=words[i], + size=[dict_size, 32], + param_attr='emb.w', + is_sparse=True) + + embs.append(emb) + + embs = layers.concat(input=embs, axis=1) + loss = layers.nce(input=embs, + label=words[label_word], + num_total_classes=dict_size, + param_attr='nce.w', + bias_attr='nce.b') + avg_loss = layers.mean(x=loss) + self.assertIsNotNone(avg_loss) + print(str(default_main_program())) + + def test_row_conv(self): + program = Program() + with program_guard(program): + x = layers.data(name='x', shape=[16], dtype='float32', lod_level=1) + out = layers.row_conv(input=x, future_context_size=2) + self.assertIsNotNone(out) + print(str(program)) + + def test_multiplex(self): + program = Program() + with program_guard(program): + x1 = layers.data(name='x1', shape=[4], dtype='float32') + x2 = layers.data(name='x2', shape=[4], dtype='float32') + index = layers.data(name='index', shape=[1], dtype='int32') + out = layers.multiplex(inputs=[x1, x2], index=index) + self.assertIsNotNone(out) + print(str(program)) + if __name__ == '__main__': unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_learning_rate_decay.py b/python/paddle/v2/fluid/tests/test_learning_rate_decay.py new file mode 100644 index 0000000000000000000000000000000000000000..dc348cf2d21693290095900f8ab63c29923b4673 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_learning_rate_decay.py @@ -0,0 +1,110 @@ +# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest + +import math +import paddle.v2.fluid.framework as framework +import paddle.v2.fluid as fluid +import paddle.v2.fluid.layers as layers +import paddle.v2.fluid.learning_rate_decay as lr_decay + + +def exponential_decay(learning_rate, + global_step, + decay_steps, + decay_rate, + staircase=False): + exponent = float(global_step) / float(decay_steps) + if staircase: + exponent = math.floor(exponent) + return learning_rate * decay_rate**exponent + + +def natural_exp_decay(learning_rate, + global_step, + decay_steps, + decay_rate, + staircase=False): + exponent = float(global_step) / float(decay_steps) + if staircase: + exponent = math.floor(exponent) + return learning_rate * math.exp(-1 * decay_rate * exponent) + + +def inverse_time_decay(learning_rate, + global_step, + decay_steps, + decay_rate, + staircase=False): + temp = float(global_step) / float(decay_steps) + if staircase: + temp = math.floor(temp) + return learning_rate / (1 + decay_rate * temp) + + +class TestLearningRateDecay(unittest.TestCase): + def check_decay(self, python_decay_fn, fluid_decay_fn, staircase): + init_lr = 1.0 + decay_steps = 5 + decay_rate = 0.5 + + global_step = layers.create_global_var( + shape=[1], value=0.0, dtype='float32', persistable=True) + + decayed_lr = fluid_decay_fn( + learning_rate=init_lr, + global_step=global_step, + decay_steps=decay_steps, + decay_rate=decay_rate, + staircase=staircase) + layers.increment(global_step, 1.0) + + place = fluid.CPUPlace() + exe = fluid.Executor(place) + + exe.run(fluid.default_startup_program()) + for step in range(10): + step_val, lr_val = exe.run(fluid.default_main_program(), + feed=[], + fetch_list=[global_step, decayed_lr]) + python_decayed_lr = python_decay_fn( + learning_rate=init_lr, + global_step=step, + decay_steps=decay_steps, + decay_rate=decay_rate, + staircase=staircase) + self.assertAlmostEqual(python_decayed_lr, lr_val[0]) + + def test_decay(self): + decay_fns = [ + (exponential_decay, lr_decay.exponential_decay, True), + (exponential_decay, lr_decay.exponential_decay, False), + (natural_exp_decay, lr_decay.natural_exp_decay, True), + (natural_exp_decay, lr_decay.natural_exp_decay, False), + (inverse_time_decay, lr_decay.inverse_time_decay, True), + (inverse_time_decay, lr_decay.inverse_time_decay, False), + ] + + for py_decay_fn, fluid_decay_fn, staircase in decay_fns: + print("decay_fn=" + str(py_decay_fn) + " staircase=" + str( + staircase)) + main_program = framework.Program() + startup_program = framework.Program() + with framework.program_guard(main_program, startup_program): + self.check_decay(py_decay_fn, fluid_decay_fn, staircase) + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_linear_chain_crf_op.py b/python/paddle/v2/fluid/tests/test_linear_chain_crf_op.py index cd917dff7f26a04cd289470b3191ad08d7935169..cbfd9d5e5b359b708e0e84c2c40ca0ef41cc35b6 100644 --- a/python/paddle/v2/fluid/tests/test_linear_chain_crf_op.py +++ b/python/paddle/v2/fluid/tests/test_linear_chain_crf_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import random import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_lod_array_length_op.py b/python/paddle/v2/fluid/tests/test_lod_array_length_op.py index f80136cb0d8bf5ae87cfa140d0b8aa895d38568d..eff28368f1a6a78b8e7e9df7e281a1eb3b11288e 100644 --- a/python/paddle/v2/fluid/tests/test_lod_array_length_op.py +++ b/python/paddle/v2/fluid/tests/test_lod_array_length_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.layers as layers from paddle.v2.fluid.executor import Executor diff --git a/python/paddle/v2/fluid/tests/test_lod_rank_table.py b/python/paddle/v2/fluid/tests/test_lod_rank_table.py index 673605d79c72ef2d1251659afeee458ee0d0ac91..eb0392e8bf7111f223a54430ab7a6646a4f35f71 100644 --- a/python/paddle/v2/fluid/tests/test_lod_rank_table.py +++ b/python/paddle/v2/fluid/tests/test_lod_rank_table.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from paddle.v2.fluid.layers import lod_rank_table, data from paddle.v2.fluid.executor import Executor import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_lod_reset_op.py b/python/paddle/v2/fluid/tests/test_lod_reset_op.py index d799dbfa217bc6012e1225216882a95fc0998544..4ee360403e88c6a4f8d5fcc6d6ce8acc865e6277 100644 --- a/python/paddle/v2/fluid/tests/test_lod_reset_op.py +++ b/python/paddle/v2/fluid/tests/test_lod_reset_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_lod_tensor_array.py b/python/paddle/v2/fluid/tests/test_lod_tensor_array.py index c593b1e06132036ecdf89ef47b1074198c346cc6..0f3ac3c03dbe43c2e977a5e98faff31d6c231acd 100644 --- a/python/paddle/v2/fluid/tests/test_lod_tensor_array.py +++ b/python/paddle/v2/fluid/tests/test_lod_tensor_array.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core import numpy diff --git a/python/paddle/v2/fluid/tests/test_lod_tensor_array_ops.py b/python/paddle/v2/fluid/tests/test_lod_tensor_array_ops.py index 5887f9799a11106656306852a7ba4f7b2ef9ebd4..c2d04db99b969430f3a70e45c92967f548ba513b 100644 --- a/python/paddle/v2/fluid/tests/test_lod_tensor_array_ops.py +++ b/python/paddle/v2/fluid/tests/test_lod_tensor_array_ops.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core import numpy diff --git a/python/paddle/v2/fluid/tests/test_log_loss_op.py b/python/paddle/v2/fluid/tests/test_log_loss_op.py index fde99bfaa16e6d260a4eefc295dbc765c3816861..338355d0c4d33a8927e689001bae3a236a0ecd3a 100644 --- a/python/paddle/v2/fluid/tests/test_log_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_log_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_logical_op.py b/python/paddle/v2/fluid/tests/test_logical_op.py index 8c9e8de739f91a94352f1d9cfa5ce225be929b42..dd67dc561b0e2740d60c3bc21fae0a10ba648c5e 100644 --- a/python/paddle/v2/fluid/tests/test_logical_op.py +++ b/python/paddle/v2/fluid/tests/test_logical_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import op_test import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_lookup_table_op.py b/python/paddle/v2/fluid/tests/test_lookup_table_op.py index 1ff6b305bc9269af6af5377801b90b2839a7c079..0c566c76c91dce8dcfc882eed998f492ae3cde76 100644 --- a/python/paddle/v2/fluid/tests/test_lookup_table_op.py +++ b/python/paddle/v2/fluid/tests/test_lookup_table_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -32,5 +33,19 @@ class TestLookupTableOp(OpTest): self.check_grad(['W'], 'Out', no_grad_set=set('Ids')) +class TestLookupTableOpWithPadding(TestLookupTableOp): + def test_check_output(self): + ids = np.squeeze(self.inputs['Ids']) + padding_idx = np.random.choice(ids, 1)[0] + self.outputs['Out'][ids == padding_idx] = np.zeros(31) + self.attrs = {'padding_idx': long(padding_idx)} + self.check_output() + + def test_check_grad(self): + # Since paddings are not trainable and fixed in forward, the gradient of + # paddings makes no sense and we don't test the gradient here. + pass + + if __name__ == "__main__": unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_lrn_op.py b/python/paddle/v2/fluid/tests/test_lrn_op.py index 051704617e733823e07db7a96d9dff5a8bb95afa..a841dcf79f9abbd9e4a995a395ae9211f62832f2 100644 --- a/python/paddle/v2/fluid/tests/test_lrn_op.py +++ b/python/paddle/v2/fluid/tests/test_lrn_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_lstm_op.py b/python/paddle/v2/fluid/tests/test_lstm_op.py index 76ea8def7cbe66f81b2c548fc2cc9fb9a46c77eb..3e79f9d8e157bc744f14ecfa7c9a6d7de4eae1f9 100644 --- a/python/paddle/v2/fluid/tests/test_lstm_op.py +++ b/python/paddle/v2/fluid/tests/test_lstm_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -41,7 +42,7 @@ def relu(x): return np.maximum(x, 0) -ACTVATION = { +ACTIVATION = { 'identity': identity, 'sigmoid': sigmoid, 'tanh': tanh, @@ -157,8 +158,8 @@ class TestLstmOp(OpTest): w_b = b[:, 0:4 * self.D] w_c = b[:, 4 * self.D:] if self.use_peepholes else None h, c = lstm(x, self.lod, h0, c0, w, w_b, w_c, self.is_reverse, - ACTVATION[self.act_gate], ACTVATION[self.act_cell], - ACTVATION[self.act_cand]) + ACTIVATION[self.act_gate], ACTIVATION[self.act_cell], + ACTIVATION[self.act_cand]) self.inputs = {'Input': (x, self.lod), 'Weight': w} diff --git a/python/paddle/v2/fluid/tests/test_lstm_unit_op.py b/python/paddle/v2/fluid/tests/test_lstm_unit_op.py index c97c1e72aaa1e83f89e24ad5f72d022edef5f7b8..d6348ea0ec9ddf6adb8b48f0d4aeb03dd9b0e895 100644 --- a/python/paddle/v2/fluid/tests/test_lstm_unit_op.py +++ b/python/paddle/v2/fluid/tests/test_lstm_unit_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_lstmp_op.py b/python/paddle/v2/fluid/tests/test_lstmp_op.py new file mode 100644 index 0000000000000000000000000000000000000000..92a954a9aa5574c3016cf9744e1765fff9e9c091 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_lstmp_op.py @@ -0,0 +1,286 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +#Licensed under the Apache License, Version 2.0 (the "License"); +#you may not use this file except in compliance with the License. +#You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +#Unless required by applicable law or agreed to in writing, software +#distributed under the License is distributed on an "AS IS" BASIS, +#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +#See the License for the specific language governing permissions and +#limitations under the License. +import unittest +import numpy as np +import test_lstm_op as LstmTest + +ACTIVATION = { + 'identity': LstmTest.identity, + 'sigmoid': LstmTest.sigmoid, + 'tanh': LstmTest.tanh, + 'relu': LstmTest.relu +} + + +# LSTM with recurrent projection Layer +def lstmp( + input, # T x 4D + lod, # 1 x N + h0=None, # N x D + c0=None, # N x D + w_r=None, # P x 4D + w_rh=None, # D x P + w_b=None, # 1 x 4D + w_c=None, # 1 x 3D + is_reverse=False, + act_gate=None, + act_cell=None, + act_cand=None, + act_proj=None): + def _step(x, w_r, w_rh, w_c, r_pre, c_pre, act_gate, act_cell, act_cand, + act_proj): + g = np.dot(r_pre, w_r) # 1 x 4D + g = g + x + g = np.reshape(g, (1, g.size)) + c, g_i, g_f, g_o = np.split(g, 4, axis=1) + if w_c is None: + g_i = act_gate(g_i) # 1 x D + g_f = act_gate(g_f) # 1 x D + else: + w_ic, w_fc, _ = np.split(w_c, 3, axis=1) + g_i = act_gate(g_i + w_ic * c_pre) # 1 x D + g_f = act_gate(g_f + w_fc * c_pre) # 1 x D + c = g_f * c_pre + g_i * act_cand(c) # 1 x D + + if w_c is None: + g_o = act_gate(g_o) # 1 x D + else: + _, _, w_oc = np.split(w_c, 3, axis=1) + g_o = act_gate(g_o + w_oc * c) # 1 x D + h = g_o * act_cell(c) + # projection + r = np.dot(h, w_rh) + r = act_proj(r) + return r, c + + def _reverse(x, lod): + y = np.zeros_like(x) + for i in range(len(lod) - 1): + b, e = lod[i], lod[i + 1] + y[b:e, :] = np.flip(x[b:e, :], 0) + return y + + offset = lod[0] + batch_size = len(offset) - 1 + # recurrent projection state + projection = [] + cell = [] + input = _reverse(input, offset) if is_reverse else input + if w_b is not None: + input = input + np.tile(w_b, (offset[-1], 1)) + for i in range(batch_size): + # compute one sequence + seq_len = offset[i + 1] - offset[i] + x = input[offset[i]:offset[i + 1], :] + r_pre = np.dot(h0[i], w_rh) # 1 x P + r_pre = act_proj(r_pre) + c_pre = c0[i] # 1 x D + for j in range(seq_len): + # compute one step + r_pre, c_pre = _step(x[j], w_r, w_rh, w_c, r_pre, c_pre, act_gate, + act_cell, act_cand, act_proj) + projection.append(r_pre.flatten()) + cell.append(c_pre.flatten()) + + projection = np.array(projection).astype('float64') + cell = np.array(cell).astype('float64') + + projection = _reverse(projection, offset) if is_reverse else projection + cell = _reverse(cell, offset) if is_reverse else cell + + assert projection.shape == (input.shape[0], w_r.shape[0]) # T x P + assert cell.shape == (input.shape[0], input.shape[1] / 4) # T x D + return projection, cell + + +class TestLstmpOp(LstmTest.TestLstmOp): + def reset_argument(self): + pass + + def setUp(self): + self.set_argument() + # projection size + self.P = 10 + self.act_proj = self.act_cell + + self.reset_argument() + self.op_type = 'lstmp' + + T = self.lod[0][-1] + N = len(self.lod[0]) - 1 + + x = np.random.normal(size=(T, 4 * self.D)).astype('float64') + if self.has_initial_state: + h0 = np.random.normal(size=(N, self.D)).astype('float64') + c0 = np.random.normal(size=(N, self.D)).astype('float64') + else: + h0 = np.zeros((N, self.D)).astype('float64') + c0 = np.zeros((N, self.D)).astype('float64') + w = np.random.normal(size=(self.P, 4 * self.D)).astype('float64') + if self.use_peepholes: + b = np.random.normal(size=(1, 7 * self.D)).astype('float64') + else: + b = np.random.normal(size=(1, 4 * self.D)).astype('float64') + + w_b = b[:, 0:4 * self.D] + w_c = b[:, 4 * self.D:] if self.use_peepholes else None + w_rh = np.random.normal(size=(self.D, self.P)).astype('float64') + r, c = lstmp(x, self.lod, h0, c0, w, w_rh, w_b, w_c, self.is_reverse, + ACTIVATION[self.act_gate], ACTIVATION[self.act_cell], + ACTIVATION[self.act_cand], ACTIVATION[self.act_proj]) + + self.inputs = {'Input': (x, self.lod), 'Weight': w, 'ProjWeight': w_rh} + + self.inputs['Bias'] = b + + if self.has_initial_state: + self.inputs['H0'] = h0 + self.inputs['C0'] = c0 + + self.outputs = { + 'Projection': (r, self.lod), + 'Cell': (c, self.lod), + } + self.attrs = { + 'use_peepholes': self.use_peepholes, + 'is_reverse': self.is_reverse, + 'gate_activation': self.act_gate, + 'cell_activation': self.act_cell, + 'candidate_activation': self.act_cand, + 'proj_activation': self.act_proj + } + + def test_check_output(self): + self.check_output(atol=1e-8) + + def test_check_grad(self): + # TODO(qingqing) remove folowing lines after the check_grad is refined. + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'Weight', 'ProjWeight', 'Bias'], ['Projection'], + max_relative_error=1e-2) + + +class TestLstmpOpHasInitial(TestLstmpOp): + def reset_argument(self): + self.has_initial_state = True + + def test_check_grad(self): + # TODO(qingqing) remove folowing lines after the check_grad is refined. + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'Weight', 'ProjWeight', 'Bias', 'H0', 'C0'], + ['Projection'], + max_relative_error=1e-2) + + def test_check_grad_ingore_bias(self): + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'ProjWeight', 'Weight'], ['Projection'], + max_relative_error=1e-2, + no_grad_set=set('Bias')) + + def test_check_grad_ingore_weight(self): + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'ProjWeight', 'Bias'], ['Projection'], + max_relative_error=1e-2, + no_grad_set=set('Weight')) + + def test_check_grad_ingore_proj_weight(self): + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'Weight', 'Bias'], ['Projection'], + max_relative_error=1e-2, + no_grad_set=set('ProjWeight')) + + def test_check_grad_ingore_input(self): + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Weight', 'ProjWeight', 'Bias'], ['Projection'], + max_relative_error=1e-2, + no_grad_set=set('Input')) + + def test_check_grad_ingore_h0(self): + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'Weight', 'ProjWeight', 'Bias', 'C0'], ['Projection'], + max_relative_error=1e-2, + no_grad_set=set('H0')) + + def test_check_grad_ingore_c0(self): + N = len(self.lod[0]) - 1 + self.outputs['OrderedP0'] = np.zeros((N, self.P)).astype('float64') + self.outputs['BatchGate'] = np.zeros((N, 4 * self.D)).astype('float64') + self.outputs['BatchHidden'] = np.zeros((N, self.D)).astype('float64') + self.outputs['BatchCellPreAct'] = np.zeros( + (N, self.D)).astype('float64') + self.check_grad( + ['Input', 'Weight', 'ProjWeight', 'Bias', 'H0'], ['Projection'], + max_relative_error=1e-2, + no_grad_set=set('C0')) + + +class TestLstmpOpRerverse(TestLstmpOp): + def reset_argument(self): + self.is_reverse = True + + +class TestLstmpOpNotUsePeepholes(TestLstmpOp): + def reset_argument(self): + self.use_peepholes = False + + +class TestLstmpOpLinearProjection(TestLstmpOp): + def reset_argument(self): + self.act_proj = 'identity' + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_margin_rank_loss_op.py b/python/paddle/v2/fluid/tests/test_margin_rank_loss_op.py index 3d8c1d19f90ed4e5945646b85cebb20bb5053f53..694ce20712864effbd2d1e1c66c388b5a3c8ec49 100644 --- a/python/paddle/v2/fluid/tests/test_margin_rank_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_margin_rank_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_math_op_patch.py b/python/paddle/v2/fluid/tests/test_math_op_patch.py new file mode 100644 index 0000000000000000000000000000000000000000..2e77639a4c886327cc8dc7053fc6c0f6c6e9dcc9 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_math_op_patch.py @@ -0,0 +1,181 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import decorators +import paddle.v2.fluid as fluid +import numpy + + +class TestMathOpPatches(unittest.TestCase): + @decorators.prog_scope() + def test_add_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = a + 10 + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(a_np + 10, b_np)) + + @decorators.prog_scope() + def test_radd_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = 10 + a + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(a_np + 10, b_np)) + + @decorators.prog_scope() + def test_sub_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = a - 10 + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(a_np - 10, b_np)) + + @decorators.prog_scope() + def test_radd_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = 10 - a + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(10 - a_np, b_np)) + + @decorators.prog_scope() + def test_mul_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = a * 10 + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(a_np * 10, b_np)) + + @decorators.prog_scope() + def test_rmul_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = 10 * a + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(10 * a_np, b_np)) + + @decorators.prog_scope() + def test_div_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = a / 10 + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(a_np / 10, b_np)) + + @decorators.prog_scope() + def test_rdiv_scalar(self): + a = fluid.layers.data(name="a", shape=[1]) + b = 10 / a + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + 1e-2 + + b_np = exe.run(fluid.default_main_program(), + feed={"a": a_np}, + fetch_list=[b]) + self.assertTrue(numpy.allclose(10 / a_np, b_np)) + + @decorators.prog_scope() + def test_div_two_tensor(self): + a = fluid.layers.data(name="a", shape=[1]) + b = fluid.layers.data(name="b", shape=[1]) + c = a / b + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = numpy.random.random(size=[10, 1]).astype('float32') + 1e-2 + c_np = exe.run(fluid.default_main_program(), + feed={"a": a_np, + 'b': b_np}, + fetch_list=[c]) + self.assertTrue(numpy.allclose(a_np / b_np, c_np)) + + @decorators.prog_scope() + def test_mul_two_tensor(self): + a = fluid.layers.data(name="a", shape=[1]) + b = fluid.layers.data(name="b", shape=[1]) + c = a * b + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = numpy.random.random(size=[10, 1]).astype('float32') + c_np = exe.run(fluid.default_main_program(), + feed={"a": a_np, + 'b': b_np}, + fetch_list=[c]) + self.assertTrue(numpy.allclose(a_np * b_np, c_np)) + + @decorators.prog_scope() + def test_add_two_tensor(self): + a = fluid.layers.data(name="a", shape=[1]) + b = fluid.layers.data(name="b", shape=[1]) + c = a + b + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = numpy.random.random(size=[10, 1]).astype('float32') + c_np = exe.run(fluid.default_main_program(), + feed={"a": a_np, + 'b': b_np}, + fetch_list=[c]) + self.assertTrue(numpy.allclose(a_np + b_np, c_np)) + + @decorators.prog_scope() + def test_sub_two_tensor(self): + a = fluid.layers.data(name="a", shape=[1]) + b = fluid.layers.data(name="b", shape=[1]) + c = a - b + place = fluid.CPUPlace() + exe = fluid.Executor(place) + a_np = numpy.random.random(size=[10, 1]).astype('float32') + b_np = numpy.random.random(size=[10, 1]).astype('float32') + c_np = exe.run(fluid.default_main_program(), + feed={"a": a_np, + 'b': b_np}, + fetch_list=[c]) + self.assertTrue(numpy.allclose(a_np - b_np, c_np)) + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_matmul_op.py b/python/paddle/v2/fluid/tests/test_matmul_op.py index 0220cfe1287ab6908f379f9f7451cd257dd7f0a6..5138af38f4dcc044d31622341b3d7fccc4f6bc90 100644 --- a/python/paddle/v2/fluid/tests/test_matmul_op.py +++ b/python/paddle/v2/fluid/tests/test_matmul_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -59,19 +60,18 @@ def reference_matmul(X, Y, transpose_X=False, transpose_Y=False): X = X.reshape((X.size, 1)) elif X.ndim == 2: X = X.T - elif X.ndim == 3: - X = np.transpose(X, (0, 2, 1)) else: - raise ValueError('X must have between 1 and 3 dimensions') + dim = [i for i in range(len(X.shape))] + dim[-1], dim[len(X.shape) - 2] = dim[len(X.shape) - 2], dim[-1] + X = np.transpose(X, tuple(dim)) if transpose_Y: if Y.ndim == 1: Y = Y.reshape((1, Y.size)) - elif Y.ndim == 2: - Y = Y.T - elif Y.ndim == 3: - Y = np.transpose(Y, (0, 2, 1)) else: - raise ValueError('Y must have between 1 and 3 dimensions') + dim = [i for i in range(len(Y.shape))] + dim[-1], dim[len(Y.shape) - 2] = dim[len(Y.shape) - 2], dim[-1] + Y = np.transpose(Y, tuple(dim)) + Out = np.matmul(X, Y) if not Out.shape: # We do not support 0-dimensional Tensors (scalars). So where @@ -96,18 +96,18 @@ class Generator(object): self.outputs = {'Out': Out} def test_check_output(self): - self.check_output(atol=1e-2) + self.check_output(atol=1e-3) def test_check_grad_normal(self): - self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.5) + self.check_grad(['X', 'Y'], 'Out', max_relative_error=1e-3) def test_check_grad_ignore_x(self): self.check_grad( - ['Y'], 'Out', max_relative_error=0.5, no_grad_set=set("X")) + ['Y'], 'Out', max_relative_error=1e-3, no_grad_set=set("X")) def test_check_grad_ignore_y(self): self.check_grad( - ['X'], 'Out', max_relative_error=0.5, no_grad_set=set('Y')) + ['X'], 'Out', max_relative_error=1e-3, no_grad_set=set('Y')) # Generate test cases for all possibilities @@ -120,13 +120,50 @@ for dim_X in [1, 2, 3]: dim_X, dim_Y, transpose_X, transpose_Y)) shape_X, shape_Y = generate_compatible_shapes( dim_X, dim_Y, transpose_X, transpose_Y) - test_class = type(test_name, (Generator, OpTest), { + globals()[test_name] = type(test_name, (Generator, OpTest), { 'shape_X': shape_X, 'shape_Y': shape_Y, 'transpose_X': transpose_X, 'transpose_Y': transpose_Y, }) - globals()[test_name] = test_class + + +# Test case n-dim +def generate_compatible_shapes(dim, transpose_X, transpose_Y): + M = 2 + N = 4 + K = 3 + shape_X = [2 for _ in range(dim - 2)] + shape_Y = [2 for _ in range(dim - 2)] + + if transpose_X: + shape_X += [K, M] + else: + shape_X += [M, K] + + if transpose_Y: + shape_Y += [N, K] + else: + shape_Y += [K, N] + + return shape_X, shape_Y + + +# Test case n-dim +for dim in [4]: + for transpose_X in [False, True]: + for transpose_Y in [False, True]: + test_name = ( + 'TestMatMulOp_dimX_{}_dim_Y_{}_transX_{}_transY_{}'.format( + dim, dim, transpose_X, transpose_Y)) + shape_X, shape_Y = generate_compatible_shapes(dim, transpose_X, + transpose_Y) + globals()[test_name] = type(test_name, (Generator, OpTest), { + 'shape_X': shape_X, + 'shape_Y': shape_Y, + 'transpose_X': transpose_X, + 'transpose_Y': transpose_Y, + }) if __name__ == "__main__": unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_maxout_op.py b/python/paddle/v2/fluid/tests/test_maxout_op.py index ed8c0d2b67199fb880b4f314a2817a128bf1ad3a..5cd7fbde84a3414829aa51f39283fed0499d39f1 100644 --- a/python/paddle/v2/fluid/tests/test_maxout_op.py +++ b/python/paddle/v2/fluid/tests/test_maxout_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_mean_op.py b/python/paddle/v2/fluid/tests/test_mean_op.py index f9d7d6921e45d10ab2e22c0764c129a8a473726a..81e842163584456b33a0dbd95cfcb405ae857b75 100644 --- a/python/paddle/v2/fluid/tests/test_mean_op.py +++ b/python/paddle/v2/fluid/tests/test_mean_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_memory_optimization_transpiler.py b/python/paddle/v2/fluid/tests/test_memory_optimization_transpiler.py index 76f3c4eb644019b122a93fc12512c3bd39a71bbe..2e9ed78ffd8253d5f2286256097a2238a99d7ba5 100644 --- a/python/paddle/v2/fluid/tests/test_memory_optimization_transpiler.py +++ b/python/paddle/v2/fluid/tests/test_memory_optimization_transpiler.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import unittest diff --git a/python/paddle/v2/fluid/tests/test_minus_op.py b/python/paddle/v2/fluid/tests/test_minus_op.py index 99c0d9056a7172ea40b7db943800a86d4cfbdb46..aee909f56c4353dc8feb7bb82198d5944376f151 100644 --- a/python/paddle/v2/fluid/tests/test_minus_op.py +++ b/python/paddle/v2/fluid/tests/test_minus_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_mnist_if_else_op.py b/python/paddle/v2/fluid/tests/test_mnist_if_else_op.py index 18e3991b94843cda7316b8e23be02475bb669653..3288a0f007c8e593831692eb9a134e78804bdc2e 100644 --- a/python/paddle/v2/fluid/tests/test_mnist_if_else_op.py +++ b/python/paddle/v2/fluid/tests/test_mnist_if_else_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.layers as layers from paddle.v2.fluid.framework import Program, program_guard, default_main_program, default_startup_program from paddle.v2.fluid.executor import Executor diff --git a/python/paddle/v2/fluid/tests/test_modified_huber_loss_op.py b/python/paddle/v2/fluid/tests/test_modified_huber_loss_op.py index 40955283e6a9bfe07613f1a49e1adbc53275436e..eb3873b9ea8b9cbf7b87c125559cc070c39946dc 100644 --- a/python/paddle/v2/fluid/tests/test_modified_huber_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_modified_huber_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_momentum_op.py b/python/paddle/v2/fluid/tests/test_momentum_op.py index 8008a5586f13f97b5671c13bafe8bd8d98e5b3ea..048eaae06ba5323517d5a32174233faaa0fd8be9 100644 --- a/python/paddle/v2/fluid/tests/test_momentum_op.py +++ b/python/paddle/v2/fluid/tests/test_momentum_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_mul_op.py b/python/paddle/v2/fluid/tests/test_mul_op.py index 3033b8ef70d2856805687fa7f09bf51fed01ff5a..83715f0e27b4e4599360356c02831c6814ef520f 100644 --- a/python/paddle/v2/fluid/tests/test_mul_op.py +++ b/python/paddle/v2/fluid/tests/test_mul_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_multihead_attention.py b/python/paddle/v2/fluid/tests/test_multihead_attention.py new file mode 100644 index 0000000000000000000000000000000000000000..a2b300a645fe21931cc12a4e7bb8ebe9b85707c9 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_multihead_attention.py @@ -0,0 +1,98 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import numpy as np + + +class TestMultiheadAttention(unittest.TestCase): + def gen_random_input(self): + """Generate random input data. + """ + # batch_size, max_sequence_length, hidden dimension + self.input_shape = (3, 13, 16) + self.queries = np.random.random(size=self.input_shape).astype("float32") + self.keys = np.random.random(size=self.input_shape).astype("float32") + + def set_program(self): + """Build the test program. + """ + queries = fluid.layers.data( + name="queries", + shape=self.input_shape, + dtype="float32", + append_batch_size=False) + queries.stop_gradient = False + keys = fluid.layers.data( + name="keys", + shape=self.input_shape, + dtype="float32", + append_batch_size=False) + keys.stop_gradient = False + + contexts = fluid.nets.scaled_dot_product_attention( + queries=queries, + keys=keys, + values=keys, + num_heads=8, + dropout_rate=0.) + out = fluid.layers.reduce_sum(contexts, dim=None) + fluid.backward.append_backward(loss=out) + + self.fetch_list = [contexts] + + def run_program(self): + """Run the test program. + """ + places = [core.CPUPlace()] + if core.is_compiled_with_cuda(): + places.append(core.CUDAPlace(0)) + + for place in places: + self.set_inputs(place) + exe = fluid.Executor(place) + + exe.run(fluid.default_startup_program()) + output = exe.run(fluid.default_main_program(), + feed=self.inputs, + fetch_list=self.fetch_list, + return_numpy=True) + self.op_output = output + + def set_inputs(self, place): + """Set the randomly generated data to the test program. + """ + self.inputs = {} + queries = fluid.Tensor() + queries.set(self.queries, place) + + keys = fluid.Tensor() + keys.set(self.keys, place) + + self.inputs["keys"] = keys + self.inputs["queries"] = queries + + def test_multihead_attention(self): + self.gen_random_input() + + self.set_program() + self.run_program() + + #fixme(caoying) add more meaningfull unittest. + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_multiplex_op.py b/python/paddle/v2/fluid/tests/test_multiplex_op.py index 5746ab391e8b4201fc05b7556ad22bbb16fcd8ab..a06aef94a5d9dbee1bcad287c18540e308ce22fa 100644 --- a/python/paddle/v2/fluid/tests/test_multiplex_op.py +++ b/python/paddle/v2/fluid/tests/test_multiplex_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_nce.py b/python/paddle/v2/fluid/tests/test_nce.py index ce66a7c6b355d9b97b7bee853025e5e9cb6e533b..9a51c1f612a0d5363d36e6642ed3b409970025b1 100644 --- a/python/paddle/v2/fluid/tests/test_nce.py +++ b/python/paddle/v2/fluid/tests/test_nce.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -108,4 +109,6 @@ class TestNCECase1(TestNCE): if __name__ == '__main__': + # FIXME(qijun) https://github.com/PaddlePaddle/Paddle/issues/7778 + exit(0) unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_net.py b/python/paddle/v2/fluid/tests/test_net.py index cc78cb4a56de8256f6c5cb41584d875f0946cd12..69d95d4f707d6ff3a66079802a312d0c847f410c 100644 --- a/python/paddle/v2/fluid/tests/test_net.py +++ b/python/paddle/v2/fluid/tests/test_net.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core as core from paddle.v2.fluid.op import Operator import unittest diff --git a/python/paddle/v2/fluid/tests/test_norm_op.py b/python/paddle/v2/fluid/tests/test_norm_op.py index b053522d72bfb95b1d3da960482857f0bf6f6f8d..dd1cd5a31c55f898239297f2815370316a6a8ccf 100644 --- a/python/paddle/v2/fluid/tests/test_norm_op.py +++ b/python/paddle/v2/fluid/tests/test_norm_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_normalization_wrapper.py b/python/paddle/v2/fluid/tests/test_normalization_wrapper.py new file mode 100644 index 0000000000000000000000000000000000000000..6b71f2a923f0cf0744d6b2190aa35830dcf15f24 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_normalization_wrapper.py @@ -0,0 +1,96 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import numpy as np + + +class TestNormalization(unittest.TestCase): + data_desc = {"name": "input", "shape": (2, 3, 7)} + + def gen_random_input(self): + """Generate random input data. + """ + self.data = np.random.random( + size=self.data_desc["shape"]).astype("float32") + + def set_program(self, axis, epsilon): + """Build the test program. + """ + data = fluid.layers.data( + name=self.data_desc["name"], + shape=self.data_desc["shape"], + dtype="float32", + append_batch_size=False) + data.stop_gradient = False + l2_norm = fluid.layers.l2_normalize(x=data, axis=axis, epsilon=epsilon) + out = fluid.layers.reduce_sum(l2_norm, dim=None) + + fluid.backward.append_backward(loss=out) + self.fetch_list = [l2_norm] + + def run_program(self): + """Run the test program. + """ + places = [core.CPUPlace()] + if core.is_compiled_with_cuda(): + places.append(core.CUDAPlace(0)) + + for place in places: + self.set_inputs(place) + exe = fluid.Executor(place) + + output = exe.run(fluid.default_main_program(), + feed=self.inputs, + fetch_list=self.fetch_list, + return_numpy=True) + self.op_output = output + + def set_inputs(self, place): + """Set the randomly generated data to the test program. + """ + self.inputs = {} + tensor = fluid.Tensor() + tensor.set(self.data, place) + self.inputs[self.data_desc["name"]] = tensor + + def l2_normalize(self, data, axis, epsilon): + """ Compute the groundtruth. + """ + output = data * np.reciprocal( + np.sum(np.square(data), axis=axis, keepdims=True)) + return output + + def test_l2_normalize(self): + """ Test the python wrapper for l2_normalize. + """ + axis = 1 + #TODO(caoying) epsilon is not supported due to lack of a maximum_op. + epsilon = 1e-6 + + self.gen_random_input() + + self.set_program(axis, epsilon) + self.run_program() + + expect_output = self.l2_normalize(self.data, axis, epsilon) + + # check output + self.assertTrue(np.allclose(self.op_output, expect_output, atol=0.001)) + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_one_hot_op.py b/python/paddle/v2/fluid/tests/test_one_hot_op.py new file mode 100644 index 0000000000000000000000000000000000000000..e51ea27d14d0637021f8902fa935beb318658018 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_one_hot_op.py @@ -0,0 +1,110 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +import math +from op_test import OpTest +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +import paddle.v2.fluid.framework as framework +from paddle.v2.fluid.framework import Program, program_guard + + +class TestOneHotOp(OpTest): + def setUp(self): + self.op_type = 'one_hot' + depth = 10 + dimension = 12 + x_lod = [[0, 4, 5, 8, 11]] + x = [np.random.randint(0, depth - 1) for i in xrange(x_lod[0][-1])] + x = np.array(x).astype('int').reshape([x_lod[0][-1], 1]) + + out = np.zeros(shape=(np.product(x.shape[:-1]), + depth)).astype('float32') + + for i in xrange(np.product(x.shape)): + out[i, x[i]] = 1.0 + + self.inputs = {'X': (x, x_lod)} + self.attrs = {'depth': depth, 'dtype': int(core.DataType.FP32)} + self.outputs = {'Out': (out, x_lod)} + + def test_check_output(self): + self.check_output() + + +class TestOneHotOp_default_dtype(OpTest): + def setUp(self): + self.op_type = 'one_hot' + depth = 10 + dimension = 12 + x_lod = [[0, 4, 5, 8, 11]] + x = [np.random.randint(0, depth - 1) for i in xrange(x_lod[0][-1])] + x = np.array(x).astype('int').reshape([x_lod[0][-1], 1]) + + out = np.zeros(shape=(np.product(x.shape[:-1]), + depth)).astype('float32') + + for i in xrange(np.product(x.shape)): + out[i, x[i]] = 1.0 + + self.inputs = {'X': (x, x_lod)} + self.attrs = {'depth': depth} + self.outputs = {'Out': (out, x_lod)} + + def test_check_output(self): + self.check_output() + + +class TestOneHotOp_exception(OpTest): + def setUp(self): + self.op_type = 'one_hot' + self.depth = 10 + self.place = core.CPUPlace() + self.dimension = 12 + self.x = core.LoDTensor() + x_lod = [[0, 4, 5, 8, 11]] + data = [np.random.randint(11, 20) for i in xrange(x_lod[0][-1])] + data = np.array(data).astype('int').reshape([x_lod[0][-1], 1]) + self.x.set(data, self.place) + self.x.set_lod(x_lod) + + def test_check_output(self): + program = Program() + with program_guard(program): + x = fluid.layers.data( + name='x', shape=[self.dimension], dtype='float32', lod_level=1) + block = program.current_block() + one_hot_out = block.create_var( + name="one_hot_out", + type=core.VarDesc.VarType.LOD_TENSOR, + dtype='float32') + block.append_op( + type='one_hot', + inputs={'X': x}, + attrs={'depth': self.depth}, + outputs={'Out': one_hot_out}) + exe = fluid.Executor(self.place) + + def run(): + exe.run(feed={'x': self.x}, + fetch_list=[one_hot_out], + return_numpy=False) + + self.assertRaises(core.EnforceNotMet, run) + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_op_support_gpu.py b/python/paddle/v2/fluid/tests/test_op_support_gpu.py index 741686a87465853b26311944e97cd88738d070fd..7de02a8fda22a3db82a2e0b5e6fa9c9f2718fa12 100644 --- a/python/paddle/v2/fluid/tests/test_op_support_gpu.py +++ b/python/paddle/v2/fluid/tests/test_op_support_gpu.py @@ -1,23 +1,25 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core class TestOpSupportGPU(unittest.TestCase): def test_case(self): - self.assertEqual(core.is_compile_gpu(), core.op_support_gpu("sum")) + self.assertEqual(core.is_compiled_with_cuda(), + core.op_support_gpu("sum")) if __name__ == '__main__': diff --git a/python/paddle/v2/fluid/tests/test_operator.py b/python/paddle/v2/fluid/tests/test_operator.py index e75ee41149c6cf4479cd62620647198bc738406d..b82cf580e8567a7d519a75f971bfa0de3ce90684 100644 --- a/python/paddle/v2/fluid/tests/test_operator.py +++ b/python/paddle/v2/fluid/tests/test_operator.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.op as op diff --git a/python/paddle/v2/fluid/tests/test_operator_desc.py b/python/paddle/v2/fluid/tests/test_operator_desc.py index ed18fafe339271ce61b891966162bfc0d4a8c48d..2c8665ffa25549c57eed2934440be676260c1f31 100644 --- a/python/paddle/v2/fluid/tests/test_operator_desc.py +++ b/python/paddle/v2/fluid/tests/test_operator_desc.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_optimizer.py b/python/paddle/v2/fluid/tests/test_optimizer.py index dbec3a59441dd1aa2e87296d9e4edd7e1a0f1306..480ee7091579ba171ca957cb4d25f0034e0534c0 100644 --- a/python/paddle/v2/fluid/tests/test_optimizer.py +++ b/python/paddle/v2/fluid/tests/test_optimizer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.framework as framework diff --git a/python/paddle/v2/fluid/tests/test_pad_op.py b/python/paddle/v2/fluid/tests/test_pad_op.py index 1036b6bcad307a6ce6b4092c84db7f494e8f5811..0bd48000555697be9822d4dfb1056cbc0414aa11 100644 --- a/python/paddle/v2/fluid/tests/test_pad_op.py +++ b/python/paddle/v2/fluid/tests/test_pad_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_parallel_op.py b/python/paddle/v2/fluid/tests/test_parallel_op.py index 3c190477d167c6aa48078869a9abb15488d21dd3..367cc8b1aaf0aff24c685031f33d35becb9eb7ef 100644 --- a/python/paddle/v2/fluid/tests/test_parallel_op.py +++ b/python/paddle/v2/fluid/tests/test_parallel_op.py @@ -1,23 +1,21 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid as fluid import numpy -import sys -# TODO(dzhwinter): get places op check need to be enhanced. -sys.exit(0) class BaseParallelForTest(unittest.TestCase): @@ -55,7 +53,7 @@ class BaseParallelForTest(unittest.TestCase): fetch=fetch, place=cpu, use_parallel=True) - if fluid.core.is_compile_gpu(): + if fluid.core.is_compiled_with_cuda(): gpu = fluid.CUDAPlace(0) result_gpu = self._run_test_impl_( callback=callback, @@ -151,23 +149,52 @@ class BaseParallelForTest(unittest.TestCase): class ParallelOpTest(BaseParallelForTest): + @staticmethod + def __network__(): + x = fluid.layers.data(shape=[784], dtype='float32', name='img') + x = yield x + hidden = fluid.layers.fc(input=x, size=200, param_attr='fc1.w') + loss = fluid.layers.mean(x=hidden) + yield loss + def test_simple_fc(self): - def __network__(): - x = fluid.layers.data(shape=[784], dtype='float32', name='img') - # FIXME: This is a bug of parallel.do - x.stop_gradient = False - x = yield x - hidden = fluid.layers.fc(input=x, size=200, param_attr='fc1.w') - loss = fluid.layers.mean(x=hidden) - yield loss + self.run_test( + callback=self.__network__, + feed={ + 'img': numpy.random.random(size=(51, 784)).astype('float32') + }, + fetch=['fc1.w@GRAD']) + + def test_fc_with_tiny_data(self): + self.run_test( + callback=self.__network__, + feed={'img': numpy.random.random(size=(1, 784)).astype('float32')}, + fetch=['fc1.w@GRAD']) + + +class ParallelOpTestMultipleInput(BaseParallelForTest): + @staticmethod + def __network__(): + x = fluid.layers.data( + shape=[784], dtype='float32', name='img1', stop_gradient=False) + y = fluid.layers.data( + shape=[784], dtype='float32', name='img2', stop_gradient=False) + yield [x, y] + x = x + y + hidden1 = fluid.layers.fc(input=x, size=200, param_attr='fc1.w') + hidden2 = fluid.layers.fc(input=hidden1, size=200, param_attr='fc2.w') + hidden3 = fluid.layers.fc(input=hidden2, size=200, param_attr='fc3.w') + loss = fluid.layers.mean(x=hidden3) + yield loss + def test_simple_fc(self): self.run_test( - callback=__network__, + callback=self.__network__, feed={ - 'img': - numpy.random.random(size=(128 * 3, 784)).astype('float32') + 'img1': numpy.random.random(size=(51, 784)).astype('float32'), + 'img2': numpy.random.random(size=(51, 784)).astype('float32') }, - fetch='fc1.w@GRAD') + fetch=['fc1.w@GRAD', 'fc2.w@GRAD', 'fc3.w@GRAD']) if __name__ == '__main__': diff --git a/python/paddle/v2/fluid/tests/test_parameter.py b/python/paddle/v2/fluid/tests/test_parameter.py index e0db318345061ad30178b69313b30f41c0c62164..dfecdf939bcdb7957b55daea627cdd178b1a3947 100644 --- a/python/paddle/v2/fluid/tests/test_parameter.py +++ b/python/paddle/v2/fluid/tests/test_parameter.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest from paddle.v2.fluid.framework import default_main_program import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_pool2d_op.py b/python/paddle/v2/fluid/tests/test_pool2d_op.py index ac8b24e7ad57b536ba9db360a5499c36b783b133..2f43be8a0ff03731262039cd8eb060d89ae4be40 100644 --- a/python/paddle/v2/fluid/tests/test_pool2d_op.py +++ b/python/paddle/v2/fluid/tests/test_pool2d_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_pool3d_op.py b/python/paddle/v2/fluid/tests/test_pool3d_op.py index 54b8df8465b44df953555bb1e4d299a48ffb5d66..c93711e051b9b12a2314f16fc822e1378bddce47 100644 --- a/python/paddle/v2/fluid/tests/test_pool3d_op.py +++ b/python/paddle/v2/fluid/tests/test_pool3d_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_pool_max_op.py b/python/paddle/v2/fluid/tests/test_pool_max_op.py index c4ec0e50cc9d11d72faccd1a07b3f1843a9bcec3..330ad24bd420401024571725a36191492bfae495 100644 --- a/python/paddle/v2/fluid/tests/test_pool_max_op.py +++ b/python/paddle/v2/fluid/tests/test_pool_max_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_positive_negative_pair_op.py b/python/paddle/v2/fluid/tests/test_positive_negative_pair_op.py index b75f7152efb0fbd56c87ff5b49d1a353aee5af25..9b5e54465559ece98387d1690ed29a582999715e 100644 --- a/python/paddle/v2/fluid/tests/test_positive_negative_pair_op.py +++ b/python/paddle/v2/fluid/tests/test_positive_negative_pair_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import itertools import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_precision_recall_op.py b/python/paddle/v2/fluid/tests/test_precision_recall_op.py index 87c7fcb4b5f06e530959f53b8d247cced0196350..188b7af559376f8535d109c245f08d8e050201d5 100644 --- a/python/paddle/v2/fluid/tests/test_precision_recall_op.py +++ b/python/paddle/v2/fluid/tests/test_precision_recall_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_prelu_op.py b/python/paddle/v2/fluid/tests/test_prelu_op.py index 38bd260bc92e2776b408a8c8b16ac905bb1de537..848036234c8cddd0cb3819322bd338eaf59dd360 100644 --- a/python/paddle/v2/fluid/tests/test_prelu_op.py +++ b/python/paddle/v2/fluid/tests/test_prelu_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_print_op.py b/python/paddle/v2/fluid/tests/test_print_op.py index 4e42863af45353b29d54daf76c9ab3608b217298..3177700dfad6567d667b890d2a58a61c8355c631 100644 --- a/python/paddle/v2/fluid/tests/test_print_op.py +++ b/python/paddle/v2/fluid/tests/test_print_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core from paddle.v2.fluid.executor import Executor diff --git a/python/paddle/v2/fluid/tests/test_prior_box_op.py b/python/paddle/v2/fluid/tests/test_prior_box_op.py new file mode 100644 index 0000000000000000000000000000000000000000..ca8d2bca74ce2d4be8160c8851e393489691ae56 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_prior_box_op.py @@ -0,0 +1,148 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +import sys +import math +from op_test import OpTest + + +class TestPriorBoxOp(OpTest): + def set_data(self): + self.init_test_params() + self.init_test_input() + self.init_test_output() + self.inputs = {'Input': self.input, 'Image': self.image} + + self.attrs = { + 'min_sizes': self.min_sizes, + 'max_sizes': self.max_sizes, + 'aspect_ratios': self.aspect_ratios, + 'variances': self.variances, + 'flip': self.flip, + 'clip': self.clip, + 'step_w': self.step_w, + 'step_h': self.step_h, + 'offset': self.offset + } + + self.outputs = {'Boxes': self.out_boxes, 'Variances': self.out_var} + + def test_check_output(self): + self.check_output() + + def test_check_grad(self): + return + + def setUp(self): + self.op_type = "prior_box" + self.set_data() + + def init_test_params(self): + self.layer_w = 4 + self.layer_h = 4 + + self.image_w = 20 + self.image_h = 20 + + self.step_w = float(self.image_w) / float(self.layer_w) + self.step_h = float(self.image_h) / float(self.layer_h) + + self.input_channels = 2 + self.image_channels = 3 + self.batch_size = 10 + + self.min_sizes = [2, 4] + self.min_sizes = np.array(self.min_sizes).astype('int64') + self.max_sizes = [5, 10] + self.max_sizes = np.array(self.max_sizes).astype('int64') + self.aspect_ratios = [2.0, 3.0] + self.flip = True + self.real_aspect_ratios = [1, 2.0, 1.0 / 2.0, 3.0, 1.0 / 3.0] + self.aspect_ratios = np.array( + self.aspect_ratios, dtype=np.float).flatten() + self.variances = [0.1, 0.1, 0.2, 0.2] + self.variances = np.array(self.variances, dtype=np.float).flatten() + + self.clip = True + + self.num_priors = len(self.real_aspect_ratios) * len(self.min_sizes) + if len(self.max_sizes) > 1: + self.num_priors += len(self.max_sizes) + self.offset = 0.5 + + def init_test_input(self): + self.image = np.random.random( + (self.batch_size, self.image_channels, self.image_w, + self.image_h)).astype('float32') + + self.input = np.random.random( + (self.batch_size, self.input_channels, self.layer_w, + self.layer_h)).astype('float32') + + def init_test_output(self): + out_dim = (self.layer_h, self.layer_w, self.num_priors, 4) + out_boxes = np.zeros(out_dim).astype('float32') + out_var = np.zeros(out_dim).astype('float32') + + idx = 0 + for h in range(self.layer_h): + for w in range(self.layer_w): + c_x = (w + self.offset) * self.step_w + c_y = (h + self.offset) * self.step_h + idx = 0 + for s in range(len(self.min_sizes)): + min_size = self.min_sizes[s] + c_w = c_h = min_size / 2. + out_boxes[h, w, idx, :] = [ + (c_x - c_w) / self.image_w, (c_y - c_h) / self.image_h, + (c_x + c_w) / self.image_w, (c_y + c_h) / self.image_h + ] + idx += 1 + + if len(self.max_sizes) > 0: + max_size = self.max_sizes[s] + # second prior: aspect_ratio = 1, + c_w = c_h = math.sqrt(min_size * max_size) / 2 + out_boxes[h, w, idx, :] = [(c_x - c_w) / self.image_w, + (c_y - c_h) / self.image_h, + (c_x + c_w) / self.image_w, + (c_y + c_h) / self.image_h] + idx += 1 + + # rest of priors + for r in range(len(self.real_aspect_ratios)): + ar = self.real_aspect_ratios[r] + if math.fabs(ar - 1.) < 1e-6: + continue + c_w = min_size * math.sqrt(ar) / 2 + c_h = (min_size / math.sqrt(ar)) / 2 + out_boxes[h, w, idx, :] = [(c_x - c_w) / self.image_w, + (c_y - c_h) / self.image_h, + (c_x + c_w) / self.image_w, + (c_y + c_h) / self.image_h] + idx += 1 + # clip the prior's coordidate such that it is within[0, 1] + if self.clip: + out_boxes = np.clip(out_boxes, 0.0, 1.0) + # set the variance. + out_var = np.tile(self.variances, (self.layer_h, self.layer_w, + self.num_priors, 1)) + self.out_boxes = out_boxes.astype('float32') + self.out_var = out_var.astype('float32') + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_profiler.py b/python/paddle/v2/fluid/tests/test_profiler.py index 4b439a16aa24fa41870b3e156247df4dbf5b5098..09b2d08401878448b4b3f3c6c03193e255e9ffeb 100644 --- a/python/paddle/v2/fluid/tests/test_profiler.py +++ b/python/paddle/v2/fluid/tests/test_profiler.py @@ -1,27 +1,29 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest +import os import numpy as np import paddle.v2.fluid as fluid import paddle.v2.fluid.profiler as profiler import paddle.v2.fluid.layers as layers -import os +import paddle.v2.fluid.core as core class TestProfiler(unittest.TestCase): def test_nvprof(self): - if not fluid.core.is_compile_gpu(): + if not fluid.core.is_compiled_with_cuda(): return epoc = 8 dshape = [4, 3, 28, 28] @@ -39,6 +41,50 @@ class TestProfiler(unittest.TestCase): exe.run(fluid.default_main_program(), feed={'data': input}) os.remove(output_file) + def net_profiler(self, state): + if state == 'GPU' and not core.is_compiled_with_cuda(): + return + startup_program = fluid.Program() + main_program = fluid.Program() + + with fluid.program_guard(main_program, startup_program): + image = fluid.layers.data(name='x', shape=[784], dtype='float32') + hidden1 = fluid.layers.fc(input=image, size=128, act='relu') + hidden2 = fluid.layers.fc(input=hidden1, size=64, act='relu') + predict = fluid.layers.fc(input=hidden2, size=10, act='softmax') + label = fluid.layers.data(name='y', shape=[1], dtype='int64') + cost = fluid.layers.cross_entropy(input=predict, label=label) + avg_cost = fluid.layers.mean(x=cost) + accuracy = fluid.evaluator.Accuracy(input=predict, label=label) + + optimizer = fluid.optimizer.Momentum(learning_rate=0.001, momentum=0.9) + opts = optimizer.minimize(avg_cost, startup_program=startup_program) + + place = fluid.CPUPlace() if state == 'CPU' else fluid.CUDAPlace(0) + exe = fluid.Executor(place) + exe.run(startup_program) + + accuracy.reset(exe) + with profiler.profiler(state, 'total') as prof: + for iter in range(10): + if iter == 2: + profiler.reset_profiler() + x = np.random.random((32, 784)).astype("float32") + y = np.random.randint(0, 10, (32, 1)).astype("int64") + + outs = exe.run(main_program, + feed={'x': x, + 'y': y}, + fetch_list=[avg_cost] + accuracy.metrics) + acc = np.array(outs[1]) + pass_acc = accuracy.eval(exe) + + def test_cpu_profiler(self): + self.net_profiler('CPU') + + def test_cuda_profiler(self): + self.net_profiler('GPU') + if __name__ == '__main__': unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_program.py b/python/paddle/v2/fluid/tests/test_program.py index bcaeede93e43235d4c0288715dc19c7205e6a7c0..9967da15937a1d11adfbaa67b3aa163917a52843 100644 --- a/python/paddle/v2/fluid/tests/test_program.py +++ b/python/paddle/v2/fluid/tests/test_program.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from __future__ import print_function import unittest diff --git a/python/paddle/v2/fluid/tests/test_protobuf.py b/python/paddle/v2/fluid/tests/test_protobuf.py index 5f0646d03603dfe3e37ac23e398812517db9ac17..48e6dedc5862ccf10751d8dd66a8230bdf85eca2 100644 --- a/python/paddle/v2/fluid/tests/test_protobuf.py +++ b/python/paddle/v2/fluid/tests/test_protobuf.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.proto.framework_pb2 as framework_pb2 import unittest diff --git a/python/paddle/v2/fluid/tests/test_protobuf_descs.py b/python/paddle/v2/fluid/tests/test_protobuf_descs.py index 24638dc0e8a74b60fa0a3ed1852093728c0bc79c..9034b2f4ef1c983ef224b14b8f602f87e6ce94b0 100644 --- a/python/paddle/v2/fluid/tests/test_protobuf_descs.py +++ b/python/paddle/v2/fluid/tests/test_protobuf_descs.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_proximal_adagrad_op.py b/python/paddle/v2/fluid/tests/test_proximal_adagrad_op.py index c197d850f978a2e0de67e0252d79c429c66c1a77..744d71bdcf438538e725e7d776b8423edd475639 100644 --- a/python/paddle/v2/fluid/tests/test_proximal_adagrad_op.py +++ b/python/paddle/v2/fluid/tests/test_proximal_adagrad_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_proximal_gd_op.py b/python/paddle/v2/fluid/tests/test_proximal_gd_op.py index 15452558252d09e8a81ab647a64340c30029b505..96540cf6cf4ed970c3e743df7621e443c49eef89 100644 --- a/python/paddle/v2/fluid/tests/test_proximal_gd_op.py +++ b/python/paddle/v2/fluid/tests/test_proximal_gd_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_rank_loss_op.py b/python/paddle/v2/fluid/tests/test_rank_loss_op.py index b4ba7920cd7aaf610075d3dee37f0b7825b387bd..f31a2c2681871f63a272a1a658ed65cc8646b8a8 100644 --- a/python/paddle/v2/fluid/tests/test_rank_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_rank_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_recurrent_op.py b/python/paddle/v2/fluid/tests/test_recurrent_op.py index bcc3457aa3ae6dce161d0d58d6fd965b0cb87f11..6d59e199e24d460791ab6273b20d30bfa733d3c1 100644 --- a/python/paddle/v2/fluid/tests/test_recurrent_op.py +++ b/python/paddle/v2/fluid/tests/test_recurrent_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.layers as layers diff --git a/python/paddle/v2/fluid/tests/test_recv_op.py b/python/paddle/v2/fluid/tests/test_recv_op.py new file mode 100644 index 0000000000000000000000000000000000000000..5c4cec028d354b99d6203281ec4c727d7e3eceac --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_recv_op.py @@ -0,0 +1,68 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest + +import paddle.v2.fluid as fluid +import paddle.v2.fluid.layers as layers +import numpy +from multiprocessing import Process +import os, sys + + +class TestRecvOp(unittest.TestCase): + def test_send(self): + # Run init_serv in a thread + place = fluid.CPUPlace() + p = Process(target=self.init_serv, args=(place, )) + p.daemon = True + p.start() + self.init_client(place) + # FIXME(typhoonzero): find a way to gracefully shutdown the server. + os.system("kill -9 %d" % p.pid) + p.join() + + def init_serv(self, place): + main = fluid.Program() + with fluid.program_guard(main): + x = layers.data( + shape=[32, 32], + dtype='float32', + name="X", + append_batch_size=False) + fluid.initializer.Constant(value=1.0)(x, main.global_block()) + serv = layers.ListenAndServ("127.0.0.1:6174", optimizer_mode=False) + with serv.do(): + o = layers.scale(x=x, scale=10.0) + main.global_block().create_var( + name=o.name, psersistable=False, dtype=o.dtype, shape=o.shape) + exe = fluid.Executor(place) + exe.run(main) + + def init_client(self, place): + main = fluid.Program() + with fluid.program_guard(main): + x = layers.data( + shape=[32, 32], + dtype='float32', + name='X', + append_batch_size=False) + fluid.initializer.Constant(value=1.0)(x, main.global_block()) + layers.Send("127.0.0.1:6174", [x], [x]) + exe = fluid.Executor(place) + exe.run(main) + + +if __name__ == "__main__": + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_reduce_op.py b/python/paddle/v2/fluid/tests/test_reduce_op.py index 57ee307ba66b47bd15864e7be3943b4f5237eb1e..c669f73a7c6de0735b3c580ed4f0ed8ba359a040 100644 --- a/python/paddle/v2/fluid/tests/test_reduce_op.py +++ b/python/paddle/v2/fluid/tests/test_reduce_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -19,7 +20,7 @@ from op_test import OpTest class TestSumOp(OpTest): def setUp(self): self.op_type = "reduce_sum" - self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")} + self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")} self.outputs = {'Out': self.inputs['X'].sum(axis=0)} def test_check_output(self): @@ -32,7 +33,7 @@ class TestSumOp(OpTest): class TestMeanOp(OpTest): def setUp(self): self.op_type = "reduce_mean" - self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float32")} + self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float64")} self.attrs = {'dim': 1} self.outputs = {'Out': self.inputs['X'].mean(axis=self.attrs['dim'])} @@ -48,7 +49,7 @@ class TestMaxOp(OpTest): def setUp(self): self.op_type = "reduce_max" - self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")} + self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")} self.attrs = {'dim': -1} self.outputs = {'Out': self.inputs['X'].max(axis=self.attrs['dim'])} @@ -61,7 +62,7 @@ class TestMinOp(OpTest): def setUp(self): self.op_type = "reduce_min" - self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")} + self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")} self.attrs = {'dim': 2} self.outputs = {'Out': self.inputs['X'].min(axis=self.attrs['dim'])} @@ -72,7 +73,7 @@ class TestMinOp(OpTest): class TestKeepDimReduce(OpTest): def setUp(self): self.op_type = "reduce_sum" - self.inputs = {'X': np.random.random((5, 6, 10)).astype("float32")} + self.inputs = {'X': np.random.random((5, 6, 10)).astype("float64")} self.attrs = {'dim': -2, 'keep_dim': True} self.outputs = { 'Out': self.inputs['X'].sum(axis=self.attrs['dim'], keepdims=True) @@ -88,7 +89,7 @@ class TestKeepDimReduce(OpTest): class Test1DReduce(OpTest): def setUp(self): self.op_type = "reduce_sum" - self.inputs = {'X': np.random.random(20).astype("float32")} + self.inputs = {'X': np.random.random(20).astype("float64")} self.outputs = {'Out': self.inputs['X'].sum(axis=0)} def test_check_output(self): @@ -101,7 +102,7 @@ class Test1DReduce(OpTest): class TestReduceAll(OpTest): def setUp(self): self.op_type = "reduce_sum" - self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float32")} + self.inputs = {'X': np.random.random((5, 6, 2, 10)).astype("float64")} self.attrs = {'reduce_all': True} self.outputs = {'Out': self.inputs['X'].sum()} diff --git a/python/paddle/v2/fluid/tests/test_registry.py b/python/paddle/v2/fluid/tests/test_registry.py index dba11896307ae52877628ab7a51854b80c7a6fec..44e50ca55ac609ed2e0a145ff12248fa18479668 100644 --- a/python/paddle/v2/fluid/tests/test_registry.py +++ b/python/paddle/v2/fluid/tests/test_registry.py @@ -1,35 +1,31 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. import unittest -import warnings import paddle.v2.fluid as fluid -import paddle.v2.fluid.framework as framework -import paddle.v2.fluid.layers as layers -import paddle.v2.fluid.registry as registry +import numpy as np +import decorators class TestRegistry(unittest.TestCase): + @decorators.prog_scope() def test_registry_layer(self): - self.layer_type = "mean" - program = framework.Program() - x = fluid.layers.data(name='X', shape=[10, 10], dtype='float32') - output = layers.mean(x) + output = fluid.layers.mean(x=x) + place = fluid.CPUPlace() exe = fluid.Executor(place) - X = np.random.random((10, 10)).astype("float32") - mean_out = exe.run(program, feed={"X": X}, fetch_list=[output]) - self.assertAlmostEqual(np.mean(X), mean_out) + mean_out = exe.run(feed={"X": X}, fetch_list=[output]) + self.assertAlmostEqual(np.mean(X), mean_out[0]) diff --git a/python/paddle/v2/fluid/tests/test_regularizer.py b/python/paddle/v2/fluid/tests/test_regularizer.py index 9eaae1904a01ca7994ad493c3b37352b3d50bab6..b33817fa41636e7a62aa5907e63a9302b2149f66 100644 --- a/python/paddle/v2/fluid/tests/test_regularizer.py +++ b/python/paddle/v2/fluid/tests/test_regularizer.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.framework as framework diff --git a/python/paddle/v2/fluid/tests/test_reorder_lod_tensor.py b/python/paddle/v2/fluid/tests/test_reorder_lod_tensor.py index 0bcdfafcf4496a4f47c69afb7acd24427cbb634c..0a223bac0ce8fd626881cef983c7cd960f2c5ba8 100644 --- a/python/paddle/v2/fluid/tests/test_reorder_lod_tensor.py +++ b/python/paddle/v2/fluid/tests/test_reorder_lod_tensor.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid as fluid import paddle.v2.fluid.core as core @@ -44,7 +45,7 @@ class TestReorderLoDTensor(unittest.TestCase): outputs = [] input_grads = [] places = [core.CPUPlace()] - if core.is_compile_gpu(): + if core.is_compiled_with_cuda(): places.append(core.CUDAPlace(0)) for place in places: self.set_inputs(place) diff --git a/python/paddle/v2/fluid/tests/test_reshape_op.py b/python/paddle/v2/fluid/tests/test_reshape_op.py index d6e6797043dc07cdc62f41a3d6dac43fc25934d0..2cc0b36460994b021c6e6fd944c80dfec6c92f2f 100644 --- a/python/paddle/v2/fluid/tests/test_reshape_op.py +++ b/python/paddle/v2/fluid/tests/test_reshape_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_rmsprop_op.py b/python/paddle/v2/fluid/tests/test_rmsprop_op.py index 27a1ea213714271fb373e270add039bc4667e6fd..b6d7c698009d1fc35a5cdbcb72708ee74c181bf4 100644 --- a/python/paddle/v2/fluid/tests/test_rmsprop_op.py +++ b/python/paddle/v2/fluid/tests/test_rmsprop_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_rnn_memory_helper_op.py b/python/paddle/v2/fluid/tests/test_rnn_memory_helper_op.py index 378d7f852304e4cc41ed760663457e470985afbc..82b54bbd1a46be763c436f075b28e466f7fbd3fa 100644 --- a/python/paddle/v2/fluid/tests/test_rnn_memory_helper_op.py +++ b/python/paddle/v2/fluid/tests/test_rnn_memory_helper_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest from paddle.v2.fluid.framework import Program diff --git a/python/paddle/v2/fluid/tests/test_roi_pool_op.py b/python/paddle/v2/fluid/tests/test_roi_pool_op.py index 6d7a698b09e288696caf9ad8460f8df32c11b009..af48848dcd06d35aa404d4cb688d7fd55c6b048b 100644 --- a/python/paddle/v2/fluid/tests/test_roi_pool_op.py +++ b/python/paddle/v2/fluid/tests/test_roi_pool_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import math diff --git a/python/paddle/v2/fluid/tests/test_row_conv_op.py b/python/paddle/v2/fluid/tests/test_row_conv_op.py index 1234d289cb2f021669b227237be9b0032f4ce935..580b08f75ebbdf0f79cc45becf84520503e9877c 100644 --- a/python/paddle/v2/fluid/tests/test_row_conv_op.py +++ b/python/paddle/v2/fluid/tests/test_row_conv_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_scale_op.py b/python/paddle/v2/fluid/tests/test_scale_op.py index 9847d3d36198c1078958e06ad87590f812e4eaa8..95cd935dda32bd5e1e8afc1e09c83affe8004cf7 100644 --- a/python/paddle/v2/fluid/tests/test_scale_op.py +++ b/python/paddle/v2/fluid/tests/test_scale_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_scatter_op.py b/python/paddle/v2/fluid/tests/test_scatter_op.py index b6c4162f6f47c5eb5b8d6a0308fc80baeb37e14c..f2936e19ae52b1b3c6d7daf02f68789af12d2b23 100644 --- a/python/paddle/v2/fluid/tests/test_scatter_op.py +++ b/python/paddle/v2/fluid/tests/test_scatter_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_scope.py b/python/paddle/v2/fluid/tests/test_scope.py index adaaf1690627e20a9e8d8b21796d061cb6fdc2dc..566a11abbe28c7fa79f07253609c1152879a0ad2 100644 --- a/python/paddle/v2/fluid/tests/test_scope.py +++ b/python/paddle/v2/fluid/tests/test_scope.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core import unittest diff --git a/python/paddle/v2/fluid/tests/test_selected_rows.py b/python/paddle/v2/fluid/tests/test_selected_rows.py index 3179a3caaecfc17a71f15f41d69cb82369ed5998..65ddf1f8f5f565ab5f361b8ef00b0032d62a58f8 100644 --- a/python/paddle/v2/fluid/tests/test_selected_rows.py +++ b/python/paddle/v2/fluid/tests/test_selected_rows.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core as core import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_seq_concat_op.py b/python/paddle/v2/fluid/tests/test_seq_concat_op.py index 1f026fd76e83a270a7b20815b6ad1f397c062bd7..ba2bb075e6d76ec4e4d74291582512f1223b8813 100644 --- a/python/paddle/v2/fluid/tests/test_seq_concat_op.py +++ b/python/paddle/v2/fluid/tests/test_seq_concat_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import sys diff --git a/python/paddle/v2/fluid/tests/test_seq_conv.py b/python/paddle/v2/fluid/tests/test_seq_conv.py index c7e508519446fc69758d0dbc91b1ea9bacb9f11b..674a2e16940d1c8d40fcd6abdb33c25f3c2fd845 100644 --- a/python/paddle/v2/fluid/tests/test_seq_conv.py +++ b/python/paddle/v2/fluid/tests/test_seq_conv.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import random diff --git a/python/paddle/v2/fluid/tests/test_seq_pool.py b/python/paddle/v2/fluid/tests/test_seq_pool.py index bb15495373fb083b1c7dc031c7286dceef7e4ecf..9dd6b2a0872f64a33df3aae1cd78bc021c1f44e7 100644 --- a/python/paddle/v2/fluid/tests/test_seq_pool.py +++ b/python/paddle/v2/fluid/tests/test_seq_pool.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_sequence_erase_op.py b/python/paddle/v2/fluid/tests/test_sequence_erase_op.py index 650984009a76a56fb65811f7bb805ca656194a35..4823836ba97d50dcc455ee0b558fdf602aa9c5e5 100644 --- a/python/paddle/v2/fluid/tests/test_sequence_erase_op.py +++ b/python/paddle/v2/fluid/tests/test_sequence_erase_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest @@ -29,7 +30,7 @@ def sequence_erase(in_seq, lod0, tokens): return np.array(out_seq).astype("int32"), new_lod0 -class TestSequenceEraseOp(OpTest): +class TestSequenceEraseOpInt32(OpTest): def setUp(self): self.op_type = "sequence_erase" in_seq = np.random.randint(0, 10, (30, 1)).astype("int32") @@ -44,5 +45,35 @@ class TestSequenceEraseOp(OpTest): self.check_output() +class TestSequenceEraseOpInt64(OpTest): + def setUp(self): + self.op_type = "sequence_erase" + in_seq = np.random.randint(0, 10, (30, 1)).astype("int64") + lod = [[0, 9, 13, 24, 30]] + tokens = [2, 3, 5] + out_seq, new_lod0 = sequence_erase(in_seq, lod[0], tokens) + self.attrs = {'tokens': tokens} + self.inputs = {'X': (in_seq, lod)} + self.outputs = {'Out': (out_seq, [new_lod0])} + + def test_check_output(self): + self.check_output() + + +class TestSequenceEraseOpEmpty(OpTest): + def setUp(self): + self.op_type = "sequence_erase" + in_seq = np.random.randint(0, 10, (30, 1)).astype("int32") + lod = [[0, 9, 13, 24, 30]] + tokens = [] + out_seq, new_lod0 = sequence_erase(in_seq, lod[0], tokens) + self.attrs = {'tokens': tokens} + self.inputs = {'X': (in_seq, lod)} + self.outputs = {'Out': (out_seq, [new_lod0])} + + def test_check_output(self): + self.check_output() + + if __name__ == '__main__': unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_sequence_expand.py b/python/paddle/v2/fluid/tests/test_sequence_expand.py index aacdabf295dc1c26c62db5ebfa6961df9fed5816..6fc045125fabf6192abc362b6b3d85623308b325 100644 --- a/python/paddle/v2/fluid/tests/test_sequence_expand.py +++ b/python/paddle/v2/fluid/tests/test_sequence_expand.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_sequence_reshape.py b/python/paddle/v2/fluid/tests/test_sequence_reshape.py new file mode 100644 index 0000000000000000000000000000000000000000..06d5af8f5e7f5383561245bbbd57ecf5f65ceec4 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_sequence_reshape.py @@ -0,0 +1,85 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy as np +import math +from op_test import OpTest + + +class TestSequenceReshape(OpTest): + def setUp(self): + self.op_type = 'sequence_reshape' + dimension = 12 + x_lod = [[0, 4, 5, 8, 11]] + x = np.random.uniform(0.1, 1, [11, 24]).astype('float32') + + self.inputs = {'X': (x, x_lod)} + self.attrs = {'new_dim': dimension} + + out, out_lod = self.compute_output(x, x_lod, dimension) + + self.outputs = {'Out': (out, out_lod)} + + def compute_output(self, x, x_lod, dimension): + x_width = x.shape[1] + out_lod = [[0]] + for i in xrange(len(x_lod[0]) - 1): + seq_len = x_lod[0][i + 1] - x_lod[0][i] + offset = (seq_len * x_width) / dimension + assert int(offset) * dimension == seq_len * x_width + out_lod[0].append(out_lod[0][-1] + int(offset)) + out = np.zeros(shape=(out_lod[0][-1], dimension)).astype('float32') + out.ravel()[:] = x.ravel()[:] + return out, out_lod + + def test_check_output(self): + self.check_output() + + def test_check_grad(self): + self.check_grad(["X"], "Out") + + +class TestSequenceReshape_reduce(TestSequenceReshape): + def setUp(self): + self.op_type = 'sequence_reshape' + dimension = 24 + x_lod = [[0, 4, 6, 8, 12]] + x = np.random.uniform(0.1, 1, [12, 12]).astype('float32') + + self.inputs = {'X': (x, x_lod)} + self.attrs = {'new_dim': dimension} + + out, out_lod = self.compute_output(x, x_lod, dimension) + + self.outputs = {'Out': (out, out_lod)} + + +class TestSequenceReshape_same(TestSequenceReshape): + def setUp(self): + self.op_type = 'sequence_reshape' + dimension = 12 + x_lod = [[0, 4, 6, 8, 12]] + x = np.random.uniform(0.1, 1, [12, 12]).astype('float32') + + self.inputs = {'X': (x, x_lod)} + self.attrs = {'new_dim': dimension} + + out, out_lod = self.compute_output(x, x_lod, dimension) + + self.outputs = {'Out': (out, out_lod)} + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_sequence_slice_op.py b/python/paddle/v2/fluid/tests/test_sequence_slice_op.py index 94062431f0b4297535c20c4b86a11a8058c8876e..bf1f21bcde4e293c30c1863bb84725632a147101 100644 --- a/python/paddle/v2/fluid/tests/test_sequence_slice_op.py +++ b/python/paddle/v2/fluid/tests/test_sequence_slice_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import sys diff --git a/python/paddle/v2/fluid/tests/test_sequence_softmax_op.py b/python/paddle/v2/fluid/tests/test_sequence_softmax_op.py index 8170e4d7f18d0fe28d8f927f122139369ee672fe..5bd780f6b5b5d0a9deda8701625d33fbe7e87abf 100644 --- a/python/paddle/v2/fluid/tests/test_sequence_softmax_op.py +++ b/python/paddle/v2/fluid/tests/test_sequence_softmax_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_sgd_op.py b/python/paddle/v2/fluid/tests/test_sgd_op.py index 4a71fb30a9c7a185be8150a476a6bb317d19e3eb..ba2ca1683f9f6d72bbd1550df89c7424d223a1d9 100644 --- a/python/paddle/v2/fluid/tests/test_sgd_op.py +++ b/python/paddle/v2/fluid/tests/test_sgd_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np import paddle.v2.fluid.core as core @@ -90,7 +91,7 @@ class TestSparseSGDOp(unittest.TestCase): def test_sparse_sgd(self): places = [core.CPUPlace()] - if core.is_compile_gpu(): + if core.is_compiled_with_cuda(): places.append(core.CUDAPlace(0)) for place in places: self.check_with_place(place) diff --git a/python/paddle/v2/fluid/tests/test_shrink_rnn_memory.py b/python/paddle/v2/fluid/tests/test_shrink_rnn_memory.py index 1825a5258fa1ab08c405d20e3e77b4d92a200e7f..4578211bac63fdf365f5e2d0de7181126c6fdaed 100644 --- a/python/paddle/v2/fluid/tests/test_shrink_rnn_memory.py +++ b/python/paddle/v2/fluid/tests/test_shrink_rnn_memory.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core from paddle.v2.fluid.executor import Executor diff --git a/python/paddle/v2/fluid/tests/test_sigmoid_cross_entropy_with_logits_op.py b/python/paddle/v2/fluid/tests/test_sigmoid_cross_entropy_with_logits_op.py index 132502c9cba5af71b2477cded1a9bc63ed842a56..f88fa62119cfc60df832ec0e14ef11930b8bcf21 100644 --- a/python/paddle/v2/fluid/tests/test_sigmoid_cross_entropy_with_logits_op.py +++ b/python/paddle/v2/fluid/tests/test_sigmoid_cross_entropy_with_logits_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np from op_test import OpTest from scipy.special import logit diff --git a/python/paddle/v2/fluid/tests/test_sign_op.py b/python/paddle/v2/fluid/tests/test_sign_op.py index f649cb9e7cd9938c697a8c36a88d5b39507e4269..c1dfa7f45d4d22795bef97f1f937bdbcdb5f5a30 100644 --- a/python/paddle/v2/fluid/tests/test_sign_op.py +++ b/python/paddle/v2/fluid/tests/test_sign_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_smooth_l1_loss_op.py b/python/paddle/v2/fluid/tests/test_smooth_l1_loss_op.py index 1052eaa8b0ec281208344175a9978e25528f4e5d..5a388bb7b37c2509f4af290479195dd20fdc63a0 100644 --- a/python/paddle/v2/fluid/tests/test_smooth_l1_loss_op.py +++ b/python/paddle/v2/fluid/tests/test_smooth_l1_loss_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_softmax_op.py b/python/paddle/v2/fluid/tests/test_softmax_op.py index d03e50b2f1edb43ada22b0357fc136c5edccd6fe..cf43e676c5421ce9fca07fe36fb6e347585d6e03 100644 --- a/python/paddle/v2/fluid/tests/test_softmax_op.py +++ b/python/paddle/v2/fluid/tests/test_softmax_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_softmax_with_cross_entropy_op.py b/python/paddle/v2/fluid/tests/test_softmax_with_cross_entropy_op.py index 330467081b41d980599dff7804e84a8bc000912a..626f34f0e07c372d6262e030c9686a73b63cb97d 100644 --- a/python/paddle/v2/fluid/tests/test_softmax_with_cross_entropy_op.py +++ b/python/paddle/v2/fluid/tests/test_softmax_with_cross_entropy_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_split_and_merge_lod_tensor_op.py b/python/paddle/v2/fluid/tests/test_split_and_merge_lod_tensor_op.py index 4e90404eca49213c1fe2a5a35a912cd048ce7af8..bc541298ed80fd8ae9db86aec384a8522a994749 100644 --- a/python/paddle/v2/fluid/tests/test_split_and_merge_lod_tensor_op.py +++ b/python/paddle/v2/fluid/tests/test_split_and_merge_lod_tensor_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.core as core import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_split_op.py b/python/paddle/v2/fluid/tests/test_split_op.py index 000c300446f7a026477eba2e854b94d20969b5a3..b80b64c41be4a45e9d3725297a526e93b399f00d 100644 --- a/python/paddle/v2/fluid/tests/test_split_op.py +++ b/python/paddle/v2/fluid/tests/test_split_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_split_selected_rows_op.py b/python/paddle/v2/fluid/tests/test_split_selected_rows_op.py new file mode 100644 index 0000000000000000000000000000000000000000..343aa20066146ae08462a92f1efaa20c4d4b5ed8 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_split_selected_rows_op.py @@ -0,0 +1,130 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import paddle.v2.fluid.core as core +import numpy as np +from paddle.v2.fluid.op import Operator + + +class TestSpliteSelectedRows(unittest.TestCase): + def get_places(self): + places = [core.CPUPlace()] + if core.is_compiled_with_cuda(): + places.append(core.CUDAPlace(0)) + return places + + def test_check_output(self): + for place in self.get_places(): + self.check_with_place(place) + + def test_check_grad(self): + for place in self.get_places(): + self.check_grad_with_place(place) + + def check_with_place(self, place): + scope = core.Scope() + rows = [0, 5, 7, 4, 20] + height = 20 + row_numel = 2 + + # initialize input variable X + x = scope.var('X').get_selected_rows() + x.set_rows(rows) + x.set_height(height) + np_array = np.ones((len(rows), row_numel)).astype("float32") + np_array[0, 0] = 2.0 + np_array[2, 1] = 4.0 + np_array[4, 1] = 8.0 + x_tensor = x.get_tensor() + x_tensor.set(np_array, place) + + height_sections = [5, 5, 5, 5, 3] + + # initialize output variables [out0, out1] + outs_name = ["out%d" % i for i in xrange(len(height_sections))] + outs = [ + scope.var(var_name).get_selected_rows() for var_name in outs_name + ] + + # expected output selected rows + expected_out0_rows = [0, 4] + expected_out1_rows = [5, 7] + expected_out4_rows = [20] + + op = Operator( + "split_selected_rows", + X="X", + Out=outs_name, + height_sections=height_sections) + + op.run(scope, place) + + self.assertEqual(outs[0].rows(), expected_out0_rows) + self.assertEqual(outs[1].rows(), expected_out1_rows) + self.assertEqual(outs[4].rows(), expected_out4_rows) + + self.assertEqual(outs[0].height(), height_sections[0]) + self.assertEqual(outs[4].height(), height_sections[4]) + + self.assertAlmostEqual(2.0, np.array(outs[0].get_tensor())[0, 0]) + self.assertAlmostEqual(4.0, np.array(outs[1].get_tensor())[1, 1]) + self.assertAlmostEqual(8.0, np.array(outs[4].get_tensor())[0, 1]) + + def check_grad_with_place(self, place): + scope = core.Scope() + height = 10 + row_numel = 2 + + # attr + height_sections = [5, 5] + + # initialize input variable X + out0_grad = scope.var("out0@GRAD").get_selected_rows() + rows0 = [0, 5] + out0_grad.set_rows(rows0) + out0_grad.set_height(height) + out0_grad_tensor = out0_grad.get_tensor() + np_array = np.ones((len(rows0), row_numel)).astype("float32") + np_array[0, 0] = 2.0 + out0_grad_tensor.set(np_array, place) + + out1_grad = scope.var("out1@GRAD").get_selected_rows() + rows1 = [7, 5] + out1_grad.set_rows(rows1) + out1_grad.set_height(height) + out1_grad_tensor = out1_grad.get_tensor() + np_array = np.ones((len(rows1), row_numel)).astype("float32") + np_array[0, 1] = 4.0 + out1_grad_tensor.set(np_array, place) + + x_grad = scope.var("X@GRAD").get_selected_rows() + + grad_op = Operator( + "sum", + X=["out0@GRAD", "out1@GRAD"], + Out="X@GRAD", + height_sections=height_sections) + + grad_op.run(scope, place) + + self.assertEqual(x_grad.rows(), rows0 + rows1) + self.assertEqual(x_grad.height(), height) + + self.assertAlmostEqual(2.0, np.array(x_grad.get_tensor())[0, 0]) + self.assertAlmostEqual(4.0, np.array(x_grad.get_tensor())[2, 1]) + + +if __name__ == "__main__": + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_spp_op.py b/python/paddle/v2/fluid/tests/test_spp_op.py index f09bb94b449f2868575d081490027efc472b5b95..e912b56de54c91a4c716e8056a6cf891de74bedf 100644 --- a/python/paddle/v2/fluid/tests/test_spp_op.py +++ b/python/paddle/v2/fluid/tests/test_spp_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_squared_l2_distance_op.py b/python/paddle/v2/fluid/tests/test_squared_l2_distance_op.py index 7b80d81d728197df7a089e35356e847234a8d3c9..8171207cd95f6b5104b6c0da3c7424016c9569b6 100644 --- a/python/paddle/v2/fluid/tests/test_squared_l2_distance_op.py +++ b/python/paddle/v2/fluid/tests/test_squared_l2_distance_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_squared_l2_norm_op.py b/python/paddle/v2/fluid/tests/test_squared_l2_norm_op.py index 80994f5937ec09d754f8782128bb36214321a3c7..b7575cb4d2cea1801cc194dd4b4dcda1aea2f545 100644 --- a/python/paddle/v2/fluid/tests/test_squared_l2_norm_op.py +++ b/python/paddle/v2/fluid/tests/test_squared_l2_norm_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy as np import unittest from numpy import linalg as LA diff --git a/python/paddle/v2/fluid/tests/test_sum_op.py b/python/paddle/v2/fluid/tests/test_sum_op.py index 366708ac839643949e36cebe29392f6d4b8d5e6a..0a15a9485d72629de6e2e8c7d76c8a5b2cdf0a14 100644 --- a/python/paddle/v2/fluid/tests/test_sum_op.py +++ b/python/paddle/v2/fluid/tests/test_sum_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_tensor.py b/python/paddle/v2/fluid/tests/test_tensor.py index 62a48b206cd3fb1adefe8a015cf3e2df17a779aa..0219bef42b3ba133dda7412c1036cf989a170a36 100644 --- a/python/paddle/v2/fluid/tests/test_tensor.py +++ b/python/paddle/v2/fluid/tests/test_tensor.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import paddle.v2.fluid.core as core import unittest import numpy @@ -107,9 +108,31 @@ class TestTensor(unittest.TestCase): scope = core.Scope() place = core.CPUPlace() lod_py = [[0, 2, 5], [0, 2, 4, 5]] - lod_tensor = core.LoDTensor(lod_py) + lod_tensor = core.LoDTensor() + + lod_tensor.set_dims([5, 2, 3, 4]) + lod_tensor.set_lod(lod_py) + lod_tensor.alloc_float(place) + tensor_array = numpy.array(lod_tensor) + tensor_array[0, 0, 0, 0] = 1.0 + tensor_array[0, 0, 0, 1] = 2.0 + lod_tensor.set(tensor_array, place) + + lod_v = numpy.array(lod_tensor) + self.assertAlmostEqual(1.0, lod_v[0, 0, 0, 0]) + self.assertAlmostEqual(2.0, lod_v[0, 0, 0, 1]) + self.assertListEqual(lod_py, lod_tensor.lod()) + + def test_lod_tensor_gpu_init(self): + if not core.is_compiled_with_cuda(): + return + scope = core.Scope() + place = core.CUDAPlace(0) + lod_py = [[0, 2, 5], [0, 2, 4, 5]] + lod_tensor = core.LoDTensor() lod_tensor.set_dims([5, 2, 3, 4]) + lod_tensor.set_lod(lod_py) lod_tensor.alloc_float(place) tensor_array = numpy.array(lod_tensor) tensor_array[0, 0, 0, 0] = 1.0 diff --git a/python/paddle/v2/fluid/tests/test_top_k_op.py b/python/paddle/v2/fluid/tests/test_top_k_op.py index 86968dba140c4e0b6e94b9bf78255bcbb753434d..a50faf0fffdcffab8b67e968450573d676d2d40a 100644 --- a/python/paddle/v2/fluid/tests/test_top_k_op.py +++ b/python/paddle/v2/fluid/tests/test_top_k_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_transpose_op.py b/python/paddle/v2/fluid/tests/test_transpose_op.py index ff2541f450ca8ce374404aef65f1caa2d5f4f00b..a16de1416f4d8d1c6cf51c3c4b57dbe97e528b45 100644 --- a/python/paddle/v2/fluid/tests/test_transpose_op.py +++ b/python/paddle/v2/fluid/tests/test_transpose_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_uniform_random_op.py b/python/paddle/v2/fluid/tests/test_uniform_random_op.py index 332ac4f07ff936569c43aaab4a698f2b8e3fb985..94cf416fad8f02cdea8017ae1350fa264ce644b1 100644 --- a/python/paddle/v2/fluid/tests/test_uniform_random_op.py +++ b/python/paddle/v2/fluid/tests/test_uniform_random_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy @@ -35,7 +36,7 @@ class TestUniformRandomOp(unittest.TestCase): self.uniform_random_test(place=core.CPUPlace()) def test_gpu(self): - if core.is_compile_gpu(): + if core.is_compiled_with_cuda(): self.uniform_random_test(place=core.CUDAPlace(0)) def uniform_random_test(self, place): diff --git a/python/paddle/v2/fluid/tests/test_unpool_op.py b/python/paddle/v2/fluid/tests/test_unpool_op.py index 988c0c75063f42f8865388979e14a00db3733400..3dd43f9ba4b0efe852ba1a7d509c6f81576b2206 100644 --- a/python/paddle/v2/fluid/tests/test_unpool_op.py +++ b/python/paddle/v2/fluid/tests/test_unpool_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import numpy as np from op_test import OpTest diff --git a/python/paddle/v2/fluid/tests/test_variable.py b/python/paddle/v2/fluid/tests/test_variable.py index 199fd4a8c26fc2ebcd50655fd2ad131b08383071..9f9748ca4e4778bf963360fb13661d3469e344bf 100644 --- a/python/paddle/v2/fluid/tests/test_variable.py +++ b/python/paddle/v2/fluid/tests/test_variable.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest from paddle.v2.fluid.framework import default_main_program, Program, convert_np_dtype_to_dtype_ import paddle.v2.fluid.core as core diff --git a/python/paddle/v2/fluid/tests/test_warpctc_op.py b/python/paddle/v2/fluid/tests/test_warpctc_op.py index 9f565676c5af1685704681758a8590ecb6f59026..55d1c73262a4d846ae2280c1967eb00f4f6c4dd7 100644 --- a/python/paddle/v2/fluid/tests/test_warpctc_op.py +++ b/python/paddle/v2/fluid/tests/test_warpctc_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import sys import unittest import numpy as np diff --git a/python/paddle/v2/fluid/tests/test_weight_normalization.py b/python/paddle/v2/fluid/tests/test_weight_normalization.py new file mode 100644 index 0000000000000000000000000000000000000000..80ad8285d8a3c2ced814cc3588a814c14ec60855 --- /dev/null +++ b/python/paddle/v2/fluid/tests/test_weight_normalization.py @@ -0,0 +1,121 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +import unittest +import numpy +import collections +import paddle.v2.fluid as fluid +import paddle.v2.fluid.core as core +from paddle.v2.fluid.initializer import ConstantInitializer +from paddle.v2.fluid.param_attr import WeightNormParamAttr + + +class TestWeightNormalization(unittest.TestCase): + batch_size = 3 + hidden_size = 5 + data_desc = (['x', [10], 0], ) + + @classmethod + def setUpClass(cls): + cls.set_program() + + @classmethod + def set_program(cls): + data = fluid.layers.data( + name=cls.data_desc[0][0], shape=cls.data_desc[0][1]) + out = fluid.layers.fc(input=data, + size=cls.hidden_size, + param_attr=WeightNormParamAttr( + dim=None, + name='weight_norm_param', + initializer=ConstantInitializer(1.0)), + bias_attr=False, + act=None) + loss = fluid.layers.reduce_sum(out) + fluid.backward.append_backward(loss=loss) + cls.fetch_list = [ + 'weight_norm_param_g', 'weight_norm_param_v', + 'weight_norm_param_g@GRAD' + ] + + def run_program(self): + outputs = [] + places = [core.CPUPlace()] + if core.is_compiled_with_cuda(): + places.append(core.CUDAPlace(0)) + for place in places: + self.set_inputs(place) + exe = fluid.Executor(place) + exe.run(fluid.default_startup_program()) + output = exe.run(fluid.default_main_program(), + feed=self.inputs, + fetch_list=self.fetch_list, + return_numpy=False) + outputs.append(output) + self.actual_outputs = outputs + + def set_data(self): + self.data = collections.OrderedDict() + for desc in self.data_desc: + data_name = desc[0] + data_shape = desc[1] + data_lod_level = desc[2] + data_lod = [] + for i in range(data_lod_level): + lod_level_i = numpy.random.randint( + low=1, + high=5, + size=self.batch_size if i == 0 else lod_level_i[-1]) + lod_level_i = [0] + numpy.cumsum(lod_level_i).tolist() + data_lod.append(lod_level_i) + data_value = numpy.random.random( + size=[data_lod[-1][-1] if data_lod else self.batch_size + ] + data_shape).astype('float32') + self.data[data_name] = (data_value, data_lod) + + def set_inputs(self, place): + self.inputs = {} + for desc in self.data_desc: + tensor = fluid.Tensor() + tensor.set(self.data[desc[0]][0], place) + if self.data[desc[0]][1]: + tensor.set_lod(self.data[desc[0]][1]) + self.inputs[desc[0]] = tensor + + def weight_normalize(self): + v = numpy.ones((self.data[self.data_desc[0][0]][0].shape[-1], + self.hidden_size)) + g = numpy.linalg.norm(v, axis=None, keepdims=True) + w = g * v / numpy.linalg.norm(v, axis=None, keepdims=True) + x = self.data[self.data_desc[0][0]][0] + out = numpy.dot(x, w) + g_grad = (numpy.dot(x.T, numpy.ones_like(out)) * (v / numpy.linalg.norm( + v, axis=None, keepdims=True))).sum(axis=None, keepdims=True) + return g, v, g_grad + + def test_weight_normalization(self): + self.set_data() + self.run_program() + expect_output = self.weight_normalize() + for actual_output in self.actual_outputs: + [ + self.assertTrue( + numpy.allclose( + numpy.array(actual), expect, atol=0.001)) + for expect, actual in zip(expect_output, actual_output) + ] + + +if __name__ == '__main__': + unittest.main() diff --git a/python/paddle/v2/fluid/tests/test_while_op.py b/python/paddle/v2/fluid/tests/test_while_op.py index 72de0a03612053ba3a07526726a65606275dc61c..9f5e1b668c0d01768138487d38f755d2832a8551 100644 --- a/python/paddle/v2/fluid/tests/test_while_op.py +++ b/python/paddle/v2/fluid/tests/test_while_op.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import paddle.v2.fluid.layers as layers from paddle.v2.fluid.executor import Executor diff --git a/python/paddle/v2/image.py b/python/paddle/v2/image.py index a6fa0cecb87e86e804815012885678a9fc557d95..e5000e440cc8d822dbd38dce3978d2722d32ebe4 100644 --- a/python/paddle/v2/image.py +++ b/python/paddle/v2/image.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. """ This file contains some common interfaces for image preprocess. Many users are confused about the image layout. We introduce @@ -176,7 +176,6 @@ def resize_short(im, size): :param size: the shorter edge size of image after resizing. :type size: int """ - assert im.shape[-1] == 1 or im.shape[-1] == 3 h, w = im.shape[:2] h_new, w_new = size, size if h > w: @@ -267,7 +266,7 @@ def random_crop(im, size, is_color=True): return im -def left_right_flip(im): +def left_right_flip(im, is_color=True): """ Flip an image along the horizontal direction. Return the flipped image. @@ -278,13 +277,15 @@ def left_right_flip(im): im = left_right_flip(im) - :paam im: input image with HWC layout + :param im: input image with HWC layout or HW layout for gray image :type im: ndarray + :param is_color: whether input image is color or not + :type is_color: bool """ - if len(im.shape) == 3: + if len(im.shape) == 3 and is_color: return im[:, ::-1, :] else: - return im[:, ::-1, :] + return im[:, ::-1] def simple_transform(im, @@ -319,11 +320,12 @@ def simple_transform(im, """ im = resize_short(im, resize_size) if is_train: - im = random_crop(im, crop_size) + im = random_crop(im, crop_size, is_color=is_color) if np.random.randint(2) == 0: - im = left_right_flip(im) + im = left_right_flip(im, is_color) else: - im = center_crop(im, crop_size) + im = center_crop(im, crop_size, is_color) + im = center_crop(im, crop_size, is_color=is_color) if len(im.shape) == 3: im = to_chw(im) @@ -331,8 +333,10 @@ def simple_transform(im, if mean is not None: mean = np.array(mean, dtype=np.float32) # mean value, may be one value per channel - if mean.ndim == 1: + if mean.ndim == 1 and is_color: mean = mean[:, np.newaxis, np.newaxis] + elif mean.ndim == 1: + mean = mean else: # elementwise mean assert len(mean.shape) == len(im) @@ -372,6 +376,6 @@ def load_and_transform(filename, mean values per channel. :type mean: numpy array | list """ - im = load_image(filename) + im = load_image(filename, is_color) im = simple_transform(im, resize_size, crop_size, is_train, is_color, mean) return im diff --git a/python/paddle/v2/inference.py b/python/paddle/v2/inference.py index 39d1bfff0c8659bb87b9b97334f377639cea9c59..78bf9807da3586fd2899cd914cd64c824814e7f3 100644 --- a/python/paddle/v2/inference.py +++ b/python/paddle/v2/inference.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import numpy import collections import topology diff --git a/python/paddle/v2/master/__init__.py b/python/paddle/v2/master/__init__.py index 09daaaa75e01969ded25dcc848df3f7b9202124e..494e4baf20e6d1c5171aa54c93a30c67c02bb02a 100644 --- a/python/paddle/v2/master/__init__.py +++ b/python/paddle/v2/master/__init__.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + from client import * __all__ = ['client'] diff --git a/python/paddle/v2/master/client.py b/python/paddle/v2/master/client.py index b874c2f349094c0c0ab9e3663fcc7491f1edd236..b3c790e39d7fa9b6595a92b6df94474b9e111cb0 100644 --- a/python/paddle/v2/master/client.py +++ b/python/paddle/v2/master/client.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import ctypes import os diff --git a/python/paddle/v2/reader/tests/__init__.py b/python/paddle/v2/reader/tests/__init__.py index 2619c1c0e9db17c38ccc6e1dd010bd9c1c5966bd..b94a21a7e406b833797f8f521c62a2351c2bc30a 100644 --- a/python/paddle/v2/reader/tests/__init__.py +++ b/python/paddle/v2/reader/tests/__init__.py @@ -1,13 +1,13 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. diff --git a/python/paddle/v2/reader/tests/creator_test.py b/python/paddle/v2/reader/tests/creator_test.py index cf190aa6645f9a5bed891a3a47c03efa03813d65..ac6cd4e9b6e19efb8cddbede0ff91438e5dbf531 100644 --- a/python/paddle/v2/reader/tests/creator_test.py +++ b/python/paddle/v2/reader/tests/creator_test.py @@ -1,3 +1,17 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + # Copyright PaddlePaddle contributors. All Rights Reserved # # Licensed under the Apache License, Version 2.0 (the "License"); diff --git a/python/paddle/v2/reader/tests/decorator_test.py b/python/paddle/v2/reader/tests/decorator_test.py index 4ba71969dffe7447b6c5b70aeb752a4e5469fb36..e41e9c78a070feab85f3d221c3f1a96b83e6dccf 100644 --- a/python/paddle/v2/reader/tests/decorator_test.py +++ b/python/paddle/v2/reader/tests/decorator_test.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import time import unittest diff --git a/python/paddle/v2/tests/test_image.py b/python/paddle/v2/tests/test_image.py index b2d773510de28ca2614e95b465c73b82aa7b0463..2b0444bb03fcdc051d0d2324c728584f01ca7ba4 100644 --- a/python/paddle/v2/tests/test_image.py +++ b/python/paddle/v2/tests/test_image.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import unittest import numpy as np diff --git a/python/paddle/v2/tests/test_layer.py b/python/paddle/v2/tests/test_layer.py index de932ad715bea8db158393c3c192ef67502e2fa3..710e8135f2de2549e653e952533924e582d938d0 100644 --- a/python/paddle/v2/tests/test_layer.py +++ b/python/paddle/v2/tests/test_layer.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import unittest import paddle.v2.activation as activation diff --git a/python/paddle/v2/tests/test_op.py b/python/paddle/v2/tests/test_op.py index 69acccddf42bb22ab54e0cf9e2a5eaef34e47b50..dd04cc4ab6698f7f99209a49a67733c75789f1d9 100644 --- a/python/paddle/v2/tests/test_op.py +++ b/python/paddle/v2/tests/test_op.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import unittest import paddle.v2.data_type as data_type diff --git a/python/paddle/v2/tests/test_paramconf_order.py b/python/paddle/v2/tests/test_paramconf_order.py index 41fea64122b81948d57cce07f00d764e4889da66..33c240b8f5a54a679289c71167551f71288d2305 100644 --- a/python/paddle/v2/tests/test_paramconf_order.py +++ b/python/paddle/v2/tests/test_paramconf_order.py @@ -1,3 +1,17 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + # Copyright PaddlePaddle contributors. All Rights Reserved # # Licensed under the Apache License, Version 2.0 (the "License"); diff --git a/python/paddle/v2/tests/test_parameters.py b/python/paddle/v2/tests/test_parameters.py index ab6863620feacd4db7f83eec976811ab0097f5e7..1fe1f09b9d38b35d0b36088f577d175342a9d9b1 100644 --- a/python/paddle/v2/tests/test_parameters.py +++ b/python/paddle/v2/tests/test_parameters.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + import unittest import sys diff --git a/python/paddle/v2/tests/test_rnn_layer.py b/python/paddle/v2/tests/test_rnn_layer.py index 192b0ee678bcee752327b8c4d41fba29ea361bb6..7920e342e1406e6b7eba5037558a87050b4ee49d 100644 --- a/python/paddle/v2/tests/test_rnn_layer.py +++ b/python/paddle/v2/tests/test_rnn_layer.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import difflib import unittest diff --git a/python/paddle/v2/tests/test_topology.py b/python/paddle/v2/tests/test_topology.py index 7fd2ee82fde21d90be541a28f23742e51a9a1665..11b4154eedc629d703a90ece3768f2e26e981665 100644 --- a/python/paddle/v2/tests/test_topology.py +++ b/python/paddle/v2/tests/test_topology.py @@ -1,4 +1,4 @@ -# Copyright PaddlePaddle contributors. All Rights Reserved +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -11,6 +11,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. + import unittest import paddle.v2.layer as layer import paddle.v2.topology as topology diff --git a/python/paddle/v2/trainer.py b/python/paddle/v2/trainer.py index 1a70a7203b7e358ce654621d686319179b511249..a0060bf227f01389093a106acb5f087a39969e74 100644 --- a/python/paddle/v2/trainer.py +++ b/python/paddle/v2/trainer.py @@ -1,16 +1,16 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. """ Module Trainer """ diff --git a/tools/manylinux1/Dockerfile.x64 b/tools/manylinux1/Dockerfile.x64 index 2c6ba650a5d7996bef212e88a16f2a159ca377e7..0f1b8331309248aaaf0ed32cf14c583a4cdb7437 100644 --- a/tools/manylinux1/Dockerfile.x64 +++ b/tools/manylinux1/Dockerfile.x64 @@ -35,7 +35,7 @@ RUN cd /opt && wget -q --no-check-certificate https://github.com/google/protobuf cd protobuf-3.1.0 && ./configure && make -j4 && make install && cd .. && rm -f protobuf-cpp-3.1.0.tar.gz -RUN yum install -y sqlite-devel zlib-devel openssl-devel boost boost-devel pcre-devel vim tk-devel tkinter libtool +RUN yum install -y sqlite-devel zlib-devel openssl-devel pcre-devel vim tk-devel tkinter libtool RUN wget -O /root/requirements.txt https://raw.githubusercontent.com/PaddlePaddle/Paddle/develop/python/requirements.txt diff --git a/tools/manylinux1/build_scripts/manylinux1-check.py b/tools/manylinux1/build_scripts/manylinux1-check.py index e4bde065a293c9d8ea8c5b150246766328138fd4..a27eab1c77c3b8e2be02d5bd492e5c2514b6d3b1 100644 --- a/tools/manylinux1/build_scripts/manylinux1-check.py +++ b/tools/manylinux1/build_scripts/manylinux1-check.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + # Logic copied from PEP 513 diff --git a/tools/manylinux1/build_scripts/python-tag-abi-tag.py b/tools/manylinux1/build_scripts/python-tag-abi-tag.py index 301fbf07a47fef03c91d9dd5f49c2894a5971319..cd2573314c5b250e00584152f693a70682c17f39 100644 --- a/tools/manylinux1/build_scripts/python-tag-abi-tag.py +++ b/tools/manylinux1/build_scripts/python-tag-abi-tag.py @@ -1,3 +1,17 @@ +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + # Utility script to print the python tag + the abi tag for a Python # See PEP 425 for exactly what these are, but an example would be: # cp27-cp27mu diff --git a/tools/manylinux1/build_scripts/ssl-check.py b/tools/manylinux1/build_scripts/ssl-check.py index 900185cef14c51bca6a929801a86728b7ffc0b4a..34a3116207f9e69d436227fce9961e099e22d34c 100644 --- a/tools/manylinux1/build_scripts/ssl-check.py +++ b/tools/manylinux1/build_scripts/ssl-check.py @@ -1,16 +1,17 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. +# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. # -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at # -# http://www.apache.org/licenses/LICENSE-2.0 +# http://www.apache.org/licenses/LICENSE-2.0 # -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + # cf. https://github.com/pypa/manylinux/issues/53 GOOD_SSL = "https://google.com" diff --git a/v1_api_demo/README.md b/v1_api_demo/README.md deleted file mode 100644 index 0460a85fae078800332982751a5d4a9644c50bd6..0000000000000000000000000000000000000000 --- a/v1_api_demo/README.md +++ /dev/null @@ -1,5 +0,0 @@ -The examples in v1_api_demo are using v1_api currently, and will be upgraded to v2_api later. -Thus, v1_api_demo is a temporary directory. We decide not to maintain it and will delete it in future. - -Please go to [PaddlePaddle/book](https://github.com/PaddlePaddle/book) and -[PaddlePaddle/models](https://github.com/PaddlePaddle/models) to learn PaddlePaddle. diff --git a/v1_api_demo/gan/.gitignore b/v1_api_demo/gan/.gitignore deleted file mode 100644 index 93a6f5080a16a601cffb0bff51af9aef3ba3bae7..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/.gitignore +++ /dev/null @@ -1,11 +0,0 @@ -output/ -uniform_params/ -cifar_params/ -mnist_params/ -*.png -.pydevproject -.project -*.log -*.pyc -data/mnist_data/ -data/cifar-10-batches-py/ diff --git a/v1_api_demo/gan/README.md b/v1_api_demo/gan/README.md deleted file mode 100644 index 1908b534b0c1f63904d5503399b961d74ce0037c..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/README.md +++ /dev/null @@ -1,13 +0,0 @@ -# Generative Adversarial Networks (GAN) - -This demo implements GAN training described in the original GAN paper (https://arxiv.org/abs/1406.2661) and DCGAN (https://arxiv.org/abs/1511.06434). - -The general training procedures are implemented in gan_trainer.py. The neural network configurations are specified in gan_conf.py (for synthetic data) and gan_conf_image.py (for image data). - -In order to run the model, first download the corresponding data by running the shell script in ./data. -Then you can run the command below. The flag -d specifies the training data (cifar, mnist or uniform) and flag --useGpu specifies whether to use gpu for training (0 is cpu, 1 is gpu). - -$python gan_trainer.py -d cifar --use_gpu 1 - -The generated images will be stored in ./cifar_samples/ -The corresponding models will be stored in ./cifar_params/ diff --git a/v1_api_demo/gan/data/download_cifar.sh b/v1_api_demo/gan/data/download_cifar.sh deleted file mode 100755 index bbadc7c10c73e45a0948018b8812f79040d14bc4..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/data/download_cifar.sh +++ /dev/null @@ -1,18 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e -wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz -tar zxf cifar-10-python.tar.gz -rm cifar-10-python.tar.gz diff --git a/v1_api_demo/gan/data/get_mnist_data.sh b/v1_api_demo/gan/data/get_mnist_data.sh deleted file mode 100755 index a77c81bf5af9ddb6634ff89460797ca543c5e517..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/data/get_mnist_data.sh +++ /dev/null @@ -1,17 +0,0 @@ -#!/usr/bin/env sh -# This script downloads the mnist data and unzips it. -set -e -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -rm -rf "$DIR/mnist_data" -mkdir "$DIR/mnist_data" -cd "$DIR/mnist_data" - -echo "Downloading..." - -for fname in train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte -do - if [ ! -e $fname ]; then - wget --no-check-certificate http://yann.lecun.com/exdb/mnist/${fname}.gz - gunzip ${fname}.gz - fi -done diff --git a/v1_api_demo/gan/gan_conf.py b/v1_api_demo/gan/gan_conf.py deleted file mode 100644 index 86ac2dffe5f4490a88e12d1fa5e8cd9fa61a69f4..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/gan_conf.py +++ /dev/null @@ -1,151 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -from paddle.trainer_config_helpers import * - -mode = get_config_arg("mode", str, "generator") -assert mode in set([ - "generator", "discriminator", "generator_training", "discriminator_training" -]) - -is_generator_training = mode == "generator_training" -is_discriminator_training = mode == "discriminator_training" -is_generator = mode == "generator" -is_discriminator = mode == "discriminator" - -# The network structure below follows the ref https://arxiv.org/abs/1406.2661 -# Here we used two hidden layers and batch_norm - -print('mode=%s' % mode) -# the dim of the noise (z) as the input of the generator network -noise_dim = 10 -# the dim of the hidden layer -hidden_dim = 10 -# the dim of the generated sample -sample_dim = 2 - -settings( - batch_size=128, - learning_rate=1e-4, - learning_method=AdamOptimizer(beta1=0.5)) - - -def discriminator(sample): - """ - discriminator ouputs the probablity of a sample is from generator - or real data. - The output has two dimenstional: dimension 0 is the probablity - of the sample is from generator and dimension 1 is the probabblity - of the sample is from real data. - """ - param_attr = ParamAttr(is_static=is_generator_training) - bias_attr = ParamAttr( - is_static=is_generator_training, initial_mean=1.0, initial_std=0) - - hidden = fc_layer( - input=sample, - name="dis_hidden", - size=hidden_dim, - bias_attr=bias_attr, - param_attr=param_attr, - act=ReluActivation()) - - hidden2 = fc_layer( - input=hidden, - name="dis_hidden2", - size=hidden_dim, - bias_attr=bias_attr, - param_attr=param_attr, - act=LinearActivation()) - - hidden_bn = batch_norm_layer( - hidden2, - act=ReluActivation(), - name="dis_hidden_bn", - bias_attr=bias_attr, - param_attr=ParamAttr( - is_static=is_generator_training, initial_mean=1.0, - initial_std=0.02), - use_global_stats=False) - - return fc_layer( - input=hidden_bn, - name="dis_prob", - size=2, - bias_attr=bias_attr, - param_attr=param_attr, - act=SoftmaxActivation()) - - -def generator(noise): - """ - generator generates a sample given noise - """ - param_attr = ParamAttr(is_static=is_discriminator_training) - bias_attr = ParamAttr( - is_static=is_discriminator_training, initial_mean=1.0, initial_std=0) - - hidden = fc_layer( - input=noise, - name="gen_layer_hidden", - size=hidden_dim, - bias_attr=bias_attr, - param_attr=param_attr, - act=ReluActivation()) - - hidden2 = fc_layer( - input=hidden, - name="gen_hidden2", - size=hidden_dim, - bias_attr=bias_attr, - param_attr=param_attr, - act=LinearActivation()) - - hidden_bn = batch_norm_layer( - hidden2, - act=ReluActivation(), - name="gen_layer_hidden_bn", - bias_attr=bias_attr, - param_attr=ParamAttr( - is_static=is_discriminator_training, - initial_mean=1.0, - initial_std=0.02), - use_global_stats=False) - - return fc_layer( - input=hidden_bn, - name="gen_layer1", - size=sample_dim, - bias_attr=bias_attr, - param_attr=param_attr, - act=LinearActivation()) - - -if is_generator_training: - noise = data_layer(name="noise", size=noise_dim) - sample = generator(noise) - -if is_discriminator_training: - sample = data_layer(name="sample", size=sample_dim) - -if is_generator_training or is_discriminator_training: - label = data_layer(name="label", size=1) - prob = discriminator(sample) - cost = cross_entropy(input=prob, label=label) - classification_error_evaluator( - input=prob, label=label, name=mode + '_error') - outputs(cost) - -if is_generator: - noise = data_layer(name="noise", size=noise_dim) - outputs(generator(noise)) diff --git a/v1_api_demo/gan/gan_conf_image.py b/v1_api_demo/gan/gan_conf_image.py deleted file mode 100644 index c469227994c1a84d1aa73e03bbc74ebeac41d30e..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/gan_conf_image.py +++ /dev/null @@ -1,298 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -from paddle.trainer_config_helpers import * - -mode = get_config_arg("mode", str, "generator") -dataSource = get_config_arg("data", str, "mnist") -assert mode in set([ - "generator", "discriminator", "generator_training", "discriminator_training" -]) - -is_generator_training = mode == "generator_training" -is_discriminator_training = mode == "discriminator_training" -is_generator = mode == "generator" -is_discriminator = mode == "discriminator" - -# The network structure below follows the dcgan paper -# (https://arxiv.org/abs/1511.06434) - -print('mode=%s' % mode) -# the dim of the noise (z) as the input of the generator network -noise_dim = 100 -# the number of filters in the layer in generator/discriminator that is -# closet to the image -gf_dim = 64 -df_dim = 64 -if dataSource == "mnist": - sample_dim = 28 # image dim - c_dim = 1 # image color -else: - sample_dim = 32 - c_dim = 3 -s2, s4 = int(sample_dim / 2), int(sample_dim / 4), -s8, s16 = int(sample_dim / 8), int(sample_dim / 16) - -settings( - batch_size=128, - learning_rate=2e-4, - learning_method=AdamOptimizer(beta1=0.5)) - - -def conv_bn(input, - channels, - imgSize, - num_filters, - output_x, - stride, - name, - param_attr, - bias_attr, - param_attr_bn, - bn, - trans=False, - act=ReluActivation()): - """ - conv_bn is a utility function that constructs a convolution/deconv layer - with an optional batch_norm layer - - :param bn: whether to use batch_norm_layer - :type bn: bool - :param trans: whether to use conv (False) or deconv (True) - :type trans: bool - """ - - # calculate the filter_size and padding size based on the given - # imgSize and ouput size - tmp = imgSize - (output_x - 1) * stride - if tmp <= 1 or tmp > 5: - raise ValueError("conv input-output dimension does not fit") - elif tmp <= 3: - filter_size = tmp + 2 - padding = 1 - else: - filter_size = tmp - padding = 0 - - print(imgSize, output_x, stride, filter_size, padding) - - if trans: - nameApx = "_convt" - else: - nameApx = "_conv" - - if bn: - conv = img_conv_layer( - input, - filter_size=filter_size, - num_filters=num_filters, - name=name + nameApx, - num_channels=channels, - act=LinearActivation(), - groups=1, - stride=stride, - padding=padding, - bias_attr=bias_attr, - param_attr=param_attr, - shared_biases=True, - layer_attr=None, - filter_size_y=None, - stride_y=None, - padding_y=None, - trans=trans) - - conv_bn = batch_norm_layer( - conv, - act=act, - name=name + nameApx + "_bn", - bias_attr=bias_attr, - param_attr=param_attr_bn, - use_global_stats=False) - - return conv_bn - else: - conv = img_conv_layer( - input, - filter_size=filter_size, - num_filters=num_filters, - name=name + nameApx, - num_channels=channels, - act=act, - groups=1, - stride=stride, - padding=padding, - bias_attr=bias_attr, - param_attr=param_attr, - shared_biases=True, - layer_attr=None, - filter_size_y=None, - stride_y=None, - padding_y=None, - trans=trans) - return conv - - -def generator(noise): - """ - generator generates a sample given noise - """ - param_attr = ParamAttr( - is_static=is_discriminator_training, initial_mean=0.0, initial_std=0.02) - bias_attr = ParamAttr( - is_static=is_discriminator_training, initial_mean=0.0, initial_std=0.0) - - param_attr_bn = ParamAttr( - is_static=is_discriminator_training, initial_mean=1.0, initial_std=0.02) - - h1 = fc_layer( - input=noise, - name="gen_layer_h1", - size=s8 * s8 * gf_dim * 4, - bias_attr=bias_attr, - param_attr=param_attr, - act=LinearActivation()) - - h1_bn = batch_norm_layer( - h1, - act=ReluActivation(), - name="gen_layer_h1_bn", - bias_attr=bias_attr, - param_attr=param_attr_bn, - use_global_stats=False) - - h2_bn = conv_bn( - h1_bn, - channels=gf_dim * 4, - output_x=s8, - num_filters=gf_dim * 2, - imgSize=s4, - stride=2, - name="gen_layer_h2", - param_attr=param_attr, - bias_attr=bias_attr, - param_attr_bn=param_attr_bn, - bn=True, - trans=True) - - h3_bn = conv_bn( - h2_bn, - channels=gf_dim * 2, - output_x=s4, - num_filters=gf_dim, - imgSize=s2, - stride=2, - name="gen_layer_h3", - param_attr=param_attr, - bias_attr=bias_attr, - param_attr_bn=param_attr_bn, - bn=True, - trans=True) - - return conv_bn( - h3_bn, - channels=gf_dim, - output_x=s2, - num_filters=c_dim, - imgSize=sample_dim, - stride=2, - name="gen_layer_h4", - param_attr=param_attr, - bias_attr=bias_attr, - param_attr_bn=param_attr_bn, - bn=False, - trans=True, - act=TanhActivation()) - - -def discriminator(sample): - """ - discriminator ouputs the probablity of a sample is from generator - or real data. - The output has two dimenstional: dimension 0 is the probablity - of the sample is from generator and dimension 1 is the probabblity - of the sample is from real data. - """ - param_attr = ParamAttr( - is_static=is_generator_training, initial_mean=0.0, initial_std=0.02) - bias_attr = ParamAttr( - is_static=is_generator_training, initial_mean=0.0, initial_std=0.0) - - param_attr_bn = ParamAttr( - is_static=is_generator_training, initial_mean=1.0, initial_std=0.02) - - h0 = conv_bn( - sample, - channels=c_dim, - imgSize=sample_dim, - num_filters=df_dim, - output_x=s2, - stride=2, - name="dis_h0", - param_attr=param_attr, - bias_attr=bias_attr, - param_attr_bn=param_attr_bn, - bn=False) - - h1_bn = conv_bn( - h0, - channels=df_dim, - imgSize=s2, - num_filters=df_dim * 2, - output_x=s4, - stride=2, - name="dis_h1", - param_attr=param_attr, - bias_attr=bias_attr, - param_attr_bn=param_attr_bn, - bn=True) - - h2_bn = conv_bn( - h1_bn, - channels=df_dim * 2, - imgSize=s4, - num_filters=df_dim * 4, - output_x=s8, - stride=2, - name="dis_h2", - param_attr=param_attr, - bias_attr=bias_attr, - param_attr_bn=param_attr_bn, - bn=True) - - return fc_layer( - input=h2_bn, - name="dis_prob", - size=2, - bias_attr=bias_attr, - param_attr=param_attr, - act=SoftmaxActivation()) - - -if is_generator_training: - noise = data_layer(name="noise", size=noise_dim) - sample = generator(noise) - -if is_discriminator_training: - sample = data_layer(name="sample", size=sample_dim * sample_dim * c_dim) - -if is_generator_training or is_discriminator_training: - label = data_layer(name="label", size=1) - prob = discriminator(sample) - cost = cross_entropy(input=prob, label=label) - classification_error_evaluator( - input=prob, label=label, name=mode + '_error') - outputs(cost) - -if is_generator: - noise = data_layer(name="noise", size=noise_dim) - outputs(generator(noise)) diff --git a/v1_api_demo/gan/gan_trainer.py b/v1_api_demo/gan/gan_trainer.py deleted file mode 100644 index 4a26c230f7a21cc6dd4a3cdb52e32730b1ce73ca..0000000000000000000000000000000000000000 --- a/v1_api_demo/gan/gan_trainer.py +++ /dev/null @@ -1,349 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import argparse -import random -import numpy -import cPickle -import sys, os -from PIL import Image - -from paddle.trainer.config_parser import parse_config -from paddle.trainer.config_parser import logger -import py_paddle.swig_paddle as api -import matplotlib.pyplot as plt - - -def plot2DScatter(data, outputfile): - ''' - Plot the data as a 2D scatter plot and save to outputfile - data needs to be two dimensinoal - ''' - x = data[:, 0] - y = data[:, 1] - logger.info("The mean vector is %s" % numpy.mean(data, 0)) - logger.info("The std vector is %s" % numpy.std(data, 0)) - - heatmap, xedges, yedges = numpy.histogram2d(x, y, bins=50) - extent = [xedges[0], xedges[-1], yedges[0], yedges[-1]] - - plt.clf() - plt.scatter(x, y) - plt.savefig(outputfile, bbox_inches='tight') - - -def CHECK_EQ(a, b): - assert a == b, "a=%s, b=%s" % (a, b) - - -def copy_shared_parameters(src, dst): - ''' - copy the parameters from src to dst - :param src: the source of the parameters - :type src: GradientMachine - :param dst: the destination of the parameters - :type dst: GradientMachine - ''' - src_params = [src.getParameter(i) for i in xrange(src.getParameterSize())] - src_params = dict([(p.getName(), p) for p in src_params]) - - for i in xrange(dst.getParameterSize()): - dst_param = dst.getParameter(i) - src_param = src_params.get(dst_param.getName(), None) - if src_param is None: - continue - src_value = src_param.getBuf(api.PARAMETER_VALUE) - dst_value = dst_param.getBuf(api.PARAMETER_VALUE) - CHECK_EQ(len(src_value), len(dst_value)) - dst_value.copyFrom(src_value) - dst_param.setValueUpdated() - - -def print_parameters(src): - src_params = [src.getParameter(i) for i in xrange(src.getParameterSize())] - - print "***************" - for p in src_params: - print "Name is %s" % p.getName() - print "value is %s \n" % p.getBuf(api.PARAMETER_VALUE).copyToNumpyArray( - ) - - -def load_mnist_data(imageFile): - f = open(imageFile, "rb") - f.read(16) - - # Define number of samples for train/test - if "train" in imageFile: - n = 60000 - else: - n = 10000 - - data = numpy.fromfile(f, 'ubyte', count=n * 28 * 28).reshape((n, 28 * 28)) - data = data / 255.0 * 2.0 - 1.0 - - f.close() - return data.astype('float32') - - -def load_cifar_data(cifar_path): - batch_size = 10000 - data = numpy.zeros((5 * batch_size, 32 * 32 * 3), dtype="float32") - for i in range(1, 6): - file = cifar_path + "/data_batch_" + str(i) - fo = open(file, 'rb') - dict = cPickle.load(fo) - fo.close() - data[(i - 1) * batch_size:(i * batch_size), :] = dict["data"] - - data = data / 255.0 * 2.0 - 1.0 - return data - - -# synthesize 2-D uniform data -def load_uniform_data(): - data = numpy.random.rand(1000000, 2).astype('float32') - return data - - -def merge(images, size): - if images.shape[1] == 28 * 28: - h, w, c = 28, 28, 1 - else: - h, w, c = 32, 32, 3 - img = numpy.zeros((h * size[0], w * size[1], c)) - for idx in xrange(size[0] * size[1]): - i = idx % size[1] - j = idx // size[1] - img[j*h:j*h+h, i*w:i*w+w, :] = \ - ((images[idx, :].reshape((h, w, c), order="F").transpose(1, 0, 2) + 1.0) / 2.0 * 255.0) - return img.astype('uint8') - - -def save_images(images, path): - merged_img = merge(images, [8, 8]) - if merged_img.shape[2] == 1: - im = Image.fromarray(numpy.squeeze(merged_img)).convert('RGB') - else: - im = Image.fromarray(merged_img, mode="RGB") - im.save(path) - - -def get_real_samples(batch_size, data_np): - return data_np[numpy.random.choice( - data_np.shape[0], batch_size, replace=False), :] - - -def get_noise(batch_size, noise_dim): - return numpy.random.normal(size=(batch_size, noise_dim)).astype('float32') - - -def get_fake_samples(generator_machine, batch_size, noise): - gen_inputs = api.Arguments.createArguments(1) - gen_inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(noise)) - gen_outputs = api.Arguments.createArguments(0) - generator_machine.forward(gen_inputs, gen_outputs, api.PASS_TEST) - fake_samples = gen_outputs.getSlotValue(0).copyToNumpyMat() - return fake_samples - - -def get_training_loss(training_machine, inputs): - outputs = api.Arguments.createArguments(0) - training_machine.forward(inputs, outputs, api.PASS_TEST) - loss = outputs.getSlotValue(0).copyToNumpyMat() - return numpy.mean(loss) - - -def prepare_discriminator_data_batch_pos(batch_size, data_np): - real_samples = get_real_samples(batch_size, data_np) - labels = numpy.ones(batch_size, dtype='int32') - inputs = api.Arguments.createArguments(2) - inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(real_samples)) - inputs.setSlotIds(1, api.IVector.createVectorFromNumpy(labels)) - return inputs - - -def prepare_discriminator_data_batch_neg(generator_machine, batch_size, noise): - fake_samples = get_fake_samples(generator_machine, batch_size, noise) - labels = numpy.zeros(batch_size, dtype='int32') - inputs = api.Arguments.createArguments(2) - inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(fake_samples)) - inputs.setSlotIds(1, api.IVector.createVectorFromNumpy(labels)) - return inputs - - -def prepare_generator_data_batch(batch_size, noise): - label = numpy.ones(batch_size, dtype='int32') - inputs = api.Arguments.createArguments(2) - inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(noise)) - inputs.setSlotIds(1, api.IVector.createVectorFromNumpy(label)) - return inputs - - -def find(iterable, cond): - for item in iterable: - if cond(item): - return item - return None - - -def get_layer_size(model_conf, layer_name): - layer_conf = find(model_conf.layers, lambda x: x.name == layer_name) - assert layer_conf is not None, "Cannot find '%s' layer" % layer_name - return layer_conf.size - - -def main(): - parser = argparse.ArgumentParser() - parser.add_argument("-d", "--data_source", help="mnist or cifar or uniform") - parser.add_argument( - "--use_gpu", default="1", help="1 means use gpu for training") - parser.add_argument("--gpu_id", default="0", help="the gpu_id parameter") - args = parser.parse_args() - data_source = args.data_source - use_gpu = args.use_gpu - assert data_source in ["mnist", "cifar", "uniform"] - assert use_gpu in ["0", "1"] - - if not os.path.exists("./%s_samples/" % data_source): - os.makedirs("./%s_samples/" % data_source) - - if not os.path.exists("./%s_params/" % data_source): - os.makedirs("./%s_params/" % data_source) - - api.initPaddle('--use_gpu=' + use_gpu, '--dot_period=10', - '--log_period=100', '--gpu_id=' + args.gpu_id, - '--save_dir=' + "./%s_params/" % data_source) - - if data_source == "uniform": - conf = "gan_conf.py" - num_iter = 10000 - else: - conf = "gan_conf_image.py" - num_iter = 1000 - - gen_conf = parse_config(conf, "mode=generator_training,data=" + data_source) - dis_conf = parse_config(conf, - "mode=discriminator_training,data=" + data_source) - generator_conf = parse_config(conf, "mode=generator,data=" + data_source) - batch_size = dis_conf.opt_config.batch_size - noise_dim = get_layer_size(gen_conf.model_config, "noise") - - if data_source == "mnist": - data_np = load_mnist_data("./data/mnist_data/train-images-idx3-ubyte") - elif data_source == "cifar": - data_np = load_cifar_data("./data/cifar-10-batches-py/") - else: - data_np = load_uniform_data() - - # this creates a gradient machine for discriminator - dis_training_machine = api.GradientMachine.createFromConfigProto( - dis_conf.model_config) - # this create a gradient machine for generator - gen_training_machine = api.GradientMachine.createFromConfigProto( - gen_conf.model_config) - - # generator_machine is used to generate data only, which is used for - # training discriminator - logger.info(str(generator_conf.model_config)) - generator_machine = api.GradientMachine.createFromConfigProto( - generator_conf.model_config) - - dis_trainer = api.Trainer.create(dis_conf, dis_training_machine) - - gen_trainer = api.Trainer.create(gen_conf, gen_training_machine) - - dis_trainer.startTrain() - gen_trainer.startTrain() - - # Sync parameters between networks (GradientMachine) at the beginning - copy_shared_parameters(gen_training_machine, dis_training_machine) - copy_shared_parameters(gen_training_machine, generator_machine) - - # constrain that either discriminator or generator can not be trained - # consecutively more than MAX_strike times - curr_train = "dis" - curr_strike = 0 - MAX_strike = 5 - - for train_pass in xrange(100): - dis_trainer.startTrainPass() - gen_trainer.startTrainPass() - for i in xrange(num_iter): - # Do forward pass in discriminator to get the dis_loss - noise = get_noise(batch_size, noise_dim) - data_batch_dis_pos = prepare_discriminator_data_batch_pos( - batch_size, data_np) - dis_loss_pos = get_training_loss(dis_training_machine, - data_batch_dis_pos) - - data_batch_dis_neg = prepare_discriminator_data_batch_neg( - generator_machine, batch_size, noise) - dis_loss_neg = get_training_loss(dis_training_machine, - data_batch_dis_neg) - - dis_loss = (dis_loss_pos + dis_loss_neg) / 2.0 - - # Do forward pass in generator to get the gen_loss - data_batch_gen = prepare_generator_data_batch(batch_size, noise) - gen_loss = get_training_loss(gen_training_machine, data_batch_gen) - - if i % 100 == 0: - print "d_pos_loss is %s d_neg_loss is %s" % (dis_loss_pos, - dis_loss_neg) - print "d_loss is %s g_loss is %s" % (dis_loss, gen_loss) - - # Decide which network to train based on the training history - # And the relative size of the loss - if (not (curr_train == "dis" and curr_strike == MAX_strike)) and \ - ((curr_train == "gen" and curr_strike == MAX_strike) or dis_loss > gen_loss): - if curr_train == "dis": - curr_strike += 1 - else: - curr_train = "dis" - curr_strike = 1 - dis_trainer.trainOneDataBatch(batch_size, data_batch_dis_neg) - dis_trainer.trainOneDataBatch(batch_size, data_batch_dis_pos) - copy_shared_parameters(dis_training_machine, - gen_training_machine) - - else: - if curr_train == "gen": - curr_strike += 1 - else: - curr_train = "gen" - curr_strike = 1 - gen_trainer.trainOneDataBatch(batch_size, data_batch_gen) - # TODO: add API for paddle to allow true parameter sharing between different GradientMachines - # so that we do not need to copy shared parameters. - copy_shared_parameters(gen_training_machine, - dis_training_machine) - copy_shared_parameters(gen_training_machine, generator_machine) - - dis_trainer.finishTrainPass() - gen_trainer.finishTrainPass() - # At the end of each pass, save the generated samples/images - fake_samples = get_fake_samples(generator_machine, batch_size, noise) - if data_source == "uniform": - plot2DScatter(fake_samples, "./%s_samples/train_pass%s.png" % - (data_source, train_pass)) - else: - save_images(fake_samples, "./%s_samples/train_pass%s.png" % - (data_source, train_pass)) - dis_trainer.finishTrain() - gen_trainer.finishTrain() - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/mnist/.gitignore b/v1_api_demo/mnist/.gitignore deleted file mode 100644 index 7e61d5e3a0cabd46d4185454d46610ac2ee2e63f..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/.gitignore +++ /dev/null @@ -1,10 +0,0 @@ -data/raw_data -data/*.list -mnist_vgg_model -plot.png -train.log -*pyc -.ipynb_checkpoints -params.pkl -params.tar -params.tar.gz diff --git a/v1_api_demo/mnist/api_train.py b/v1_api_demo/mnist/api_train.py deleted file mode 100644 index e42c6cbb7e0eed4f3a3625f18d79b3de64fd8e26..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/api_train.py +++ /dev/null @@ -1,209 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -""" -A very basic example for how to use current Raw SWIG API to train mnist network. - -Current implementation uses Raw SWIG, which means the API call is directly \ -passed to C++ side of Paddle. - -The user api could be simpler and carefully designed. -""" -import random - -import numpy as np -import paddle.v2 as paddle_v2 -import py_paddle.swig_paddle as api -from paddle.trainer_config_helpers import * -from py_paddle import DataProviderConverter - -from mnist_util import read_from_mnist - - -def init_parameter(network): - assert isinstance(network, api.GradientMachine) - for each_param in network.getParameters(): - assert isinstance(each_param, api.Parameter) - array_size = len(each_param) - array = np.random.uniform(-1.0, 1.0, array_size).astype('float32') - each_param.getBuf(api.PARAMETER_VALUE).copyFromNumpyArray(array) - - -def generator_to_batch(generator, batch_size): - ret_val = list() - for each_item in generator: - ret_val.append(each_item) - if len(ret_val) == batch_size: - yield ret_val - ret_val = list() - if len(ret_val) != 0: - yield ret_val - - -class BatchPool(object): - def __init__(self, generator, batch_size): - self.data = list(generator) - self.batch_size = batch_size - - def __call__(self): - random.shuffle(self.data) - for offset in xrange(0, len(self.data), self.batch_size): - limit = min(offset + self.batch_size, len(self.data)) - yield self.data[offset:limit] - - -def input_order_converter(generator): - for each_item in generator: - yield each_item['pixel'], each_item['label'] - - -def main(): - api.initPaddle("-use_gpu=false", "-trainer_count=4") # use 4 cpu cores - - optimizer = paddle_v2.optimizer.Adam( - learning_rate=1e-4, - batch_size=1000, - model_average=ModelAverage(average_window=0.5), - regularization=L2Regularization(rate=0.5)) - - # Create Local Updater. Local means not run in cluster. - # For a cluster training, here we can change to createRemoteUpdater - # in future. - updater = optimizer.create_local_updater() - assert isinstance(updater, api.ParameterUpdater) - - # define network - images = paddle_v2.layer.data( - name='pixel', type=paddle_v2.data_type.dense_vector(784)) - label = paddle_v2.layer.data( - name='label', type=paddle_v2.data_type.integer_value(10)) - hidden1 = paddle_v2.layer.fc(input=images, size=200) - hidden2 = paddle_v2.layer.fc(input=hidden1, size=200) - inference = paddle_v2.layer.fc(input=hidden2, - size=10, - act=paddle_v2.activation.Softmax()) - cost = paddle_v2.layer.classification_cost(input=inference, label=label) - - # Create Simple Gradient Machine. - model_config = paddle_v2.layer.parse_network(cost) - m = api.GradientMachine.createFromConfigProto(model_config, - api.CREATE_MODE_NORMAL, - optimizer.enable_types()) - - # This type check is not useful. Only enable type hint in IDE. - # Such as PyCharm - assert isinstance(m, api.GradientMachine) - - # Initialize Parameter by numpy. - init_parameter(network=m) - - # Initialize ParameterUpdater. - updater.init(m) - - # DataProvider Converter is a utility convert Python Object to Paddle C++ - # Input. The input format is as same as Paddle's DataProvider. - converter = DataProviderConverter(input_types=[images.type, label.type]) - - train_file = './data/raw_data/train' - test_file = './data/raw_data/t10k' - - # start gradient machine. - # the gradient machine must be started before invoke forward/backward. - # not just for training, but also for inference. - m.start() - - # evaluator can print error rate, etc. It is a C++ class. - batch_evaluator = m.makeEvaluator() - test_evaluator = m.makeEvaluator() - - # Get Train Data. - # TrainData will stored in a data pool. Currently implementation is not care - # about memory, speed. Just a very naive implementation. - train_data_generator = input_order_converter(read_from_mnist(train_file)) - train_data = BatchPool(train_data_generator, 512) - - # outArgs is Neural Network forward result. Here is not useful, just passed - # to gradient_machine.forward - outArgs = api.Arguments.createArguments(0) - - for pass_id in xrange(2): # we train 2 passes. - updater.startPass() - - for batch_id, data_batch in enumerate(train_data()): - # data_batch is input images. - # here, for online learning, we could get data_batch from network. - - # Start update one batch. - pass_type = updater.startBatch(len(data_batch)) - - # Start BatchEvaluator. - # batch_evaluator can be used between start/finish. - batch_evaluator.start() - - # forwardBackward is a shortcut for forward and backward. - # It is sometimes faster than invoke forward/backward separately, - # because in GradientMachine, it may be async. - m.forwardBackward(converter(data_batch), outArgs, pass_type) - - for each_param in m.getParameters(): - updater.update(each_param) - - # Get cost. We use numpy to calculate total cost for this batch. - cost_vec = outArgs.getSlotValue(0) - cost_vec = cost_vec.copyToNumpyMat() - cost = cost_vec.sum() / len(data_batch) - - # Make evaluator works. - m.eval(batch_evaluator) - - # Print logs. - print 'Pass id', pass_id, 'Batch id', batch_id, 'with cost=', \ - cost, batch_evaluator - - batch_evaluator.finish() - # Finish batch. - # * will clear gradient. - # * ensure all values should be updated. - updater.finishBatch(cost) - - # testing stage. use test data set to test current network. - updater.apply() - test_evaluator.start() - test_data_generator = input_order_converter(read_from_mnist(test_file)) - for data_batch in generator_to_batch(test_data_generator, 512): - # in testing stage, only forward is needed. - m.forward(converter(data_batch), outArgs, api.PASS_TEST) - m.eval(test_evaluator) - - # print error rate for test data set - print 'Pass', pass_id, ' test evaluator: ', test_evaluator - test_evaluator.finish() - updater.restore() - - updater.catchUpWith() - params = m.getParameters() - for each_param in params: - assert isinstance(each_param, api.Parameter) - value = each_param.getBuf(api.PARAMETER_VALUE) - value = value.copyToNumpyArray() - - # Here, we could save parameter to every where you want - print each_param.getName(), value - - updater.finishPass() - - m.finish() - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/mnist/data/generate_list.py b/v1_api_demo/mnist/data/generate_list.py deleted file mode 100644 index 49981cc7a93308bc96ad5097eba749440e958525..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/data/generate_list.py +++ /dev/null @@ -1,21 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -o = open("./" + "train.list", "w") -o.write("./data/raw_data/train" + "\n") -o.close() - -o = open("./" + "test.list", "w") -o.write("./data/raw_data/t10k" + "\n") -o.close() diff --git a/v1_api_demo/mnist/data/get_mnist_data.sh b/v1_api_demo/mnist/data/get_mnist_data.sh deleted file mode 100755 index 5a2e34026d4fe7f8315d4f5453bec7c4ee4f6885..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/data/get_mnist_data.sh +++ /dev/null @@ -1,21 +0,0 @@ -#!/usr/bin/env sh -# This scripts downloads the mnist data and unzips it. -set -e -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -rm -rf "$DIR/raw_data" -mkdir "$DIR/raw_data" -cd "$DIR/raw_data" - -echo "Downloading..." - -for fname in train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte -do - if [ ! -e $fname ]; then - wget --no-check-certificate http://yann.lecun.com/exdb/mnist/${fname}.gz - gunzip ${fname}.gz - fi -done - -cd $DIR -rm -f *.list -python generate_list.py diff --git a/v1_api_demo/mnist/light_mnist.py b/v1_api_demo/mnist/light_mnist.py deleted file mode 100644 index 33409054357d2f0c6a765b3ab3164eb2e584467e..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/light_mnist.py +++ /dev/null @@ -1,79 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -is_predict = get_config_arg("is_predict", bool, False) - -####################Data Configuration ################## - -if not is_predict: - data_dir = './data/' - define_py_data_sources2( - train_list=data_dir + 'train.list', - test_list=data_dir + 'test.list', - module='mnist_provider', - obj='process') - -######################Algorithm Configuration ############# -settings(batch_size=50, learning_rate=0.001, learning_method=AdamOptimizer()) - -#######################Network Configuration ############# - -data_size = 1 * 28 * 28 -label_size = 10 -img = data_layer(name='pixel', size=data_size) - - -# light cnn -# A shallower cnn model: [CNN, BN, ReLU, Max-Pooling] x4 + FC x1 -# Easier to train for mnist dataset and quite efficient -# Final performance is close to deeper ones on tasks such as digital and character classification -def light_cnn(input_image, num_channels, num_classes): - def __light__(ipt, - num_filter=128, - times=1, - conv_filter_size=3, - dropouts=0, - num_channels_=None): - return img_conv_group( - input=ipt, - num_channels=num_channels_, - pool_size=2, - pool_stride=2, - conv_padding=0, - conv_num_filter=[num_filter] * times, - conv_filter_size=conv_filter_size, - conv_act=ReluActivation(), - conv_with_batchnorm=True, - conv_batchnorm_drop_rate=dropouts, - pool_type=MaxPooling()) - - tmp = __light__(input_image, num_filter=128, num_channels_=num_channels) - tmp = __light__(tmp, num_filter=128) - tmp = __light__(tmp, num_filter=128) - tmp = __light__(tmp, num_filter=128, conv_filter_size=1) - - tmp = fc_layer(input=tmp, size=num_classes, act=SoftmaxActivation()) - return tmp - - -predict = light_cnn(input_image=img, num_channels=1, num_classes=label_size) - -if not is_predict: - lbl = data_layer(name="label", size=label_size) - inputs(img, lbl) - outputs(classification_cost(input=predict, label=lbl)) -else: - outputs(predict) diff --git a/v1_api_demo/mnist/mnist_provider.py b/v1_api_demo/mnist/mnist_provider.py deleted file mode 100644 index 4192339837620aada84b64a92fef3e05953971c2..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/mnist_provider.py +++ /dev/null @@ -1,25 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -from paddle.trainer.PyDataProvider2 import * -from mnist_util import read_from_mnist - - -# Define a py data provider -@provider( - input_types={'pixel': dense_vector(28 * 28), - 'label': integer_value(10)}, - cache=CacheType.CACHE_PASS_IN_MEM) -def process(settings, filename): # settings is not used currently. - for each in read_from_mnist(filename): - yield each diff --git a/v1_api_demo/mnist/mnist_util.py b/v1_api_demo/mnist/mnist_util.py deleted file mode 100644 index 3fd88ae7edc821296ca0accbf6dedc083e411744..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/mnist_util.py +++ /dev/null @@ -1,30 +0,0 @@ -import numpy - -__all__ = ['read_from_mnist'] - - -def read_from_mnist(filename): - imgf = filename + "-images-idx3-ubyte" - labelf = filename + "-labels-idx1-ubyte" - f = open(imgf, "rb") - l = open(labelf, "rb") - - f.read(16) - l.read(8) - - # Define number of samples for train/test - if "train" in filename: - n = 60000 - else: - n = 10000 - - images = numpy.fromfile( - f, 'ubyte', count=n * 28 * 28).reshape((n, 28 * 28)).astype('float32') - images = images / 255.0 * 2.0 - 1.0 - labels = numpy.fromfile(l, 'ubyte', count=n).astype("int") - - for i in xrange(n): - yield {"pixel": images[i, :], 'label': labels[i]} - - f.close() - l.close() diff --git a/v1_api_demo/mnist/train.sh b/v1_api_demo/mnist/train.sh deleted file mode 100755 index ca2b1ad9eb960685b95b0f294a9b929e1a4acab1..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/train.sh +++ /dev/null @@ -1,32 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e -config=vgg_16_mnist.py -output=./mnist_vgg_model -log=train.log - -paddle train \ ---config=$config \ ---dot_period=10 \ ---log_period=100 \ ---test_all_data_in_one_period=1 \ ---use_gpu=0 \ ---trainer_count=1 \ ---num_passes=100 \ ---save_dir=$output \ -2>&1 | tee $log -paddle usage -l $log -e $? -n "mnist_train" >/dev/null 2>&1 - -python -m paddle.utils.plotcurve -i $log > plot.png diff --git a/v1_api_demo/mnist/vgg_16_mnist.py b/v1_api_demo/mnist/vgg_16_mnist.py deleted file mode 100644 index a819b391c690fb473801eb2e7ba3161cc31b5b4b..0000000000000000000000000000000000000000 --- a/v1_api_demo/mnist/vgg_16_mnist.py +++ /dev/null @@ -1,50 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -is_predict = get_config_arg("is_predict", bool, False) - -####################Data Configuration ################## - -if not is_predict: - data_dir = './data/' - define_py_data_sources2( - train_list=data_dir + 'train.list', - test_list=data_dir + 'test.list', - module='mnist_provider', - obj='process') - -######################Algorithm Configuration ############# -settings( - batch_size=128, - learning_rate=0.1 / 128.0, - learning_method=MomentumOptimizer(0.9), - regularization=L2Regularization(0.0005 * 128)) - -#######################Network Configuration ############# - -data_size = 1 * 28 * 28 -label_size = 10 -img = data_layer(name='pixel', size=data_size) - -# small_vgg is predined in trainer_config_helpers.network -predict = small_vgg(input_image=img, num_channels=1, num_classes=label_size) - -if not is_predict: - lbl = data_layer(name="label", size=label_size) - inputs(img, lbl) - outputs(classification_cost(input=predict, label=lbl)) -else: - outputs(predict) diff --git a/v1_api_demo/model_zoo/embedding/.gitignore b/v1_api_demo/model_zoo/embedding/.gitignore deleted file mode 100644 index 908f5a3fb2f7c34368ea24d0fc3ac9cac29a4fdb..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/embedding/.gitignore +++ /dev/null @@ -1,2 +0,0 @@ -baidu.dict -model_*.emb diff --git a/v1_api_demo/model_zoo/embedding/extract_para.py b/v1_api_demo/model_zoo/embedding/extract_para.py deleted file mode 100755 index 570b90c1f772c8f6abfc6cda02560fd3471ef0b6..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/embedding/extract_para.py +++ /dev/null @@ -1,113 +0,0 @@ -#!/bin/env python -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" -Example: - python extract_para.py --preModel PREMODEL --preDict PREDICT \ - --usrModel USRMODEL --usrDict USRDICT -d DIM - -Options: - -h, --help show this help message and exit - --preModel PREMODEL the name of pretrained embedding model - --preDict PREDICT the name of pretrained dictionary - --usrModel usrModel the name of output usr embedding model - --usrDict usrDict the name of user specified dictionary - -d DIM dimension of parameter -""" -from optparse import OptionParser -import struct - - -def get_row_index(preDict, usrDict): - """ - Get the row positions for all words in user dictionary from pre-trained dictionary. - return: a list of row positions - Example: preDict='a\nb\nc\n', usrDict='a\nc\n', then return [0,2] - """ - pos = [] - index = dict() - with open(preDict, "r") as f: - for line_index, line in enumerate(f): - word = line.strip().split()[0] - index[word] = line_index - with open(usrDict, "r") as f: - for line in f: - word = line.strip().split()[0] - pos.append(index[word]) - return pos - - -def extract_parameters_by_usrDict(preModel, preDict, usrModel, usrDict, - paraDim): - """ - Extract desired parameters from a pretrained embedding model based on user dictionary - """ - if paraDim not in [32, 64, 128, 256]: - raise RuntimeError("We only support 32, 64, 128, 256 dimensions now") - - fi = open(preModel, "rb") - fo = open(usrModel, "wb") - - # write filehead - rowIndex = get_row_index(preDict, usrDict) - newHead = struct.pack("iil", 0, 4, len(rowIndex) * paraDim) - fo.write(newHead) - bytes = 4 * paraDim - for i in range(0, len(rowIndex)): - # find the absolute position of input file - fi.seek(rowIndex[i] * bytes + 16, 0) - fo.write(fi.read(bytes)) - - print "extract parameters finish, total", len(rowIndex), "lines" - fi.close() - - -def main(): - """ - Main entry for running paraconvert.py - """ - usage = "usage: \n" \ - "python %prog --preModel PREMODEL --preDict PREDICT" \ - " --usrModel USRMODEL --usrDict USRDICT -d DIM" - parser = OptionParser(usage) - parser.add_option( - "--preModel", - action="store", - dest="preModel", - help="the name of pretrained embedding model") - parser.add_option( - "--preDict", - action="store", - dest="preDict", - help="the name of pretrained dictionary") - parser.add_option( - "--usrModel", - action="store", - dest="usrModel", - help="the name of output usr embedding model") - parser.add_option( - "--usrDict", - action="store", - dest="usrDict", - help="the name of user specified dictionary") - parser.add_option( - "-d", action="store", dest="dim", help="dimension of parameter") - (options, args) = parser.parse_args() - extract_parameters_by_usrDict(options.preModel, options.preDict, - options.usrModel, options.usrDict, - int(options.dim)) - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/model_zoo/embedding/paraconvert.py b/v1_api_demo/model_zoo/embedding/paraconvert.py deleted file mode 100755 index ce7a70efc43d7f85708f1e12bb94739f3588370c..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/embedding/paraconvert.py +++ /dev/null @@ -1,159 +0,0 @@ -#!/bin/env python -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" -Example: - python paraconvert.py --b2t -i INPUT -o OUTPUT -d DIM - python paraconvert.py --t2b -i INPUT -o OUTPUT - -Options: - -h, --help show this help message and exit - --b2t convert parameter file of embedding model from binary to text - --t2b convert parameter file of embedding model from text to binary - -i INPUT input parameter file name - -o OUTPUT output parameter file name - -d DIM dimension of parameter -""" -from optparse import OptionParser -import struct - - -def binary2text(input, output, paraDim): - """ - Convert a binary parameter file of embedding model to be a text file. - input: the name of input binary parameter file, the format is: - 1) the first 16 bytes is filehead: - version(4 bytes): version of paddle, default = 0 - floatSize(4 bytes): sizeof(float) = 4 - paraCount(8 bytes): total number of parameter - 2) the next (paraCount * 4) bytes is parameters, each has 4 bytes - output: the name of output text parameter file, for example: - 0,4,32156096 - -0.7845433,1.1937413,-0.1704215,... - 0.0000909,0.0009465,-0.0008813,... - ... - the format is: - 1) the first line is filehead: - version=0, floatSize=4, paraCount=32156096 - 2) other lines print the paramters - a) each line prints paraDim paramters splitted by ',' - b) there is paraCount/paraDim lines (embedding words) - paraDim: dimension of parameters - """ - fi = open(input, "rb") - fo = open(output, "w") - """ - """ - version, floatSize, paraCount = struct.unpack("iil", fi.read(16)) - newHead = ','.join([str(version), str(floatSize), str(paraCount)]) - print >> fo, newHead - - bytes = 4 * int(paraDim) - format = "%df" % int(paraDim) - context = fi.read(bytes) - line = 0 - - while context: - numbers = struct.unpack(format, context) - lst = [] - for i in numbers: - lst.append('%8.7f' % i) - print >> fo, ','.join(lst) - context = fi.read(bytes) - line += 1 - fi.close() - fo.close() - print "binary2text finish, total", line, "lines" - - -def get_para_count(input): - """ - Compute the total number of embedding parameters in input text file. - input: the name of input text file - """ - numRows = 1 - paraDim = 0 - with open(input) as f: - line = f.readline() - paraDim = len(line.split(",")) - for line in f: - numRows += 1 - return numRows * paraDim - - -def text2binary(input, output, paddle_head=True): - """ - Convert a text parameter file of embedding model to be a binary file. - input: the name of input text parameter file, for example: - -0.7845433,1.1937413,-0.1704215,... - 0.0000909,0.0009465,-0.0008813,... - ... - the format is: - 1) it doesn't have filehead - 2) each line stores the same dimension of parameters, - the separator is commas ',' - output: the name of output binary parameter file, the format is: - 1) the first 16 bytes is filehead: - version(4 bytes), floatSize(4 bytes), paraCount(8 bytes) - 2) the next (paraCount * 4) bytes is parameters, each has 4 bytes - """ - fi = open(input, "r") - fo = open(output, "wb") - - newHead = struct.pack("iil", 0, 4, get_para_count(input)) - fo.write(newHead) - - count = 0 - for line in fi: - line = line.strip().split(",") - for i in range(0, len(line)): - binary_data = struct.pack("f", float(line[i])) - fo.write(binary_data) - count += 1 - fi.close() - fo.close() - print "text2binary finish, total", count, "lines" - - -def main(): - """ - Main entry for running paraconvert.py - """ - usage = "usage: \n" \ - "python %prog --b2t -i INPUT -o OUTPUT -d DIM \n" \ - "python %prog --t2b -i INPUT -o OUTPUT" - parser = OptionParser(usage) - parser.add_option( - "--b2t", - action="store_true", - help="convert parameter file of embedding model from binary to text") - parser.add_option( - "--t2b", - action="store_true", - help="convert parameter file of embedding model from text to binary") - parser.add_option( - "-i", action="store", dest="input", help="input parameter file name") - parser.add_option( - "-o", action="store", dest="output", help="output parameter file name") - parser.add_option( - "-d", action="store", dest="dim", help="dimension of parameter") - (options, args) = parser.parse_args() - if options.b2t: - binary2text(options.input, options.output, options.dim) - if options.t2b: - text2binary(options.input, options.output) - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/model_zoo/embedding/pre_DictAndModel.sh b/v1_api_demo/model_zoo/embedding/pre_DictAndModel.sh deleted file mode 100755 index f61c65a935c76032a06613cfe0b50f1c90bc50d9..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/embedding/pre_DictAndModel.sh +++ /dev/null @@ -1,32 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e -set -x -BASE_URL='http://paddlepaddle.cdn.bcebos.com/model_zoo/embedding' - -DOWNLOAD_ITEMS=(baidu.dict model_32.emb model_64.emb model_128.emb model_256.emb) -ITEM_MD5=(fa03a12321eaab6c30a8fcc9442eaea3 - f88c8325ee6da6187f1080e8fe66c1cd - 927cf70f27f860aff1a5703ebf7f1584 - a52e43655cd25d279777ed509a1ae27b - b92c67fe9ff70fea53596080e351ac80) - -for ((i=0; i<${#ITEM_MD5[@]}; i++)) -do - FILENAME=${DOWNLOAD_ITEMS[${i}]} - REAL_MD5=`wget ${BASE_URL}/${FILENAME} -O - | tee ${FILENAME} | md5sum | cut -d ' ' -f 1` - EXPECTED_MD5=${ITEM_MD5[${i}]} - [ "${EXPECTED_MD5}" = "${REAL_MD5}" ] -done diff --git a/v1_api_demo/model_zoo/resnet/.gitignore b/v1_api_demo/model_zoo/resnet/.gitignore deleted file mode 100644 index 7a64209b62340a5c5a51626821028e63ed5e588e..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/.gitignore +++ /dev/null @@ -1,5 +0,0 @@ -fea_output/ -features/ -model.list -ResNet_50.dot -ResNet_50.png diff --git a/v1_api_demo/model_zoo/resnet/classify.py b/v1_api_demo/model_zoo/resnet/classify.py deleted file mode 100755 index 6074cc1d3a85e13e3e8d336d81e22104f9d8e7cf..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/classify.py +++ /dev/null @@ -1,312 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import os -import sys -import cPickle -import logging -from PIL import Image -import numpy as np -from optparse import OptionParser - -import paddle.utils.image_util as image_util - -from py_paddle import swig_paddle, DataProviderConverter -from paddle.trainer.PyDataProvider2 import dense_vector -from paddle.trainer.config_parser import parse_config - -logging.basicConfig( - format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s') -logging.getLogger().setLevel(logging.INFO) - - -class ImageClassifier(): - def __init__(self, - train_conf, - model_dir=None, - resize_dim=256, - crop_dim=224, - use_gpu=True, - mean_file=None, - output_layer=None, - oversample=False, - is_color=True): - """ - train_conf: network configure. - model_dir: string, directory of model. - resize_dim: int, resized image size. - crop_dim: int, crop size. - mean_file: string, image mean file. - oversample: bool, oversample means multiple crops, namely five - patches (the four corner patches and the center - patch) as well as their horizontal reflections, - ten crops in all. - """ - self.train_conf = train_conf - self.model_dir = model_dir - if model_dir is None: - self.model_dir = os.path.dirname(train_conf) - - self.resize_dim = resize_dim - self.crop_dims = [crop_dim, crop_dim] - self.oversample = oversample - self.is_color = is_color - - self.output_layer = output_layer - if self.output_layer: - assert isinstance(self.output_layer, basestring) - self.output_layer = self.output_layer.split(",") - - self.transformer = image_util.ImageTransformer(is_color=is_color) - self.transformer.set_transpose((2, 0, 1)) - self.transformer.set_channel_swap((2, 1, 0)) - - self.mean_file = mean_file - if self.mean_file is not None: - mean = np.load(self.mean_file)['data_mean'] - mean = mean.reshape(3, self.crop_dims[0], self.crop_dims[1]) - self.transformer.set_mean(mean) # mean pixel - else: - # if you use three mean value, set like: - # this three mean value is calculated from ImageNet. - self.transformer.set_mean(np.array([103.939, 116.779, 123.68])) - - conf_args = "is_test=1,use_gpu=%d,is_predict=1" % (int(use_gpu)) - conf = parse_config(train_conf, conf_args) - swig_paddle.initPaddle("--use_gpu=%d" % (int(use_gpu))) - self.network = swig_paddle.GradientMachine.createFromConfigProto( - conf.model_config) - assert isinstance(self.network, swig_paddle.GradientMachine) - self.network.loadParameters(self.model_dir) - - data_size = 3 * self.crop_dims[0] * self.crop_dims[1] - slots = [dense_vector(data_size)] - self.converter = DataProviderConverter(slots) - - def get_data(self, img_path): - """ - 1. load image from img_path. - 2. resize or oversampling. - 3. transformer data: transpose, channel swap, sub mean. - return K x H x W ndarray. - - img_path: image path. - """ - image = image_util.load_image(img_path, self.is_color) - # Another way to extract oversampled features is that - # cropping and averaging from large feature map which is - # calculated by large size of image. - # This way reduces the computation. - if self.oversample: - # image_util.resize_image: short side is self.resize_dim - image = image_util.resize_image(image, self.resize_dim) - image = np.array(image) - input = np.zeros( - (1, image.shape[0], image.shape[1], 3), dtype=np.float32) - input[0] = image.astype(np.float32) - input = image_util.oversample(input, self.crop_dims) - else: - image = image.resize(self.crop_dims, Image.ANTIALIAS) - input = np.zeros( - (1, self.crop_dims[0], self.crop_dims[1], 3), dtype=np.float32) - input[0] = np.array(image).astype(np.float32) - - data_in = [] - for img in input: - img = self.transformer.transformer(img).flatten() - data_in.append([img.tolist()]) - # paddle input: [[[]],[[]],...], [[]] is one sample. - return data_in - - def forward(self, input_data): - """ - return output arguments which are the Outputs() in network configure. - - input_data: py_paddle input data. - call forward. - """ - in_arg = self.converter(input_data) - return self.network.forwardTest(in_arg) - - def forward(self, data, output_layer): - """ - return output arguments which are the Outputs() in network configure. - - input_data: py_paddle input data. - call forward. - """ - input = self.converter(data) - self.network.forwardTest(input) - output = self.network.getLayerOutputs(output_layer) - res = {} - if isinstance(output_layer, basestring): - output_layer = [output_layer] - for name in output_layer: - # For oversampling, average predictions across crops. - # If not, the shape of output[name]: (1, class_number), - # the mean is also applicable. - res[name] = output[name]['value'].mean(0) - - return res - - def predict(self, data_file): - """ - call forward and predicting. - - data_file: input image list. - """ - image_files = open(data_file, 'rb').readlines() - results = {} - if self.output_layer is None: - self.output_layer = ["output"] - for line in image_files: - image = line.split()[0] - data = self.get_data(image) - prob = self.forward(data, self.output_layer) - lab = np.argsort(-prob[self.output_layer[0]]) - results[image] = lab[0] - logging.info("Label of %s is: %d", image, lab[0]) - return results - - def extract(self, data_file, output_dir, batch_size=10000): - """ - extract and save features of output layers, which are - specify in Outputs() in network configure. - - data_file: file name of input data. - output_dir: saved directory of extracted features. - batch_size: sample number of one batch file. - """ - if not os.path.exists(output_dir): - os.mkdir(output_dir) - - sample_num = 0 - batch_num = 0 - image_feature = {} - image_files = open(data_file, 'rb').readlines() - for idx, line in enumerate(image_files): - image = line.split()[0] - data = self.get_data(image) - feature = self.forward(data, self.output_layer) - # save extracted features - file_name = image.split("/")[-1] - image_feature[file_name] = feature - sample_num += 1 - if sample_num == batch_size: - batch_name = os.path.join(output_dir, 'batch_%d' % (batch_num)) - self.save_file(image_feature, batch_name) - logging.info('Finish batch %d', batch_num) - batch_num += 1 - sample_num = 0 - image_feature = {} - if idx % 1000 == 0: - logging.info('%d/%d, %s', idx, len(image_files), file_name) - if sample_num > 0: - batch_name = os.path.join(output_dir, 'batch_%d' % (batch_num)) - self.save_file(image_feature, batch_name) - logging.info('Finish batch %d', batch_num) - logging.info('Done: make image feature batch') - - def save_file(self, data, file): - of = open(file, 'wb') - cPickle.dump(data, of, protocol=cPickle.HIGHEST_PROTOCOL) - - -def option_parser(): - """ - Main entry for predciting - """ - usage = "%prog -c config -i data_list -w model_dir [options]" - parser = OptionParser(usage="usage: %s" % usage) - parser.add_option( - "-j", - "--job", - action="store", - dest="job_type", - help="job type: predict, extract\ - predict: predicting,\ - extract: extract features") - parser.add_option( - "-c", - "--conf", - action="store", - dest="train_conf", - help="network config") - parser.add_option( - "-i", "--data", action="store", dest="data_file", help="image list") - parser.add_option( - "-w", - "--model", - action="store", - dest="model_path", - default=None, - help="model path") - parser.add_option( - "-g", - "--use_gpu", - action="store", - dest="use_gpu", - default=True, - help="Whether to use gpu mode.") - parser.add_option( - "-o", - "--output_dir", - action="store", - dest="output_dir", - default="output", - help="output path") - parser.add_option( - "-m", - "--mean", - action="store", - dest="mean", - default=None, - help="mean file.") - parser.add_option( - "-p", - "--multi_crop", - action="store_true", - dest="multi_crop", - default=False, - help="Wether to use multiple crops on image.") - parser.add_option("-l", "--output_layer", action="store", - dest="output_layer", default=None, - help="--job=extract, specify layers to extract "\ - "features, --job=predict, specify layer of " - "classification probability, output in resnet.py.") - return parser.parse_args() - - -def main(): - """ - 1. parse input arguments. - 2. predicting or extract features according job type. - """ - options, args = option_parser() - obj = ImageClassifier( - options.train_conf, - options.model_path, - use_gpu=options.use_gpu, - mean_file=options.mean, - output_layer=options.output_layer, - oversample=options.multi_crop) - if options.job_type == "predict": - obj.predict(options.data_file) - - elif options.job_type == "extract": - obj.extract(options.data_file, options.output_dir) - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/model_zoo/resnet/example/.gitignore b/v1_api_demo/model_zoo/resnet/example/.gitignore deleted file mode 100644 index 4a2b5962a6800f251cba655c026331f14648c86e..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/example/.gitignore +++ /dev/null @@ -1 +0,0 @@ -*image_list_provider_copy_1.py diff --git a/v1_api_demo/model_zoo/resnet/example/cat.jpg b/v1_api_demo/model_zoo/resnet/example/cat.jpg deleted file mode 100644 index 47b01db90eddc46ff845f10bc2accaf2364c272d..0000000000000000000000000000000000000000 Binary files a/v1_api_demo/model_zoo/resnet/example/cat.jpg and /dev/null differ diff --git a/v1_api_demo/model_zoo/resnet/example/dog.jpg b/v1_api_demo/model_zoo/resnet/example/dog.jpg deleted file mode 100644 index b9cc33cf069da5c453b97dbb7383838edd07c199..0000000000000000000000000000000000000000 Binary files a/v1_api_demo/model_zoo/resnet/example/dog.jpg and /dev/null differ diff --git a/v1_api_demo/model_zoo/resnet/example/image_list_provider.py b/v1_api_demo/model_zoo/resnet/example/image_list_provider.py deleted file mode 100644 index 2cd8eb8bf850f41282ed5db2885dc0b7218c79f7..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/example/image_list_provider.py +++ /dev/null @@ -1,102 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.utils.image_util import * -from paddle.trainer.PyDataProvider2 import * - - -def hook(settings, image_size, crop_size, color, file_list, is_train, **kwargs): - """ - Description: Init with a list of data file - file_list is the name list of input files. - kwargs["load_data_args"] is the value of 'load_data_args' - which can be set in config. - Each args is separated by a column. - image_size: the crop image size. - mean_meta: the path of the meta file to store the mean image. - mean_value: can be mean value, not a file. - can not set mean_meta and mean_value at the same time. - color: 'color' means a color image. Otherwise, it means a gray image. - is_train: whether the data provider is used for training. - Data argumentation might be different for training and testing. - """ - settings.img_size = image_size - settings.crop_size = crop_size - settings.mean_img_size = settings.crop_size - settings.color = color # default is color - settings.is_train = is_train - - settings.is_swap_channel = kwargs.get('swap_channel', None) - if settings.is_swap_channel is not None: - settings.swap_channel = settings.is_swap_channel - settings.is_swap_channel = True - - if settings.color: - settings.img_input_size = settings.crop_size * settings.crop_size * 3 - else: - settings.img_input_size = settings.crop_size * settings.crop_size - - settings.file_list = file_list - settings.mean_meta = kwargs.get('mean_meta', None) - settings.mean_value = kwargs.get('mean_value', None) - # can not specify both mean_meta and mean_value. - assert not (settings.mean_meta and settings.mean_value) - if not settings.mean_meta: - settings.mean_value = kwargs.get('mean_value') - sz = settings.crop_size * settings.crop_size - settings.img_mean = np.zeros(sz * 3, dtype=np.single) - for idx, value in enumerate(settings.mean_value): - settings.img_mean[idx * sz:(idx + 1) * sz] = value - settings.img_mean = settings.img_mean.reshape(3, settings.crop_size, - settings.crop_size) - - else: - settings.img_mean = load_meta(settings.mean_meta, - settings.mean_img_size, - settings.crop_size, settings.color) - - settings.input_types = [ - dense_vector(settings.img_input_size), # image feature - integer_value(1) - ] # labels - - settings.logger.info('Image short side: %s', settings.img_size) - settings.logger.info('Crop size: %s', settings.crop_size) - settings.logger.info('Meta path: %s', settings.mean_meta) - if settings.is_swap_channel: - settings.logger.info('swap channel: %s', settings.swap_channel) - settings.logger.info('DataProvider Initialization finished') - - -@provider(init_hook=hook, should_shuffle=False) -def processData(settings, file_list): - """ - The main function for loading data. - Load the batch, iterate all the images and labels in this batch. - file_name: the batch file name. - """ - img_path, lab = file_list.strip().split(' ') - img = Image.open(img_path) - img.load() - img = img.resize((settings.img_size, settings.img_size), Image.ANTIALIAS) - img = np.array(img).astype(np.float32) - if len(img.shape) == 3: - img = np.swapaxes(img, 1, 2) - img = np.swapaxes(img, 1, 0) - # swap channel - if settings.is_swap_channel: - img = img[settings.swap_channel, :, :] - img_feat = preprocess_img(img, settings.img_mean, settings.crop_size, - settings.is_train, settings.color) - yield img_feat.tolist(), int(lab.strip()) diff --git a/v1_api_demo/model_zoo/resnet/example/test.list b/v1_api_demo/model_zoo/resnet/example/test.list deleted file mode 100644 index 30bbf630b640a26239fc104c9c08f6ebc9dfaa82..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/example/test.list +++ /dev/null @@ -1,2 +0,0 @@ -example/dog.jpg 0 -example/cat.jpg 0 diff --git a/v1_api_demo/model_zoo/resnet/extract_fea_c++.sh b/v1_api_demo/model_zoo/resnet/extract_fea_c++.sh deleted file mode 100755 index 5447aa92dfb5facd3433eb4a1893e96e3c786c73..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/extract_fea_c++.sh +++ /dev/null @@ -1,40 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -#set names of layer which you want to extract feature -#in Outputs() of resnet.py -#like: Outputs("res5_3_branch2c_conv", "res5_3_branch2c_bn") -layer_num=50 -configure=./resnet.py -model_path=./model/resnet_$layer_num -fea_dir=fea_output -#Output is text file. -#Each line is one sample's features. -#If you set N layer names in Outputs() -#each line contains N features sperated by ";". - -# create model list file. -model_list=./model.list -touch $model_list | echo $model_path > $model_list - -paddle train \ - --local=true \ - --job=test \ - --config=$configure \ - --model_list=$model_list \ - --use_gpu=1 \ - --predict_output_dir=$fea_dir \ - --config_args=is_test=1,layer_num=$layer_num diff --git a/v1_api_demo/model_zoo/resnet/extract_fea_py.sh b/v1_api_demo/model_zoo/resnet/extract_fea_py.sh deleted file mode 100755 index 2e87152f7f8598f487870291271cdee646105044..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/extract_fea_py.sh +++ /dev/null @@ -1,29 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -#Note if you use CPU mode, you need to set use_gpu=0 in classify.py. like this: -#conf_args = "is_test=0,use_gpu=1,is_predict=1" -#conf = parse_config(train_conf, conf_args) -#swig_paddle.initPaddle("--use_gpu=0") -python classify.py \ - --job=extract \ - --conf=resnet.py \ - --use_gpu=1 \ - --mean=model/mean_meta_224/mean.meta \ - --model=model/resnet_50 \ - --data=./example/test.list \ - --output_layer="res5_3_branch2c_conv,res5_3_branch2c_bn" \ - --output_dir=features diff --git a/v1_api_demo/model_zoo/resnet/get_model.sh b/v1_api_demo/model_zoo/resnet/get_model.sh deleted file mode 100755 index b33d8178ab7859fc0b0d514fb19bec2c28a77c3d..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/get_model.sh +++ /dev/null @@ -1,32 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -cd $DIR - -mkdir model -cd model - -echo "Downloading ResNet models..." - -for file in resnet_50.tar.gz resnet_101.tar.gz resnet_152.tar.gz mean_meta_224.tar.gz -do - wget http://paddlepaddle.bj.bcebos.com/model_zoo/imagenet/$file - tar -xvf $file - rm $file -done - -echo "Done." diff --git a/v1_api_demo/model_zoo/resnet/load_feature.py b/v1_api_demo/model_zoo/resnet/load_feature.py deleted file mode 100644 index 5d3d0c0d30ef710c37c98e93a51b2f813d636b59..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/load_feature.py +++ /dev/null @@ -1,63 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import os -import sys -import cPickle -import logging - -logging.basicConfig( - format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s') -logging.getLogger().setLevel(logging.INFO) - - -def load_feature_c(file): - """ - Load feature extracted by C++ interface. - Return a list. - file: feature file. - """ - features = [] - f = open(file, 'r') - for line in f: - sample = [] - for slot in line.strip().split(";"): - fea = [float(val) for val in slot.strip().split()] - if fea: - sample.append(fea) - features.append(sample) - f.close() - return features - - -def load_feature_py(feature_dir): - """ - Load feature extracted by python interface. - Return a dictionary. - feature_dir: directory of feature file. - """ - file_list = os.listdir(feature_dir) - file_list = [os.path.join(feature_dir, f) for f in file_list] - features = {} - for file_name in file_list: - with open(file_name, 'rb') as f: - feature = cPickle.load(f) - features.update(feature) - logging.info('Load feature file %s', file_name) - return features - - -if __name__ == '__main__': - print load_feature_py(sys.argv[1]) - #print load_feature_c(sys.argv[1]) diff --git a/v1_api_demo/model_zoo/resnet/net_diagram.sh b/v1_api_demo/model_zoo/resnet/net_diagram.sh deleted file mode 100755 index 1b06ffa44eec8a0f312420c35699d3902f9a6400..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/net_diagram.sh +++ /dev/null @@ -1,39 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -:' -Visual deep residual network -1. Using make_model_diagram.py to generate dot file. -2. Using graphviz to convert dot file. - -Usage: -./net_diagram.sh -' - -set -e - -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -cd $DIR - -img_type=png -img_fileprefix=ResNet_50 -conf_filename=resnet.py -dot_filename=ResNet_50.dot -config_str="layer_num=50,data_provider=0" - -python -m paddle.utils.make_model_diagram $conf_filename $dot_filename $config_str - -# If you have installed graphviz, running like this: -# dot -Tpng -o ResNet.png ResNet.dot diff --git a/v1_api_demo/model_zoo/resnet/predict.sh b/v1_api_demo/model_zoo/resnet/predict.sh deleted file mode 100755 index 2b67b17c48c60cc8a7b7c46a1c80a3f2bf281870..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/predict.sh +++ /dev/null @@ -1,23 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -python classify.py \ - --job=predict \ - --conf=resnet.py\ - --model=model/resnet_50 \ - --multi_crop \ - --use_gpu=1 \ - --data=./example/test.list diff --git a/v1_api_demo/model_zoo/resnet/resnet.py b/v1_api_demo/model_zoo/resnet/resnet.py deleted file mode 100644 index 6fdd97fefc62392c93ecffae0fc918e8dc4b18c5..0000000000000000000000000000000000000000 --- a/v1_api_demo/model_zoo/resnet/resnet.py +++ /dev/null @@ -1,271 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * -""" -paper: https://arxiv.org/abs/1512.03385 -""" -is_test = get_config_arg("is_test", bool, False) -is_predict = get_config_arg("is_predict", bool, False) -data_provider = get_config_arg("data_provider", bool, True) -layer_num = get_config_arg("layer_num", int, 50) - -if not is_predict and data_provider: - train_list = 'train.list' if not is_test else None - # mean.meta is mean file of ImageNet dataset. - # mean.meta size : 3 x 224 x 224. - # If you use three mean value, set like: - # "mean_value:103.939,116.779,123.68;" - args = { - 'mean_meta': "model/mean_meta_224/mean.meta", - 'image_size': 224, - 'crop_size': 224, - 'color': True, - 'swap_channel:': [2, 1, 0] - } - define_py_data_sources2( - train_list, - 'example/test.list', - module="example.image_list_provider", - obj="processData", - args=args) - -batch_size = 1 -learning_rate = 0.1 / batch_size -momentum = 0.9 -weight_decay = 0.0001 * batch_size -default_momentum(momentum) -default_decay_rate(weight_decay) - -Settings( - algorithm='sgd', - batch_size=batch_size, - learning_rate=learning_rate, - - # set the appropriate parameters according your schedule - learning_method='momentum', - learning_rate_decay_a=0.5, - learning_rate_decay_b=1200000 * 10, - learning_rate_schedule="discexp", ) - - -def conv_bn_layer(name, - input, - filter_size, - num_filters, - stride, - padding, - channels=None, - active_type=ReluActivation()): - """ - A wrapper for conv layer with batch normalization layers. - Note: - conv layer has no activation. - """ - - tmp = img_conv_layer( - name=name + "_conv", - input=input, - filter_size=filter_size, - num_channels=channels, - num_filters=num_filters, - stride=stride, - padding=padding, - act=LinearActivation(), - bias_attr=False) - return batch_norm_layer( - name=name + "_bn", input=tmp, act=active_type, use_global_stats=is_test) - - -def bottleneck_block(name, input, num_filters1, num_filters2): - """ - A wrapper for bottlenect building block in ResNet. - Last conv_bn_layer has no activation. - Addto layer has activation of relu. - """ - last_name = conv_bn_layer( - name=name + '_branch2a', - input=input, - filter_size=1, - num_filters=num_filters1, - stride=1, - padding=0) - last_name = conv_bn_layer( - name=name + '_branch2b', - input=last_name, - filter_size=3, - num_filters=num_filters1, - stride=1, - padding=1) - last_name = conv_bn_layer( - name=name + '_branch2c', - input=last_name, - filter_size=1, - num_filters=num_filters2, - stride=1, - padding=0, - active_type=LinearActivation()) - - return addto_layer( - name=name + "_addto", input=[input, last_name], act=ReluActivation()) - - -def mid_projection(name, input, num_filters1, num_filters2, stride=2): - """ - A wrapper for middile projection in ResNet. - projection shortcuts are used for increasing dimensions, - and other shortcuts are identity - branch1: projection shortcuts are used for increasing - dimensions, has no activation. - branch2x: bottleneck building block, shortcuts are identity. - """ - # stride = 2 - branch1 = conv_bn_layer( - name=name + '_branch1', - input=input, - filter_size=1, - num_filters=num_filters2, - stride=stride, - padding=0, - active_type=LinearActivation()) - - last_name = conv_bn_layer( - name=name + '_branch2a', - input=input, - filter_size=1, - num_filters=num_filters1, - stride=stride, - padding=0) - last_name = conv_bn_layer( - name=name + '_branch2b', - input=last_name, - filter_size=3, - num_filters=num_filters1, - stride=1, - padding=1) - - last_name = conv_bn_layer( - name=name + '_branch2c', - input=last_name, - filter_size=1, - num_filters=num_filters2, - stride=1, - padding=0, - active_type=LinearActivation()) - - return addto_layer( - name=name + "_addto", input=[branch1, last_name], act=ReluActivation()) - - -def deep_res_net(res2_num=3, res3_num=4, res4_num=6, res5_num=3): - """ - A wrapper for 50,101,152 layers of ResNet. - res2_num: number of blocks stacked in conv2_x - res3_num: number of blocks stacked in conv3_x - res4_num: number of blocks stacked in conv4_x - res5_num: number of blocks stacked in conv5_x - """ - # For ImageNet - # conv1: 112x112 - img = data_layer(name='input', size=224 * 224 * 3) - tmp = conv_bn_layer( - "conv1", - img, - filter_size=7, - channels=3, - num_filters=64, - stride=2, - padding=3) - tmp = img_pool_layer(name="pool1", input=tmp, pool_size=3, stride=2) - - # conv2_x: 56x56 - tmp = mid_projection( - name="res2_1", input=tmp, num_filters1=64, num_filters2=256, stride=1) - for i in xrange(2, res2_num + 1, 1): - tmp = bottleneck_block( - name="res2_" + str(i), input=tmp, num_filters1=64, num_filters2=256) - - # conv3_x: 28x28 - tmp = mid_projection( - name="res3_1", input=tmp, num_filters1=128, num_filters2=512) - for i in xrange(2, res3_num + 1, 1): - tmp = bottleneck_block( - name="res3_" + str(i), - input=tmp, - num_filters1=128, - num_filters2=512) - - # conv4_x: 14x14 - tmp = mid_projection( - name="res4_1", input=tmp, num_filters1=256, num_filters2=1024) - for i in xrange(2, res4_num + 1, 1): - tmp = bottleneck_block( - name="res4_" + str(i), - input=tmp, - num_filters1=256, - num_filters2=1024) - - # conv5_x: 7x7 - tmp = mid_projection( - name="res5_1", input=tmp, num_filters1=512, num_filters2=2048) - for i in xrange(2, res5_num + 1, 1): - tmp = bottleneck_block( - name="res5_" + str(i), - input=tmp, - num_filters1=512, - num_filters2=2048) - - tmp = img_pool_layer( - name='avgpool', - input=tmp, - pool_size=7, - stride=1, - pool_type=AvgPooling()) - - output = fc_layer( - name='output', input=tmp, size=1000, act=SoftmaxActivation()) - - if not is_predict: - classification_cost( - input=output, label=data_layer( - name='label', size=1)) - - -def res_net_50(): - deep_res_net(3, 4, 6, 3) - - -def res_net_101(): - deep_res_net(3, 4, 23, 3) - - -def res_net_152(): - deep_res_net(3, 8, 36, 3) - - -if not is_predict: - Inputs("input", "label") -else: - Inputs("input") -# Outputs("cost-softmax" if not is_predict else "output") -Outputs("res5_3_branch2c_conv", "res5_3_branch2c_bn") - -if layer_num == 50: - res_net_50() -elif layer_num == 101: - res_net_101() -elif layer_num == 152: - res_net_152() -else: - print("Wrong layer number.") diff --git a/v1_api_demo/quick_start/.gitignore b/v1_api_demo/quick_start/.gitignore deleted file mode 100644 index f71662563ff96d6227dd568d9951a90b0d09456e..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/.gitignore +++ /dev/null @@ -1,15 +0,0 @@ -*.pyc -data/dict.txt -data/dict_all.txt -data/labels.list -data/mosesdecoder-master/ -data/reviews_Electronics_5.json.gz -data/test.list -data/test.txt -data/train.list -data/train.txt -data/pred.list -data/pred.txt -dataprovider_copy_1.py -train.log -output diff --git a/v1_api_demo/quick_start/api_predict.py b/v1_api_demo/quick_start/api_predict.py deleted file mode 100755 index 9bdffe1006281c58a595e2771561ba62e4c2d6bd..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/api_predict.py +++ /dev/null @@ -1,147 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import os, sys -import numpy as np -from optparse import OptionParser -from py_paddle import swig_paddle, DataProviderConverter -from paddle.trainer.PyDataProvider2 import sparse_binary_vector -from paddle.trainer.config_parser import parse_config -""" -Usage: run following command to show help message. - python api_predict.py -h -""" - - -class QuickStartPrediction(): - def __init__(self, train_conf, dict_file, model_dir=None, label_file=None): - """ - train_conf: trainer configure. - dict_file: word dictionary file name. - model_dir: directory of model. - """ - self.train_conf = train_conf - self.dict_file = dict_file - self.word_dict = {} - self.dict_dim = self.load_dict() - self.model_dir = model_dir - if model_dir is None: - self.model_dir = os.path.dirname(train_conf) - - self.label = None - if label_file is not None: - self.load_label(label_file) - - conf = parse_config(train_conf, "is_predict=1") - self.network = swig_paddle.GradientMachine.createFromConfigProto( - conf.model_config) - self.network.loadParameters(self.model_dir) - input_types = [sparse_binary_vector(self.dict_dim)] - self.converter = DataProviderConverter(input_types) - - def load_dict(self): - """ - Load dictionary from self.dict_file. - """ - for line_count, line in enumerate(open(self.dict_file, 'r')): - self.word_dict[line.strip().split('\t')[0]] = line_count - return len(self.word_dict) - - def load_label(self, label_file): - """ - Load label. - """ - self.label = {} - for v in open(label_file, 'r'): - self.label[int(v.split('\t')[1])] = v.split('\t')[0] - - def get_index(self, data): - """ - transform word into integer index according to the dictionary. - """ - words = data.strip().split() - word_slot = [self.word_dict[w] for w in words if w in self.word_dict] - return word_slot - - def batch_predict(self, data_batch): - input = self.converter(data_batch) - output = self.network.forwardTest(input) - prob = output[0]["id"].tolist() - print("predicting labels is:") - print prob - - -def option_parser(): - usage = "python predict.py -n config -w model_dir -d dictionary -i input_file " - parser = OptionParser(usage="usage: %s [options]" % usage) - parser.add_option( - "-n", - "--tconf", - action="store", - dest="train_conf", - help="network config") - parser.add_option( - "-d", - "--dict", - action="store", - dest="dict_file", - help="dictionary file") - parser.add_option( - "-b", - "--label", - action="store", - dest="label", - default=None, - help="dictionary file") - parser.add_option( - "-c", - "--batch_size", - type="int", - action="store", - dest="batch_size", - default=1, - help="the batch size for prediction") - parser.add_option( - "-w", - "--model", - action="store", - dest="model_path", - default=None, - help="model path") - return parser.parse_args() - - -def main(): - options, args = option_parser() - train_conf = options.train_conf - batch_size = options.batch_size - dict_file = options.dict_file - model_path = options.model_path - label = options.label - swig_paddle.initPaddle("--use_gpu=0") - predict = QuickStartPrediction(train_conf, dict_file, model_path, label) - - batch = [] - labels = [] - for line in sys.stdin: - [label, text] = line.split("\t") - labels.append(int(label)) - batch.append([predict.get_index(text)]) - print("labels is:") - print labels - predict.batch_predict(batch) - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/quick_start/api_predict.sh b/v1_api_demo/quick_start/api_predict.sh deleted file mode 100755 index 4d9aa9e8854ed79446a47dbc593f419cdda077b4..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/api_predict.sh +++ /dev/null @@ -1,30 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -#Note the default model is pass-00002, you shold make sure the model path -#exists or change the mode path. -#only test on trainer_config.lr.py -model=output/model/pass-00001/ -config=trainer_config.lr.py -label=data/labels.list -dict=data/dict.txt -batch_size=20 -head -n$batch_size data/test.txt | python api_predict.py \ - --tconf=$config\ - --model=$model \ - --label=$label \ - --dict=$dict \ - --batch_size=$batch_size diff --git a/v1_api_demo/quick_start/api_train.py b/v1_api_demo/quick_start/api_train.py deleted file mode 100644 index 5699789daa4051661b0a72c69f4668f2d8bb9cb2..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/api_train.py +++ /dev/null @@ -1,122 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import argparse -import itertools -import random - -from paddle.trainer.config_parser import parse_config -from py_paddle import swig_paddle as api -from py_paddle import DataProviderConverter -from paddle.trainer.PyDataProvider2 \ - import integer_value, integer_value_sequence, sparse_binary_vector - - -def parse_arguments(): - parser = argparse.ArgumentParser() - parser.add_argument( - "--train_data", type=str, required=False, help="train data file") - parser.add_argument("--test_data", type=str, help="test data file") - parser.add_argument( - "--config", type=str, required=True, help="config file name") - parser.add_argument("--dict_file", required=True, help="dictionary file") - parser.add_argument( - "--seq", default=1, type=int, help="whether use sequence training") - parser.add_argument( - "--use_gpu", default=0, type=int, help="whether use GPU for training") - parser.add_argument( - "--trainer_count", - default=1, - type=int, - help="Number of threads for training") - parser.add_argument( - "--num_passes", default=5, type=int, help="Number of training passes") - return parser.parse_args() - - -UNK_IDX = 0 - - -def load_data(file_name, word_dict): - with open(file_name, 'r') as f: - for line in f: - label, comment = line.strip().split('\t') - words = comment.split() - word_slot = [word_dict.get(w, UNK_IDX) for w in words] - yield word_slot, int(label) - - -def load_dict(dict_file): - word_dict = dict() - with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - return word_dict - - -def main(): - options = parse_arguments() - api.initPaddle("--use_gpu=%s" % options.use_gpu, - "--trainer_count=%s" % options.trainer_count) - - word_dict = load_dict(options.dict_file) - train_dataset = list(load_data(options.train_data, word_dict)) - if options.test_data: - test_dataset = list(load_data(options.test_data, word_dict)) - else: - test_dataset = None - - trainer_config = parse_config(options.config, - "dict_file=%s" % options.dict_file) - # No need to have data provider for trainer - trainer_config.ClearField('data_config') - trainer_config.ClearField('test_data_config') - - # create a GradientMachine from the model configuratin - model = api.GradientMachine.createFromConfigProto( - trainer_config.model_config) - # create a trainer for the gradient machine - trainer = api.Trainer.create(trainer_config, model) - - # create a data converter which converts data to PaddlePaddle - # internal format - input_types = [ - integer_value_sequence(len(word_dict)) if options.seq else - sparse_binary_vector(len(word_dict)), integer_value(2) - ] - converter = DataProviderConverter(input_types) - - batch_size = trainer_config.opt_config.batch_size - trainer.startTrain() - for train_pass in xrange(options.num_passes): - trainer.startTrainPass() - random.shuffle(train_dataset) - for pos in xrange(0, len(train_dataset), batch_size): - batch = itertools.islice(train_dataset, pos, pos + batch_size) - size = min(batch_size, len(train_dataset) - pos) - trainer.trainOneDataBatch(size, converter(batch)) - trainer.finishTrainPass() - if test_dataset: - trainer.startTestPeriod() - for pos in xrange(0, len(test_dataset), batch_size): - batch = itertools.islice(test_dataset, pos, pos + batch_size) - size = min(batch_size, len(test_dataset) - pos) - trainer.testOneDataBatch(size, converter(batch)) - trainer.finishTestPeriod() - trainer.finishTrain() - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/quick_start/api_train.sh b/v1_api_demo/quick_start/api_train.sh deleted file mode 100755 index 9b2a4e2f224b1677c458ede66a6a3bac09d8ad61..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/api_train.sh +++ /dev/null @@ -1,29 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -# Note: if using trainer_config.emb.py, trainer_config.cnn.py -# or trainer_config.lstm.py, you need to change --seq to --seq=1 -# because they are sequence models. -python api_train.py \ - --config=trainer_config.lr.py \ - --trainer_count=2 \ - --num_passes=15 \ - --use_gpu=0 \ - --seq=0 \ - --train_data=data/train.txt \ - --test_data=data/test.txt \ - --dict_file=data/dict.txt \ - 2>&1 | tee 'train.log' diff --git a/v1_api_demo/quick_start/cluster/cluster_train.sh b/v1_api_demo/quick_start/cluster/cluster_train.sh deleted file mode 100755 index a7b1f01064b29cf6abc4cd6b706ee466a6d6da36..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/cluster/cluster_train.sh +++ /dev/null @@ -1,45 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -# Should run pserver.sh before run this script. -bin_dir=$(cd `dirname $0`; pwd) -home_dir=$(cd "${bin_dir}/.."; pwd) -source "$bin_dir/env.sh" - -model_dir="$bin_dir/output" -log_file="$bin_dir/train.log" - -pushd "$home_dir" -cfg=trainer_config.lr.py -paddle train \ - --start_pserver=false \ - --config=$cfg \ - --save_dir=${model_dir} \ - --trainer_count=4 \ - --local=0 \ - --log_period=100 \ - --num_passes=15 \ - --use_gpu=false \ - --show_parameter_stats_period=100 \ - --test_all_data_in_one_period=1 \ - --num_gradient_servers=1 \ - --nics=`get_nics` \ - --port=7164 \ - --ports_num=1 \ - --pservers="127.0.0.1" \ - --comment="paddle_trainer" \ - 2>&1 | tee "$log_file" -popd diff --git a/v1_api_demo/quick_start/cluster/env.sh b/v1_api_demo/quick_start/cluster/env.sh deleted file mode 100644 index a404993835d0e479f65c89c5561855293b7b66f0..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/cluster/env.sh +++ /dev/null @@ -1,28 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -function get_nics() { - machine=`uname -s` - local nics="" - if [ "$machine" == "Linux" ]; then - nics="lo" - elif [ "$machine" == "Darwin" ]; then - nics="lo0" - else - nics="unsupport" - fi - echo $nics -} diff --git a/v1_api_demo/quick_start/cluster/pserver.sh b/v1_api_demo/quick_start/cluster/pserver.sh deleted file mode 100755 index b187c1d9b9108a607ed310253d54ecc096f0e792..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/cluster/pserver.sh +++ /dev/null @@ -1,26 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e -bin_dir=$(cd `dirname $0`; pwd) -source "$bin_dir/env.sh" - -paddle pserver \ - --nics=`get_nics` \ - --port=7164 \ - --ports_num=1 \ - --ports_num_for_sparse=1 \ - --num_gradient_servers=1 \ - --comment="paddle_pserver" \ - 2>&1 | tee 'pserver.log' diff --git a/v1_api_demo/quick_start/data/README.md b/v1_api_demo/quick_start/data/README.md deleted file mode 100644 index 63abcf7ebf31903213e44cf492b93e09f61db14e..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/data/README.md +++ /dev/null @@ -1,9 +0,0 @@ -This dataset consists of electronics product reviews associated with -binary labels (positive/negative) for sentiment classification. - -The preprocessed data can be downloaded by script `get_data.sh`. -The data was derived from reviews_Electronics_5.json.gz at - -http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz - -If you want to process the raw data, you can use the script `proc_from_raw_data/get_data.sh`. diff --git a/v1_api_demo/quick_start/data/get_data.sh b/v1_api_demo/quick_start/data/get_data.sh deleted file mode 100755 index a09a18f919e5a84f1f7c889a43f0a5fbf4a60a77..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/data/get_data.sh +++ /dev/null @@ -1,27 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -cd $DIR - -# Download the preprocessed data -wget http://paddlepaddle.bj.bcebos.com/demo/quick_start_preprocessed_data/preprocessed_data.tar.gz - -# Extract package -tar zxvf preprocessed_data.tar.gz - -# Remove compressed package -rm preprocessed_data.tar.gz diff --git a/v1_api_demo/quick_start/data/proc_from_raw_data/get_data.sh b/v1_api_demo/quick_start/data/proc_from_raw_data/get_data.sh deleted file mode 100755 index d976eaebfaa600778e0ab6bb0adbd7159f1cce2f..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/data/proc_from_raw_data/get_data.sh +++ /dev/null @@ -1,79 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -# 1. size of pos : neg = 1:1. -# 2. size of testing set = min(25k, len(all_data) * 0.1), others is traning set. -# 3. distinct train set and test set. - -set -e - -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -cd $DIR - -# Download data -echo "Downloading Amazon Electronics reviews data..." -# http://jmcauley.ucsd.edu/data/amazon/ -wget http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz -echo "Downloading mosesdecoder..." -# https://github.com/moses-smt/mosesdecoder -wget https://github.com/moses-smt/mosesdecoder/archive/master.zip - -unzip master.zip -rm master.zip - -################## -# Preprocess data -echo "Preprocess data..." -export LC_ALL=C -UNAME_STR=`uname` - -if [ ${UNAME_STR} == 'Linux' ]; then - SHUF_PROG='shuf' -else - SHUF_PROG='gshuf' -fi - -mkdir -p tmp -python preprocess.py -i reviews_Electronics_5.json.gz -# uniq and shuffle -cd tmp -echo 'Uniq and shuffle...' -cat pos_*|sort|uniq|${SHUF_PROG}> pos.shuffed -cat neg_*|sort|uniq|${SHUF_PROG}> neg.shuffed - -min_len=`sed -n '$=' neg.shuffed` -test_num=$((min_len/10)) -if [ $test_num -gt 12500 ];then - test_num=12500 -fi -train_num=$((min_len-test_num)) - -head -n$train_num pos.shuffed >train.pos -head -n$train_num neg.shuffed >train.neg -tail -n$test_num pos.shuffed >test.pos -tail -n$test_num neg.shuffed >test.neg - -cat train.pos train.neg | ${SHUF_PROG} >../train.txt -cat test.pos test.neg | ${SHUF_PROG} >../test.txt - -cd - -echo 'train.txt' > train.list -echo 'test.txt' > test.list - -# use 30k dict -rm -rf tmp -mv dict.txt dict_all.txt -cat dict_all.txt | head -n 30001 > dict.txt -echo 'Done.' diff --git a/v1_api_demo/quick_start/data/proc_from_raw_data/preprocess.py b/v1_api_demo/quick_start/data/proc_from_raw_data/preprocess.py deleted file mode 100755 index 5706351a21fbd15d9bbf197156bb0fdabcb07295..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/data/proc_from_raw_data/preprocess.py +++ /dev/null @@ -1,236 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# -*- coding: UTF-8 -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" -1. Tokenize the words and punctuation -2. pos sample : rating score 5; neg sample: rating score 1-2. - -Usage: - python preprocess.py -i data_file [random seed] -""" - -import sys -import os -import operator -import gzip -from subprocess import Popen, PIPE -from optparse import OptionParser -import json -from multiprocessing import Queue -from multiprocessing import Pool -import multiprocessing - -batch_size = 5000 -word_count = {} -num_tokenize = max(1, - multiprocessing.cpu_count() - 2) # parse + tokenize + save -max_queue_size = 8 -parse_queue = Queue(maxsize=max_queue_size + num_tokenize) -tokenize_queue = Queue(maxsize=max_queue_size + num_tokenize) - - -def create_dict(data): - """ - Create dictionary based on data, and saved in data_dir/dict.txt. - The first line is unk \t -1. - data: list, input data by batch. - """ - for seq in data: - try: - for w in seq.lower().split(): - if w not in word_count: - word_count[w] = 1 - else: - word_count[w] += 1 - except: - sys.stderr.write(seq + "\tERROR\n") - - -def parse(path): - """ - Open .gz file. - """ - sys.stderr.write(path) - g = gzip.open(path, 'r') - for l in g: - yield json.loads(l) - g.close() - - -def tokenize(sentences): - """ - Use tokenizer.perl to tokenize input sentences. - tokenizer.perl is tool of Moses. - sentences : a list of input sentences. - return: a list of processed text. - """ - dir = './mosesdecoder-master/scripts/tokenizer/tokenizer.perl' - if not os.path.exists(dir): - sys.exit( - "The ./mosesdecoder-master/scripts/tokenizer/tokenizer.perl does not exists." - ) - tokenizer_cmd = [dir, '-l', 'en', '-q', '-'] - assert isinstance(sentences, list) - text = "\n".join(sentences) - tokenizer = Popen(tokenizer_cmd, stdin=PIPE, stdout=PIPE) - tok_text, _ = tokenizer.communicate(text) - toks = tok_text.split('\n')[:-1] - return toks - - -def save_data(instance, data_dir, pre_fix, batch_num): - """ - save data by batch - """ - label = ['1' if pre_fix == 'pos' else '0' for i in range(len(instance))] - lines = ['%s\t%s' % (label[i], instance[i]) for i in range(len(label))] - file_name = os.path.join(data_dir, "%s_%s.txt" % (pre_fix, batch_num)) - file(file_name, 'w').write('\n'.join(lines) + '\n') - - -def tokenize_batch(id): - """ - tokenize data by batch - """ - while True: - num_batch, instance, pre_fix = parse_queue.get() - if num_batch == -1: ### parse_queue finished - tokenize_queue.put((-1, None, None)) - sys.stderr.write("Thread %s finish\n" % (id)) - break - tokenize_instance = tokenize(instance) - tokenize_queue.put((num_batch, tokenize_instance, pre_fix)) - sys.stderr.write('.') - - -def save_batch(data_dir, num_tokenize, data_dir_dict): - """ - save data by batch - build dict.txt - """ - token_count = 0 - while True: - num_batch, instance, pre_fix = tokenize_queue.get() - if num_batch == -1: - token_count += 1 - if token_count == num_tokenize: #### tokenize finished. - break - else: - continue - save_data(instance, data_dir, pre_fix, num_batch) - create_dict(instance) ## update dict - - sys.stderr.write("save file finish\n") - f = open(data_dir_dict, 'w') - f.write('%s\t%s\n' % ('unk', '-1')) - for k, v in sorted(word_count.items(), key=operator.itemgetter(1), \ - reverse=True): - f.write('%s\t%s\n' % (k, v)) - f.close() - sys.stderr.write("build dict finish\n") - - -def parse_batch(data, num_tokenize): - """ - parse data by batch - parse -> tokenize -> save - """ - raw_txt = parse(data) - neg, pos = [], [] - count = 0 - sys.stderr.write("extract raw data\n") - for l in raw_txt: - rating = l["overall"] - text = l["reviewText"].lower() # # convert words to lower case - if rating == 5.0 and text: - pos.append(text) - if rating < 3.0 and text: - neg.append(text) - if len(pos) == batch_size or len(neg) == batch_size: - if len(pos) == batch_size: - batch = pos - pre_fix = 'pos' - else: - batch = neg - pre_fix = 'neg' - - parse_queue.put((count, batch, pre_fix)) - count += 1 - if pre_fix == 'pos': - pos = [] - else: - neg = [] - - if len(pos) > 0: - parse_queue.put((count, pos, 'pos')) - count += 1 - if len(neg) > 0: - parse_queue.put((count, neg, 'neg')) - count += 1 - for i in range(num_tokenize): - parse_queue.put((-1, None, None)) #### for tokenize's input finished - sys.stderr.write("parsing finish\n") - - -def option_parser(): - parser = OptionParser(usage="usage: python preprcoess.py "\ - "-i data_path [options]") - parser.add_option( - "-i", "--data", action="store", dest="input", help="Input data path.") - parser.add_option( - "-s", - "--seed", - action="store", - dest="seed", - default=1024, - help="Set random seed.") - return parser.parse_args() - - -def main(): - reload(sys) - sys.setdefaultencoding('utf-8') - options, args = option_parser() - data = options.input - seed = options.seed - data_dir_dict = os.path.join(os.path.dirname(data), 'dict.txt') - data_dir = os.path.join(os.path.dirname(data), 'tmp') - pool = Pool(processes=num_tokenize + 2) - pool.apply_async(parse_batch, args=(data, num_tokenize)) - for i in range(num_tokenize): - pool.apply_async(tokenize_batch, args=(str(i), )) - pool.apply_async(save_batch, args=(data_dir, num_tokenize, data_dir_dict)) - pool.close() - pool.join() - - file(os.path.join(os.path.dirname(data), 'labels.list'), - 'w').write('neg\t0\npos\t1\n') - - -if __name__ == '__main__': - main() diff --git a/v1_api_demo/quick_start/dataprovider_bow.py b/v1_api_demo/quick_start/dataprovider_bow.py deleted file mode 100644 index 2745495586449b5d1eb64ae570f73eb6b14dbdfe..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/dataprovider_bow.py +++ /dev/null @@ -1,86 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer.PyDataProvider2 import * - -# id of the word not in dictionary -UNK_IDX = 0 - - -# initializer is called by the framework during initialization. -# It allows the user to describe the data types and setup the -# necessary data structure for later use. -# `settings` is an object. initializer need to properly fill settings.input_types. -# initializer can also store other data structures needed to be used at process(). -# In this example, dictionary is stored in settings. -# `dictionay` and `kwargs` are arguments passed from trainer_config.lr.py -def initializer(settings, dictionary, **kwargs): - # Put the word dictionary into settings - settings.word_dict = dictionary - - # setting.input_types specifies what the data types the data provider - # generates. - settings.input_types = { - # The first input is a sparse_binary_vector, - # which means each dimension of the vector is either 0 or 1. It is the - # bag-of-words (BOW) representation of the texts. - 'word': sparse_binary_vector(len(dictionary)), - # The second input is an integer. It represents the category id of the - # sample. 2 means there are two labels in the dataset. - # (1 for positive and 0 for negative) - 'label': integer_value(2) - } - - -# Delaring a data provider. It has an initializer 'data_initialzer'. -# It will cache the generated data of the first pass in memory, so that -# during later pass, no on-the-fly data generation will be needed. -# `setting` is the same object used by initializer() -# `file_name` is the name of a file listed train_list or test_list file given -# to define_py_data_sources2(). See trainer_config.lr.py. -@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM) -def process(settings, file_name): - # Open the input data file. - with open(file_name, 'r') as f: - # Read each line. - for line in f: - # Each line contains the label and text of the comment, separated by \t. - label, comment = line.strip().split('\t') - - # Split the words into a list. - words = comment.split() - - # convert the words into a list of ids by looking them up in word_dict. - word_vector = [settings.word_dict.get(w, UNK_IDX) for w in words] - - # Return the features for the current comment. The first is a list - # of ids representing a 0-1 binary sparse vector of the text, - # the second is the integer id of the label. - yield {'word': word_vector, 'label': int(label)} - - -def predict_initializer(settings, dictionary, **kwargs): - settings.word_dict = dictionary - settings.input_types = {'word': sparse_binary_vector(len(dictionary))} - - -# Declaring a data provider for prediction. The difference with process -# is that label is not generated. -@provider(init_hook=predict_initializer, should_shuffle=False) -def process_predict(settings, file_name): - with open(file_name, 'r') as f: - for line in f: - comment = line.strip().split() - word_vector = [settings.word_dict.get(w, UNK_IDX) for w in comment] - yield {'word': word_vector} diff --git a/v1_api_demo/quick_start/dataprovider_emb.py b/v1_api_demo/quick_start/dataprovider_emb.py deleted file mode 100755 index ddfa3ce9b73555cb3b7f5a44314ca35b12d41ede..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/dataprovider_emb.py +++ /dev/null @@ -1,52 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer.PyDataProvider2 import * - -UNK_IDX = 0 - - -def initializer(settings, dictionary, **kwargs): - settings.word_dict = dictionary - settings.input_types = { - # Define the type of the first input as sequence of integer. - # The value of the integers range from 0 to len(dictrionary)-1 - 'word': integer_value_sequence(len(dictionary)), - # Define the second input for label id - 'label': integer_value(2) - } - - -@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM) -def process(settings, file_name): - with open(file_name, 'r') as f: - for line in f: - label, comment = line.strip().split('\t') - words = comment.split() - word_slot = [settings.word_dict.get(w, UNK_IDX) for w in words] - yield {'word': word_slot, 'label': int(label)} - - -def predict_initializer(settings, dictionary, **kwargs): - settings.word_dict = dictionary - settings.input_types = {'word': integer_value_sequence(len(dictionary))} - - -@provider(init_hook=predict_initializer, should_shuffle=False) -def process_predict(settings, file_name): - with open(file_name, 'r') as f: - for line in f: - comment = line.strip().split() - word_slot = [settings.word_dict.get(w, UNK_IDX) for w in comment] - yield {'word': word_slot} diff --git a/v1_api_demo/quick_start/predict.sh b/v1_api_demo/quick_start/predict.sh deleted file mode 100755 index e47c2dd01fb5c919203964e298018e6dc2bd366e..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/predict.sh +++ /dev/null @@ -1,32 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -cfg=trainer_config.lr.py -#cfg=trainer_config.emb.py -#cfg=trainer_config.cnn.py -#cfg=trainer_config.lstm.py -model="output/pass-00003" -paddle train \ - --config=$cfg \ - --use_gpu=false \ - --job=test \ - --init_model_path=$model \ - --config_args=is_predict=1 \ - --predict_output_dir=. \ -2>&1 | tee 'predict.log' -paddle usage -l 'predict.log' -e $? -n "quick_start_predict_${cfg}" >/dev/null 2>&1 - -mv rank-00000 result.txt diff --git a/v1_api_demo/quick_start/train.sh b/v1_api_demo/quick_start/train.sh deleted file mode 100755 index 01697fed48054be8ad98a01d4cbb5029e6a1ead0..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/train.sh +++ /dev/null @@ -1,34 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -cfg=trainer_config.lr.py -#cfg=trainer_config.emb.py -#cfg=trainer_config.cnn.py -#cfg=trainer_config.lstm.py -#cfg=trainer_config.bidi-lstm.py -#cfg=trainer_config.db-lstm.py -#cfg=trainer_config.resnet-lstm.py -paddle train \ - --config=$cfg \ - --save_dir=./output \ - --trainer_count=4 \ - --log_period=100 \ - --num_passes=15 \ - --use_gpu=false \ - --show_parameter_stats_period=100 \ - --test_all_data_in_one_period=1 \ - 2>&1 | tee 'train.log' -paddle usage -l "train.log" -e $? -n "quick_start_${cfg}" >/dev/null 2>&1 diff --git a/v1_api_demo/quick_start/trainer_config.bidi-lstm.py b/v1_api_demo/quick_start/trainer_config.bidi-lstm.py deleted file mode 100644 index 3deff4aa00b1ea5d66097514867d1a392393a523..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.bidi-lstm.py +++ /dev/null @@ -1,74 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -dict_file = "./data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_emb", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -bias_attr = ParamAttr(initial_std=0., l2_rate=0.) -data = data_layer(name="word", size=len(word_dict)) -emb = embedding_layer(input=data, size=128) - -bi_lstm = bidirectional_lstm(input=emb, size=128) -dropout = dropout_layer(input=bi_lstm, dropout_rate=0.5) - -output = fc_layer( - input=dropout, size=2, bias_attr=bias_attr, act=SoftmaxActivation()) - -if is_predict: - maxid = maxid_layer(output) - outputs([maxid, output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) - outputs(cls) diff --git a/v1_api_demo/quick_start/trainer_config.cnn.py b/v1_api_demo/quick_start/trainer_config.cnn.py deleted file mode 100644 index e09e41484d30db385a1d276b7f346b444fe79d3d..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.cnn.py +++ /dev/null @@ -1,68 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -dict_file = "./data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_emb", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -data = data_layer(name="word", size=len(word_dict)) -embedding = embedding_layer(input=data, size=128) -conv = sequence_conv_pool(input=embedding, context_len=3, hidden_size=512) -output = fc_layer(input=conv, size=2, act=SoftmaxActivation()) -if is_predict: - maxid = maxid_layer(output) - outputs([maxid, output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) - outputs(cls) diff --git a/v1_api_demo/quick_start/trainer_config.db-lstm.py b/v1_api_demo/quick_start/trainer_config.db-lstm.py deleted file mode 100644 index fba802b4600b33cfbfd0820cce1f47e4d0f948ae..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.db-lstm.py +++ /dev/null @@ -1,74 +0,0 @@ -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -dict_file = "./data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_emb", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -bias_attr = ParamAttr(initial_std=0., l2_rate=0.) - -data = data_layer(name="word", size=len(word_dict)) -emb = embedding_layer(input=data, size=128) - -hidden_0 = mixed_layer(size=128, input=[full_matrix_projection(input=emb)]) -lstm_0 = lstmemory(input=hidden_0, layer_attr=ExtraAttr(drop_rate=0.1)) - -input_layers = [hidden_0, lstm_0] - -for i in range(1, 8): - fc = fc_layer(input=input_layers, size=128) - lstm = lstmemory( - input=fc, - layer_attr=ExtraAttr(drop_rate=0.1), - reverse=(i % 2) == 1, ) - input_layers = [fc, lstm] - -lstm_last = pooling_layer(input=lstm, pooling_type=MaxPooling()) - -output = fc_layer( - input=lstm_last, size=2, bias_attr=bias_attr, act=SoftmaxActivation()) - -if is_predict: - maxid = maxid_layer(output) - outputs([maxid, output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) - outputs(cls) diff --git a/v1_api_demo/quick_start/trainer_config.emb.py b/v1_api_demo/quick_start/trainer_config.emb.py deleted file mode 100644 index f69f98ff7fc885d3fe16d3aaf66967389b3b3240..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.emb.py +++ /dev/null @@ -1,64 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -dict_file = "./data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_emb", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, learning_rate=2e-3, learning_method=AdamOptimizer()) - -data = data_layer(name="word", size=len(word_dict)) -embedding = embedding_layer(input=data, size=128) -avg = pooling_layer(input=embedding, pooling_type=AvgPooling()) -output = fc_layer(input=avg, size=2, act=SoftmaxActivation()) -if is_predict: - maxid = maxid_layer(output) - outputs([maxid, output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) - outputs(cls) diff --git a/v1_api_demo/quick_start/trainer_config.lr.py b/v1_api_demo/quick_start/trainer_config.lr.py deleted file mode 100644 index b7b694940e338acbc40ffd3e5597f209bf07488f..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.lr.py +++ /dev/null @@ -1,85 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -dict_file = get_config_arg('dict_file', str, "./data/dict.txt") -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' - -# define the data sources for the model. -# We need to use different process for training and prediction. -# For training, the input data includes both word IDs and labels. -# For prediction, the input data only includs word Ids. -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_bow", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -# Define the data for text features. The size of the data layer is the number -# of words in the dictionary. -data = data_layer(name="word", size=len(word_dict)) - -# Define a fully connected layer with logistic activation. -# (also called softmax activation). -output = fc_layer(input=data, size=2, act=SoftmaxActivation()) - -if not is_predict: - # For training, we need label and cost - - # define the category id for each example. - # The size of the data layer is the number of labels. - label = data_layer(name="label", size=2) - - # Define cross-entropy classification loss and error. - cls = classification_cost(input=output, label=label) - outputs(cls) -else: - # For prediction, no label is needed. We need to output - # We need to output classification result, and class probabilities. - maxid = maxid_layer(output) - outputs([maxid, output]) diff --git a/v1_api_demo/quick_start/trainer_config.lstm.py b/v1_api_demo/quick_start/trainer_config.lstm.py deleted file mode 100644 index 8967d78807b9bbf990f5dd36240c18199b86954e..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.lstm.py +++ /dev/null @@ -1,70 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -dict_file = "./data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_emb", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -data = data_layer(name="word", size=len(word_dict)) -emb = embedding_layer(input=data, size=128) -lstm = simple_lstm( - input=emb, size=128, lstm_cell_attr=ExtraAttr(drop_rate=0.25)) -lstm_max = pooling_layer(input=lstm, pooling_type=MaxPooling()) -output = fc_layer(input=lstm_max, size=2, act=SoftmaxActivation()) -if is_predict: - maxid = maxid_layer(output) - outputs([maxid, output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) - outputs(cls) diff --git a/v1_api_demo/quick_start/trainer_config.resnet-lstm.py b/v1_api_demo/quick_start/trainer_config.resnet-lstm.py deleted file mode 100644 index 32d0596f250c0f0c5a4004d3af7adb794b3f0f1b..0000000000000000000000000000000000000000 --- a/v1_api_demo/quick_start/trainer_config.resnet-lstm.py +++ /dev/null @@ -1,104 +0,0 @@ -# Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserve. -# -#Licensed under the Apache License, Version 2.0 (the "License"); -#you may not use this file except in compliance with the License. -#You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -#Unless required by applicable law or agreed to in writing, software -#distributed under the License is distributed on an "AS IS" BASIS, -#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -#See the License for the specific language governing permissions and -#limitations under the License. -# edit-mode: -*- python -*- - -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -""" -This configuration is a demonstration of how to implement the stacked LSTM -with residual connections, i.e. an LSTM layer takes the sum of the hidden states -and inputs of the previous LSTM layer instead of only the hidden states. -This architecture is from: -Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V. Le, Mohammad Norouzi, -Wolfgang Macherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, -Jeff Klingner, Apurva Shah, Melvin Johnson, Xiaobing Liu, Lukasz Kaiser, -Stephan Gouws, Yoshikiyo Kato, Taku Kudo, Hideto Kazawa, Keith Stevens, -George Kurian, Nishant Patil, Wei Wang, Cliff Young, Jason Smith, Jason Riesa, -Alex Rudnick, Oriol Vinyals, Greg Corrado, Macduff Hughes, Jeffrey Dean. 2016. -Google's Neural Machine Translation System: Bridging the Gap between Human and -Machine Translation. In arXiv https://arxiv.org/pdf/1609.08144v2.pdf -Different from the architecture described in the paper, we use a stack single -direction LSTM layers as the first layer instead of bi-directional LSTM. Also, -since this is a demo code, to reduce computation time, we stacked 4 layers -instead of 8 layers. -""" - -from paddle.trainer_config_helpers import * - -dict_file = "./data/dict.txt" -word_dict = dict() -with open(dict_file, 'r') as f: - for i, line in enumerate(f): - w = line.strip().split()[0] - word_dict[w] = i - -is_predict = get_config_arg('is_predict', bool, False) -trn = 'data/train.list' if not is_predict else None -tst = 'data/test.list' if not is_predict else 'data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, - test_list=tst, - module="dataprovider_emb", - obj=process, - args={"dictionary": word_dict}) - -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=2e-3, - learning_method=AdamOptimizer(), - regularization=L2Regularization(8e-4), - gradient_clipping_threshold=25) - -bias_attr = ParamAttr(initial_std=0., l2_rate=0.) - -data = data_layer(name="word", size=len(word_dict)) -emb = embedding_layer(input=data, size=128) -lstm = simple_lstm(input=emb, size=128, lstm_cell_attr=ExtraAttr(drop_rate=0.1)) - -previous_input, previous_hidden_state = emb, lstm - -for i in range(3): - # The input to the current layer is the sum of the hidden state - # and input of the previous layer. - current_input = addto_layer(input=[previous_input, previous_hidden_state]) - hidden_state = simple_lstm( - input=current_input, size=128, lstm_cell_attr=ExtraAttr(drop_rate=0.1)) - previous_input, previous_hidden_state = current_input, hidden_state - -lstm = previous_hidden_state - -lstm_last = pooling_layer(input=lstm, pooling_type=MaxPooling()) -output = fc_layer( - input=lstm_last, size=2, bias_attr=bias_attr, act=SoftmaxActivation()) - -if is_predict: - maxid = maxid_layer(output) - outputs([maxid, output]) -else: - label = data_layer(name="label", size=2) - cls = classification_cost(input=output, label=label) - outputs(cls) diff --git a/v1_api_demo/sequence_tagging/data/get_data.sh b/v1_api_demo/sequence_tagging/data/get_data.sh deleted file mode 100755 index 0cdb394035e782b3a647f7f13e79d55b5d3dff48..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/data/get_data.sh +++ /dev/null @@ -1,21 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -cd $DIR - -wget http://www.cnts.ua.ac.be/conll2000/chunking/train.txt.gz -wget http://www.cnts.ua.ac.be/conll2000/chunking/test.txt.gz diff --git a/v1_api_demo/sequence_tagging/data/test.list b/v1_api_demo/sequence_tagging/data/test.list deleted file mode 100644 index 073c0a0c9063ac55f762ac261746aa73057d70e8..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/data/test.list +++ /dev/null @@ -1 +0,0 @@ -data/test.txt.gz diff --git a/v1_api_demo/sequence_tagging/data/train.list b/v1_api_demo/sequence_tagging/data/train.list deleted file mode 100644 index 43c24d5f6484a90fe883ad5516fe100d27c9ce47..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/data/train.list +++ /dev/null @@ -1 +0,0 @@ -data/train.txt.gz diff --git a/v1_api_demo/sequence_tagging/dataprovider.py b/v1_api_demo/sequence_tagging/dataprovider.py deleted file mode 100644 index bb4b4465bc7e032c50c1d21263651e2578af67be..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/dataprovider.py +++ /dev/null @@ -1,260 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer.PyDataProvider2 import * -import gzip -import logging - -logging.basicConfig( - format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s', ) -logger = logging.getLogger('paddle') -logger.setLevel(logging.INFO) - -OOV_POLICY_IGNORE = 0 -OOV_POLICY_USE = 1 -OOV_POLICY_ERROR = 2 - -num_original_columns = 3 - -# Feature combination patterns. -# [[-1,0], [0,0]] means previous token at column 0 and current token at -# column 0 are combined as one feature. -patterns = [ - [[-2, 0]], - [[-1, 0]], - [[0, 0]], - [[1, 0]], - [[2, 0]], - [[-1, 0], [0, 0]], - [[0, 0], [1, 0]], - [[-2, 1]], - [[-1, 1]], - [[0, 1]], - [[1, 1]], - [[2, 1]], - [[-2, 1], [-1, 1]], - [[-1, 1], [0, 1]], - [[0, 1], [1, 1]], - [[1, 1], [2, 1]], - [[-2, 1], [-1, 1], [0, 1]], - [[-1, 1], [0, 1], [1, 1]], - [[0, 1], [1, 1], [2, 1]], -] - -dict_label = { - 'B-ADJP': 0, - 'I-ADJP': 1, - 'B-ADVP': 2, - 'I-ADVP': 3, - 'B-CONJP': 4, - 'I-CONJP': 5, - 'B-INTJ': 6, - 'I-INTJ': 7, - 'B-LST': 8, - 'I-LST': 9, - 'B-NP': 10, - 'I-NP': 11, - 'B-PP': 12, - 'I-PP': 13, - 'B-PRT': 14, - 'I-PRT': 15, - 'B-SBAR': 16, - 'I-SBAR': 17, - 'B-UCP': 18, - 'I-UCP': 19, - 'B-VP': 20, - 'I-VP': 21, - 'O': 22 -} - - -def make_features(sequence): - length = len(sequence) - num_features = len(sequence[0]) - - def get_features(pos): - if pos < 0: - return ['#B%s' % -pos] * num_features - if pos >= length: - return ['#E%s' % (pos - length + 1)] * num_features - return sequence[pos] - - for i in xrange(length): - for pattern in patterns: - fname = '/'.join([get_features(i + pos)[f] for pos, f in pattern]) - sequence[i].append(fname) - - -''' -Source file format: -Each line is for one timestep. The features are separated by space. -An empty line indicates end of a sequence. - -cutoff: a list of numbers. If count of a feature is smaller than this, - it will be ignored. -if oov_policy[i] is OOV_POLICY_USE, id 0 is reserved for OOV features of -i-th column. - -return a list of dict for each column -''' - - -def create_dictionaries(filename, cutoff, oov_policy): - def add_to_dict(sequence, dicts): - num_features = len(dicts) - for features in sequence: - l = len(features) - assert l == num_features, "Wrong number of features " + line - for i in xrange(l): - if features[i] in dicts[i]: - dicts[i][features[i]] += 1 - else: - dicts[i][features[i]] = 1 - - num_features = len(cutoff) - dicts = [] - for i in xrange(num_features): - dicts.append(dict()) - - f = gzip.open(filename, 'rb') - - sequence = [] - - for line in f: - line = line.strip() - if not line: - make_features(sequence) - add_to_dict(sequence, dicts) - sequence = [] - continue - features = line.split(' ') - sequence.append(features) - - for i in xrange(num_features): - dct = dicts[i] - n = 1 if oov_policy[i] == OOV_POLICY_USE else 0 - todo = [] - for k, v in dct.iteritems(): - if v < cutoff[i]: - todo.append(k) - else: - dct[k] = n - n += 1 - - if oov_policy[i] == OOV_POLICY_USE: - # placeholder so that len(dct) will be the number of features - # including OOV - dct['#OOV#'] = 0 - - logger.info('column %d dict size=%d, ignored %d' % (i, n, len(todo))) - for k in todo: - del dct[k] - - f.close() - return dicts - - -def initializer(settings, **xargs): - cutoff = [3, 1, 0] - cutoff += [3] * len(patterns) - oov_policy = [OOV_POLICY_IGNORE, OOV_POLICY_ERROR, OOV_POLICY_ERROR] - oov_policy += [OOV_POLICY_IGNORE] * len(patterns) - dicts = create_dictionaries('data/train.txt.gz', cutoff, oov_policy) - dicts[2] = dict_label - settings.dicts = dicts - settings.oov_policy = oov_policy - input_types = [] - num_features = len(dicts) - for i in xrange(num_original_columns): - input_types.append(integer_sequence(len(dicts[i]))) - logger.info("slot %s size=%s" % (i, len(dicts[i]))) - if patterns: - dim = 0 - for i in xrange(num_original_columns, num_features): - dim += len(dicts[i]) - input_types.append(sparse_binary_vector_sequence(dim)) - logger.info("feature size=%s" % dim) - settings.input_types = input_types - - -''' -if oov_policy[i] == OOV_POLICY_USE, features in i-th column which are not -existed in dicts[i] will be assigned to id 0. -if oov_policy[i] == OOV_POLICY_ERROR, all features in i-th column MUST exist -in dicts[i]. -''' - - -@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM) -def process(settings, filename): - input_file = filename - dicts = settings.dicts - oov_policy = settings.oov_policy - - def gen_sample(sequence): - num_features = len(dicts) - sample = [list() for i in xrange(num_original_columns)] - if patterns: - sample.append([]) - for features in sequence: - assert len(features) == num_features, \ - "Wrong number of features: " + line - for i in xrange(num_original_columns): - id = dicts[i].get(features[i], -1) - if id != -1: - sample[i].append(id) - elif oov_policy[i] == OOV_POLICY_IGNORE: - sample[i].append(0xffffffff) - elif oov_policy[i] == OOV_POLICY_ERROR: - logger.fatal("Unknown token: %s" % features[i]) - else: - sample[i].append(0) - - if patterns: - dim = 0 - vec = [] - for i in xrange(num_original_columns, num_features): - id = dicts[i].get(features[i], -1) - if id != -1: - vec.append(dim + id) - elif oov_policy[i] == OOV_POLICY_IGNORE: - pass - elif oov_policy[i] == OOV_POLICY_ERROR: - logger.fatal("Unknown token: %s" % features[i]) - else: - vec.ids.append(dim + 0) - - dim += len(dicts[i]) - sample[-1].append(vec) - return sample - - num_features = len(dicts) - f = gzip.open(input_file, 'rb') - - num_sequences = 0 - sequence = [] - for line in f: - line = line.strip() - if not line: - make_features(sequence) - yield gen_sample(sequence) - sequence = [] - num_sequences += 1 - continue - features = line.split(' ') - sequence.append(features) - - f.close() - - logger.info("num_sequences=%s" % num_sequences) diff --git a/v1_api_demo/sequence_tagging/linear_crf.py b/v1_api_demo/sequence_tagging/linear_crf.py deleted file mode 100644 index ea012ba1ae9c790ccefd3dd5f066aa92202128a2..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/linear_crf.py +++ /dev/null @@ -1,83 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -import math - -define_py_data_sources2( - train_list="data/train.list", - test_list="data/test.list", - module="dataprovider", - obj="process") - -batch_size = 1 -settings( - learning_method=MomentumOptimizer(), - batch_size=batch_size, - regularization=L2Regularization(batch_size * 1e-4), - model_average=ModelAverage(0.5), - learning_rate=1e-1, - learning_rate_decay_a=1e-5, - learning_rate_decay_b=0.25, ) - -num_label_types = 23 - - -def get_simd_size(size): - return int(math.ceil(float(size) / 8)) * 8 - - -# Currently, in order to use sparse_update=True, -# the size has to be aligned. -num_label_types = get_simd_size(num_label_types) - -features = data_layer(name="features", size=76328) -word = data_layer(name="word", size=6778) -pos = data_layer(name="pos", size=44) -chunk = data_layer(name="chunk", size=num_label_types) - -crf_input = fc_layer( - input=features, - size=num_label_types, - act=LinearActivation(), - bias_attr=False, - param_attr=ParamAttr( - initial_std=0, sparse_update=True)) - -crf = crf_layer( - input=crf_input, - label=chunk, - param_attr=ParamAttr( - name="crfw", initial_std=0), ) - -crf_decoding = crf_decoding_layer( - size=num_label_types, - input=crf_input, - label=chunk, - param_attr=ParamAttr(name="crfw"), ) - -sum_evaluator( - name="error", - input=crf_decoding, ) - -chunk_evaluator( - name="chunk_f1", - input=crf_decoding, - label=chunk, - chunk_scheme="IOB", - num_chunk_types=11, ) - -inputs(word, pos, chunk, features) -outputs(crf) diff --git a/v1_api_demo/sequence_tagging/readme.md b/v1_api_demo/sequence_tagging/readme.md deleted file mode 100644 index 2e17fffb83c532f5e5fec1227f169c97c1f20e22..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/readme.md +++ /dev/null @@ -1,45 +0,0 @@ -# Sequence Tagging - -This demo is a sequence model for assigning tags to each token in a sentence. The task is described at CONLL2000 Text Chunking task. - -## Download data -```bash -cd demo/sequence_tagging -./data/get_data.sh -``` - -## Train model -```bash -cd demo/sequence_tagging -./train.sh -``` - -## Model description - -We provide two models. One is a linear CRF model (linear_crf.py) with is equivalent to the one at leon.bottou.org/projects/sgd. The second one is a stacked bidirectional RNN and CRF model (rnn_crf.py). -
- - - - - - - - - - - - - - - - - - - - - - -
Model nameNumber of parametersF1 score
linear_crf 1.8M 0.937
rnn_crf 960K 0.941
-
-
diff --git a/v1_api_demo/sequence_tagging/rnn_crf.py b/v1_api_demo/sequence_tagging/rnn_crf.py deleted file mode 100644 index 937a34df103663ecf0f0827bbfb9d82823c9b902..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/rnn_crf.py +++ /dev/null @@ -1,121 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * - -import math - -define_py_data_sources2( - train_list="data/train.list", - test_list="data/test.list", - module="dataprovider", - obj="process") - -batch_size = 16 -settings( - learning_method=MomentumOptimizer(), - batch_size=batch_size, - regularization=L2Regularization(batch_size * 1e-5), - model_average=ModelAverage(0.5), - learning_rate=2e-3, - learning_rate_decay_a=5e-7, - learning_rate_decay_b=0.5, ) - -word_dim = 128 -hidden_dim = 128 -with_rnn = True - -initial_std = 1 / math.sqrt(hidden_dim) -param_attr = ParamAttr(initial_std=initial_std) -cpu_layer_attr = ExtraLayerAttribute(device=-1) - -default_device(0) - -num_label_types = 23 - -features = data_layer(name="features", size=76328) -word = data_layer(name="word", size=6778) -pos = data_layer(name="pos", size=44) -chunk = data_layer( - name="chunk", size=num_label_types, layer_attr=cpu_layer_attr) - -emb = embedding_layer( - input=word, size=word_dim, param_attr=ParamAttr(initial_std=0)) - -hidden1 = mixed_layer( - size=hidden_dim, - act=STanhActivation(), - bias_attr=True, - input=[ - full_matrix_projection(emb), table_projection( - pos, param_attr=param_attr) - ]) - -if with_rnn: - rnn1 = recurrent_layer( - act=ReluActivation(), - bias_attr=True, - input=hidden1, - param_attr=ParamAttr(initial_std=0), ) - -hidden2 = mixed_layer( - size=hidden_dim, - act=STanhActivation(), - bias_attr=True, - input=[full_matrix_projection(hidden1)] + - ([full_matrix_projection( - rnn1, param_attr=ParamAttr(initial_std=0))] if with_rnn else []), ) - -if with_rnn: - rnn2 = recurrent_layer( - reverse=True, - act=ReluActivation(), - bias_attr=True, - input=hidden2, - param_attr=ParamAttr(initial_std=0), ) - -crf_input = mixed_layer( - size=num_label_types, - bias_attr=False, - input=[full_matrix_projection(hidden2), ] + - ([full_matrix_projection( - rnn2, param_attr=ParamAttr(initial_std=0))] if with_rnn else []), ) - -crf = crf_layer( - input=crf_input, - label=chunk, - param_attr=ParamAttr( - name="crfw", initial_std=0), - layer_attr=cpu_layer_attr, ) - -crf_decoding = crf_decoding_layer( - size=num_label_types, - input=crf_input, - label=chunk, - param_attr=ParamAttr(name="crfw"), - layer_attr=cpu_layer_attr, ) - -sum_evaluator( - name="error", - input=crf_decoding, ) - -chunk_evaluator( - name="chunk_f1", - input=crf_decoding, - label=chunk, - chunk_scheme="IOB", - num_chunk_types=11, ) - -inputs(word, pos, chunk, features) -outputs(crf) diff --git a/v1_api_demo/sequence_tagging/train.sh b/v1_api_demo/sequence_tagging/train.sh deleted file mode 100755 index 37e196c84200dc26ccb523076a81dbc393b1280f..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/train.sh +++ /dev/null @@ -1,12 +0,0 @@ -#!/bin/bash - -paddle train \ - --config rnn_crf.py \ - --parallel_nn=1 \ - --use_gpu=1 \ - --dot_period=10 \ - --log_period=1000 \ - --test_period=0 \ - --num_passes=10 \ -2>&1 | tee 'train.log' -paddle usage -l 'train.log' -e $? -n "sequence_tagging_train" >/dev/null 2>&1 diff --git a/v1_api_demo/sequence_tagging/train_linear.sh b/v1_api_demo/sequence_tagging/train_linear.sh deleted file mode 100755 index ad6e2d8ee7f813c69f9dd250c6f7bbb4403a0ed5..0000000000000000000000000000000000000000 --- a/v1_api_demo/sequence_tagging/train_linear.sh +++ /dev/null @@ -1,11 +0,0 @@ -#!/bin/bash - -paddle train \ - --config linear_crf.py \ - --use_gpu=0 \ - --dot_period=100 \ - --log_period=10000 \ - --test_period=0 \ - --num_passes=10 -2>&1 | tee 'train_linear.log' -paddle usage -l 'train_linear.log' -e $? -n "sequence_tagging_train_linear" >/dev/null 2>&1 diff --git a/v1_api_demo/traffic_prediction/README b/v1_api_demo/traffic_prediction/README deleted file mode 100644 index 4c95188583513c332b7d7cb0a32d59336208e1aa..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/README +++ /dev/null @@ -1,7 +0,0 @@ -run by: -cd ./data -sh get_data.sh -cd .. -sh train.sh -sh predict.sh - diff --git a/v1_api_demo/traffic_prediction/data/get_data.sh b/v1_api_demo/traffic_prediction/data/get_data.sh deleted file mode 100755 index f2fa548d4709c0361334f117bfb49e18d83c32f4..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/data/get_data.sh +++ /dev/null @@ -1,34 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -set -e -set -x - -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -cd $DIR - -#download the dataset -echo "Downloading traffic data..." -wget http://paddlepaddle.cdn.bcebos.com/demo/traffic/traffic_data.tar.gz - -#extract package -echo "Unzipping..." -tar -zxvf traffic_data.tar.gz - -echo "data/speeds.csv" > train.list -echo "data/speeds.csv" > test.list -echo "data/speeds.csv" > pred.list - -echo "Done." diff --git a/v1_api_demo/traffic_prediction/dataprovider.py b/v1_api_demo/traffic_prediction/dataprovider.py deleted file mode 100644 index c7883b6950c369ee67c39b80ce1cefbbf9350459..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/dataprovider.py +++ /dev/null @@ -1,82 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer.PyDataProvider2 import * -import sys -import numpy as np -TERM_NUM = 24 -FORECASTING_NUM = 24 -LABEL_VALUE_NUM = 4 - - -def initHook(settings, file_list, **kwargs): - """ - Init hook is invoked before process data. It will set obj.slots and store data meta. - - :param settings: global object. It will passed to process routine. - :type obj: object - :param file_list: the meta file object, which passed from trainer_config.py,but unused in this function. - :param kwargs: unused other arguments. - """ - del kwargs #unused - - settings.pool_size = sys.maxint - #Use a time seires of the past as feature. - #Dense_vector's expression form is [float,float,...,float] - settings.input_types = [dense_vector(TERM_NUM)] - #There are next FORECASTING_NUM fragments you need predict. - #Every predicted condition at time point has four states. - for i in range(FORECASTING_NUM): - settings.input_types.append(integer_value(LABEL_VALUE_NUM)) - - -@provider( - init_hook=initHook, cache=CacheType.CACHE_PASS_IN_MEM, should_shuffle=True) -def process(settings, file_name): - with open(file_name) as f: - #abandon fields name - f.next() - for row_num, line in enumerate(f): - speeds = map(int, line.rstrip('\r\n').split(",")[1:]) - # Get the max index. - end_time = len(speeds) - # Scanning and generating samples - for i in range(TERM_NUM, end_time - FORECASTING_NUM): - # For dense slot - pre_spd = map(float, speeds[i - TERM_NUM:i]) - - # Integer value need predicting, values start from 0, so every one minus 1. - fol_spd = [j - 1 for j in speeds[i:i + FORECASTING_NUM]] - - # Predicting label is missing, abandon the sample. - if -1 in fol_spd: - continue - yield [pre_spd] + fol_spd - - -def predict_initHook(settings, file_list, **kwargs): - settings.pool_size = sys.maxint - settings.input_types = [dense_vector(TERM_NUM)] - - -@provider(init_hook=predict_initHook, should_shuffle=False) -def process_predict(settings, file_name): - with open(file_name) as f: - #abandon fields name - f.next() - for row_num, line in enumerate(f): - speeds = map(int, line.rstrip('\r\n').split(",")) - end_time = len(speeds) - pre_spd = map(float, speeds[end_time - TERM_NUM:end_time]) - yield pre_spd diff --git a/v1_api_demo/traffic_prediction/gen_result.py b/v1_api_demo/traffic_prediction/gen_result.py deleted file mode 100644 index 3da70b30315f863fd3582583e9a29540a09c1e7f..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/gen_result.py +++ /dev/null @@ -1,61 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -res = [] -with open('./rank-00000') as f: - for line in f: - pred = map(int, line.strip('\r\n;').split(";")) - #raw prediction range from 0 to 3 - res.append([i + 1 for i in pred]) - -file_name = open('./data/pred.list').read().strip('\r\n') - -FORECASTING_NUM = 24 -header = [ - 'id', - '201604200805', - '201604200810', - '201604200815', - '201604200820', - '201604200825', - '201604200830', - '201604200835', - '201604200840', - '201604200845', - '201604200850', - '201604200855', - '201604200900', - '201604200905', - '201604200910', - '201604200915', - '201604200920', - '201604200925', - '201604200930', - '201604200935', - '201604200940', - '201604200945', - '201604200950', - '201604200955', - '201604201000', -] -################### -## To CSV format ## -################### -with open(file_name) as f: - f.next() - print ','.join(header) - for row_num, line in enumerate(f): - fields = line.rstrip('\r\n').split(',') - linkid = fields[0] - print linkid + ',' + ','.join(map(str, res[row_num])) diff --git a/v1_api_demo/traffic_prediction/predict.sh b/v1_api_demo/traffic_prediction/predict.sh deleted file mode 100755 index 2dbd5e8805dd97d35c7d58917f8ec6b5033bda03..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/predict.sh +++ /dev/null @@ -1,30 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -cfg=trainer_config.py -# pass choice -model="output/pass-00000" -paddle train \ - --config=$cfg \ - --use_gpu=false \ - --job=test \ - --init_model_path=$model \ - --config_args=is_predict=1 \ - --predict_output_dir=. - -python gen_result.py > result.csv - -rm -rf rank-00000 diff --git a/v1_api_demo/traffic_prediction/train.sh b/v1_api_demo/traffic_prediction/train.sh deleted file mode 100755 index 48dfc5604f80042598c5c779bd450a5808fdfb64..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/train.sh +++ /dev/null @@ -1,27 +0,0 @@ -#!/bin/bash -# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -set -e - -cfg=trainer_config.py -paddle train \ - --config=$cfg \ - --save_dir=./output \ - --trainer_count=4 \ - --log_period=1000 \ - --dot_period=10 \ - --num_passes=10 \ - --use_gpu=false \ - --show_parameter_stats_period=3000 \ - 2>&1 | tee 'train.log' diff --git a/v1_api_demo/traffic_prediction/trainer_config.py b/v1_api_demo/traffic_prediction/trainer_config.py deleted file mode 100755 index 52d678624aff7ca2264c3c20e320004217d14397..0000000000000000000000000000000000000000 --- a/v1_api_demo/traffic_prediction/trainer_config.py +++ /dev/null @@ -1,52 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -from paddle.trainer_config_helpers import * - -################################### DATA Configuration ############################################# -is_predict = get_config_arg('is_predict', bool, False) -trn = './data/train.list' if not is_predict else None -tst = './data/test.list' if not is_predict else './data/pred.list' -process = 'process' if not is_predict else 'process_predict' -define_py_data_sources2( - train_list=trn, test_list=tst, module="dataprovider", obj=process) -################################### Parameter Configuaration ####################################### -TERM_NUM = 24 -FORECASTING_NUM = 24 -emb_size = 16 -batch_size = 128 if not is_predict else 1 -settings( - batch_size=batch_size, - learning_rate=1e-3, - learning_method=RMSPropOptimizer()) -################################### Algorithm Configuration ######################################## - -output_label = [] - -link_encode = data_layer(name='link_encode', size=TERM_NUM) -for i in xrange(FORECASTING_NUM): - # Each task share same weight. - link_param = ParamAttr( - name='_link_vec.w', initial_max=1.0, initial_min=-1.0) - link_vec = fc_layer(input=link_encode, size=emb_size, param_attr=link_param) - score = fc_layer(input=link_vec, size=4, act=SoftmaxActivation()) - if is_predict: - maxid = maxid_layer(score) - output_label.append(maxid) - else: - # Multi-task training. - label = data_layer(name='label_%dmin' % ((i + 1) * 5), size=4) - cls = classification_cost( - input=score, name="cost_%dmin" % ((i + 1) * 5), label=label) - output_label.append(cls) -outputs(output_label) diff --git a/v1_api_demo/vae/README.md b/v1_api_demo/vae/README.md deleted file mode 100644 index e55d483b023773900729622a6cac44116fc79c76..0000000000000000000000000000000000000000 --- a/v1_api_demo/vae/README.md +++ /dev/null @@ -1,13 +0,0 @@ -#Variational Autoencoder (VAE) - -This demo implements VAE training described in the original paper (https://arxiv.org/abs/1312.6114). - - -In order to run the model, first download the MNIST dataset by running the shell script in ./data. - -Then you can run the command below. The flag --useGpu specifies whether to use gpu for training (0 is cpu, 1 is gpu). - -$python vae_train.py [--use_gpu 1] - -The generated images will be stored in ./samples/ -The corresponding models will be stored in ./params/ diff --git a/v1_api_demo/vae/data/get_mnist_data.sh b/v1_api_demo/vae/data/get_mnist_data.sh deleted file mode 100755 index a77c81bf5af9ddb6634ff89460797ca543c5e517..0000000000000000000000000000000000000000 --- a/v1_api_demo/vae/data/get_mnist_data.sh +++ /dev/null @@ -1,17 +0,0 @@ -#!/usr/bin/env sh -# This script downloads the mnist data and unzips it. -set -e -DIR="$( cd "$(dirname "$0")" ; pwd -P )" -rm -rf "$DIR/mnist_data" -mkdir "$DIR/mnist_data" -cd "$DIR/mnist_data" - -echo "Downloading..." - -for fname in train-images-idx3-ubyte train-labels-idx1-ubyte t10k-images-idx3-ubyte t10k-labels-idx1-ubyte -do - if [ ! -e $fname ]; then - wget --no-check-certificate http://yann.lecun.com/exdb/mnist/${fname}.gz - gunzip ${fname}.gz - fi -done diff --git a/v1_api_demo/vae/dataloader.py b/v1_api_demo/vae/dataloader.py deleted file mode 100644 index e9ff95d44f825cd941b5687f754618e66d491e7f..0000000000000000000000000000000000000000 --- a/v1_api_demo/vae/dataloader.py +++ /dev/null @@ -1,60 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import numpy as np - - -class MNISTloader(): - def __init__(self, - data_path="./data/mnist_data/", - batch_size=60, - process='train'): - self.batch_size = batch_size - self.data_path = data_path - self._pointer = 0 - self.image_batches = np.array([]) - self.process = process - - def _extract_images(self, filename, n): - f = open(filename, 'rb') - f.read(16) - data = np.fromfile(f, 'ubyte', count=n * 28 * 28).reshape((n, 28 * 28)) - #Mapping data into [-1, 1] - data = data / 255. * 2. - 1 - data_batches = np.split(data, 60000 / self.batch_size, 0) - - f.close() - - return data_batches - - @property - def pointer(self): - return self._pointer - - def load_data(self): - TRAIN_IMAGES = '%s/train-images-idx3-ubyte' % self.data_path - TEST_IMAGES = '%s/t10k-images-idx3-ubyte' % self.data_path - - if self.process == 'train': - self.image_batches = self._extract_images(TRAIN_IMAGES, 60000) - else: - self.image_batches = self._extract_images(TEST_IMAGES, 10000) - - def next_batch(self): - batch = self.image_batches[self._pointer] - self._pointer = (self._pointer + 1) % (60000 / self.batch_size) - return np.array(batch) - - def reset_pointer(self): - self._pointer = 0 diff --git a/v1_api_demo/vae/vae_conf.py b/v1_api_demo/vae/vae_conf.py deleted file mode 100644 index 301dd23793d19ec5946cc7bb07e32c53c04a972b..0000000000000000000000000000000000000000 --- a/v1_api_demo/vae/vae_conf.py +++ /dev/null @@ -1,116 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -from paddle.trainer_config_helpers import * -import numpy as np - -is_generating = get_config_arg("is_generating", bool, False) - -settings(batch_size=32, learning_rate=1e-3, learning_method=AdamOptimizer()) - -X_dim = 28 * 28 -h_dim = 128 -z_dim = 100 - - -def reparameterization(mu, logvar): - eps = ParamAttr(initial_mean=0., initial_std=1) - with mixed_layer() as sigma: - sigma += dotmul_projection(layer_math.exp(logvar) * 0.5, param_attr=eps) - return mu + sigma - - -def q_func(X): - """ - xavier initialization - """ - param_attr = ParamAttr( - name='share.w', initial_mean=0., initial_std=1. / np.sqrt(X_dim / 2.)) - mu_param = ParamAttr( - name='mu.w', initial_mean=0., initial_std=1. / np.sqrt(h_dim / 2.)) - logvar_param = ParamAttr( - name='logvar.w', initial_mean=0., initial_std=1. / np.sqrt(h_dim / 2.)) - - bias_attr = ParamAttr(name='share.bias', initial_mean=0., initial_std=0.) - mu_bias = ParamAttr(name='mu.bias', initial_mean=0., initial_std=0.) - logvar_bias = ParamAttr(name='logvar.bias', initial_mean=0., initial_std=0.) - - share_layer = fc_layer( - X, - size=h_dim, - param_attr=param_attr, - bias_attr=bias_attr, - act=ReluActivation()) - - return (fc_layer( - share_layer, - size=z_dim, - param_attr=mu_param, - bias_attr=mu_bias, - act=LinearActivation()), fc_layer( - share_layer, - size=z_dim, - param_attr=logvar_param, - bias_attr=logvar_bias, - act=LinearActivation())) - - -def generator(z): - - hidden_param = ParamAttr( - name='hidden.w', initial_mean=0., initial_std=1. / np.sqrt(z_dim / 2.)) - hidden_bias = ParamAttr(name='hidden.bias', initial_mean=0., initial_std=0.) - prob_param = ParamAttr( - name='prob.w', initial_mean=0., initial_std=1. / np.sqrt(h_dim / 2.)) - prob_bias = ParamAttr(name='prob.bias', initial_mean=0., initial_std=0.) - - hidden_layer = fc_layer( - z, - size=h_dim, - act=ReluActivation(), - param_attr=hidden_param, - bias_attr=hidden_bias) - prob = fc_layer( - hidden_layer, - size=X_dim, - act=SigmoidActivation(), - param_attr=prob_param, - bias_attr=prob_bias) - - return prob - - -def reconstruct_error(prob, X): - cost = multi_binary_label_cross_entropy(input=prob, label=X) - return cost - - -def KL_loss(mu, logvar): - with mixed_layer() as mu_square: - mu_square += dotmul_operator(mu, mu, scale=1.) - - cost = 0.5 * sum_cost(layer_math.exp(logvar) + mu_square - 1. - logvar) - - return cost - - -if not is_generating: - x_batch = data_layer(name='x_batch', size=X_dim) - mu, logvar = q_func(x_batch) - z_samples = reparameterization(mu, logvar) - prob = generator(z_samples) - outputs(reconstruct_error(prob, x_batch) + KL_loss(mu, logvar)) -else: - z_samples = data_layer(name='noise', size=z_dim) - outputs(generator(z_samples)) diff --git a/v1_api_demo/vae/vae_train.py b/v1_api_demo/vae/vae_train.py deleted file mode 100644 index 1babb011c77b92861cc680a2e1aaa8c9ae5d97b5..0000000000000000000000000000000000000000 --- a/v1_api_demo/vae/vae_train.py +++ /dev/null @@ -1,175 +0,0 @@ -# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -import argparse -import random -import numpy as np -import cPickle -import sys, os -from PIL import Image - -from paddle.trainer.config_parser import parse_config -from paddle.trainer.config_parser import logger -import py_paddle.swig_paddle as api -import dataloader -import matplotlib.pyplot as plt - - -def plot_samples(samples): - fig = plt.figure(figsize=(4, 4)) - gs = gridspec.GridSpec(4, 4) - gs.update(wspace=0.05, hspace=0.05) - for i, sample in enumerate(samples): - plt.subplot(gs[i]) - plt.axis('off') - plt.imshow(sample.reshape(28, 28), cmap='Greys_r') - - return fig - - -def CHECK_EQ(a, b): - assert a == b, "a=%s, b=%s" % (a, b) - - -def get_fake_samples(generator_machine, batch_size, noise): - gen_inputs = api.Arguments.createArguments(1) - gen_inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(noise)) - gen_outputs = api.Arguments.createArguments(0) - generator_machine.forward(gen_inputs, gen_outputs, api.PASS_TEST) - fake_samples = gen_outputs.getSlotValue(0).copyToNumpyMat() - return fake_samples - - -def copy_shared_parameters(src, dst): - ''' - copy the parameters from src to dst - :param src: the source of the parameters - :type src: GradientMachine - :param dst: the destination of the parameters - :type dst: GradientMachine - ''' - src_params = [src.getParameter(i) for i in xrange(src.getParameterSize())] - src_params = dict([(p.getName(), p) for p in src_params]) - - for i in xrange(dst.getParameterSize()): - dst_param = dst.getParameter(i) - src_param = src_params.get(dst_param.getName(), None) - if src_param is None: - continue - src_value = src_param.getBuf(api.PARAMETER_VALUE) - dst_value = dst_param.getBuf(api.PARAMETER_VALUE) - CHECK_EQ(len(src_value), len(dst_value)) - dst_value.copyFrom(src_value) - dst_param.setValueUpdated() - - -def find(iterable, cond): - for item in iterable: - if cond(item): - return item - return None - - -def get_layer_size(model_conf, layer_name): - layer_conf = find(model_conf.layers, lambda x: x.name == layer_name) - assert layer_conf is not None, "Cannot find '%s' layer" % layer_name - return layer_conf.size - - -def main(): - parser = argparse.ArgumentParser() - parser.add_argument( - "--use_gpu", default="1", help="1 means use gpu for training") - parser.add_argument("--gpu_id", default="0", help="the gpu_id parameter") - args = parser.parse_args() - use_gpu = args.use_gpu - assert use_gpu in ["0", "1"] - - if not os.path.exists("./samples/"): - os.makedirs("./samples/") - - if not os.path.exists("./params/"): - os.makedirs("./params/") - - api.initPaddle('--use_gpu=' + use_gpu, '--dot_period=10', - '--log_period=1000', '--gpu_id=' + args.gpu_id, - '--save_dir=' + "./params/") - - conf = "vae_conf.py" - - trainer_conf = parse_config(conf, "is_generating=False") - gener_conf = parse_config(conf, "is_generating=True") - - batch_size = trainer_conf.opt_config.batch_size - - noise_dim = get_layer_size(gener_conf.model_config, "noise") - - mnist = dataloader.MNISTloader(batch_size=batch_size) - mnist.load_data() - - training_machine = api.GradientMachine.createFromConfigProto( - trainer_conf.model_config) - - generator_machine = api.GradientMachine.createFromConfigProto( - gener_conf.model_config) - - trainer = api.Trainer.create(trainer_conf, training_machine) - - trainer.startTrain() - - for train_pass in xrange(100): - trainer.startTrainPass() - mnist.reset_pointer() - i = 0 - it = 0 - while mnist.pointer != 0 or i == 0: - X = mnist.next_batch().astype('float32') - - inputs = api.Arguments.createArguments(1) - inputs.setSlotValue(0, api.Matrix.createDenseFromNumpy(X)) - - trainer.trainOneDataBatch(batch_size, inputs) - - if it % 1000 == 0: - - outputs = api.Arguments.createArguments(0) - training_machine.forward(inputs, outputs, api.PASS_TEST) - loss = np.mean(outputs.getSlotValue(0).copyToNumpyMat()) - print "\niter: {}".format(str(it).zfill(3)) - print "VAE loss: {}".format(str(loss).zfill(3)) - - #Sync parameters between networks (GradientMachine) at the beginning - copy_shared_parameters(training_machine, generator_machine) - - z_samples = np.random.randn(batch_size, - noise_dim).astype('float32') - samples = get_fake_samples(generator_machine, batch_size, - z_samples) - - #Generating the first 16 images for a picture. - figure = plot_samples(samples[:16]) - plt.savefig( - "./samples/{}_{}.png".format( - str(train_pass).zfill(3), str(i).zfill(3)), - bbox_inches='tight') - plt.close(figure) - i += 1 - it += 1 - - trainer.finishTrainPass() - trainer.finishTrain() - - -if __name__ == '__main__': - main()