Skip to content

P
Paddle

Crayon鑫 / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
提交 021b3a44 编写于 作者: H hohdiy 提交者： GitHub
浏览文件
操作

Merge pull request #1 from PaddlePaddle/develop 

merge PaddlePaddle/Paddle
上级 4f1bf30d 17f7125b
显示空白变更内容
内联 并排
Showing with 10214 addition and 5079 deletion
.dockerignore 0 → 120000
浏览文件 @ 021b3a44
.gitignore
\ No newline at end of file
.gitignore
浏览文件 @ 021b3a44
...@@ -8,3 +8,4 @@ build/ ...@@ -8,3 +8,4 @@ build/
.cproject .cproject
.pydevproject .pydevproject
Makefile Makefile
.test_env/
CMakeLists.txt
浏览文件 @ 021b3a44
cmake_minimum_required(VERSION 2.8) cmake_minimum_required(VERSION 2.8)
project(paddle CXX C) project(paddle CXX C)
set(PADDLE_MAJOR_VERSION 0)
set(PADDLE_MINOR_VERSION 9)
set(PADDLE_PATCH_VERSION 0a0)
set(PADDLE_VERSION ${PADDLE_MAJOR_VERSION}.${PADDLE_MINOR_VERSION}.${PADDLE_PATCH_VERSION})
set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake") set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake")
set(PROJ_ROOT ${CMAKE_SOURCE_DIR}) set(PROJ_ROOT ${CMAKE_SOURCE_DIR})
...@@ -12,6 +8,17 @@ include(package) ...@@ -12,6 +8,17 @@ include(package)
find_package(SWIG 2.0) find_package(SWIG 2.0)
find_package(CUDA QUIET) find_package(CUDA QUIET)
find_package(Protobuf REQUIRED) find_package(Protobuf REQUIRED)
# Check protobuf library version.
execute_process(COMMAND ${PROTOBUF_PROTOC_EXECUTABLE} --version
OUTPUT_VARIABLE PROTOBUF_VERSION)
string(REPLACE "libprotoc " "" PROTOBUF_VERSION ${PROTOBUF_VERSION})
set(PROTOBUF_3 OFF)
if (${PROTOBUF_VERSION} VERSION_GREATER "3.0.0" OR ${PROTOBUF_VERSION} VERSION_EQUAL "3.0.0")
set(PROTOBUF_3 ON)
endif()
find_package(PythonLibs 2.7 REQUIRED) find_package(PythonLibs 2.7 REQUIRED)
find_package(PythonInterp 2.7 REQUIRED) find_package(PythonInterp 2.7 REQUIRED)
find_package(ZLIB REQUIRED) find_package(ZLIB REQUIRED)
...@@ -64,31 +71,11 @@ include(check_packages) ...@@ -64,31 +71,11 @@ include(check_packages)
include(swig) include(swig)
include(coveralls) include(coveralls)
# add PaddlePaddle version # Set PaddlePaddle version to Git tag name or Git commit ID.
if(DEFINED ENV{PADDLE_VERSION}) find_package(Git REQUIRED)
add_definitions(-DPADDLE_VERSION=\"$ENV{PADDLE_VERSION}\") # version.cmake will get the current PADDLE_VERSION
else() include(version)
if(EXISTS ${PROJ_ROOT}/.svn/) add_definitions(-DPADDLE_VERSION=\"${PADDLE_VERSION}\")
find_package(Subversion REQUIRED)
if(SUBVERSION_FOUND)
Subversion_WC_INFO(${PROJ_ROOT} Project)
add_definitions(-DPADDLE_VERSION=${Project_WC_REVISION})
endif()
elseif(EXISTS ${PROJ_ROOT}/.git/)
find_package(Git REQUIRED)
execute_process(
COMMAND ${GIT_EXECUTABLE} log -1 --format=%H
WORKING_DIRECTORY ${PROJ_ROOT}
OUTPUT_VARIABLE GIT_SHA1
RESULT_VARIABLE GIT_RESULT
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
if(NOT ${GIT_RESULT})
add_definitions(-DPADDLE_VERSION=\"${GIT_SHA1}\")
else()
message(WARNING "Cannot add paddle version from git tag")
endif()
endif()
endif()
if(NOT WITH_GPU) if(NOT WITH_GPU)
......
RELEASE.md 0 → 100644
浏览文件 @ 021b3a44
# Release v0.9.0
## New Features:
* New Layers
* bilinear interpolation layer.
* spatial pyramid-pool layer.
* de-convolution layer.
* maxout layer.
* Support rectangle padding, stride, window and input for Pooling Operation.
* Add —job=time in trainer, which can be used to print time info without compiler option -WITH_TIMER=ON.
* Expose cost_weight/nce_layer in `trainer_config_helpers`
* Add FAQ, concepts, h-rnn docs.
* Add Bidi-LSTM and DB-LSTM to quick start demo @alvations
* Add usage track scripts.
## Improvements
* Add Travis-CI for Mac OS X. Enable swig unittest in Travis-CI. Skip Travis-CI when only docs are changed.
* Add code coverage tools.
* Refine convolution layer to speedup and reduce GPU memory.
* Speed up PyDataProvider2
* Add ubuntu deb package build scripts.
* Make Paddle use git-flow branching model.
* PServer support no parameter blocks.
## Bug Fixes
* add zlib link to py_paddle
* add input sparse data check for sparse layer at runtime
* Bug fix for sparse matrix multiplication
* Fix floating-point overflow problem of tanh
* Fix some nvcc compile options
* Fix a bug in yield dictionary in DataProvider
* Fix SRL hang when exit.
# Release v0.8.0beta.1
New features:
* Mac OSX is supported by source code. #138
* Both GPU and CPU versions of PaddlePaddle are supported.
* Support CUDA 8.0
* Enhance `PyDataProvider2`
* Add dictionary yield format. `PyDataProvider2` can yield a dictionary with key is data_layer's name, value is features.
* Add `min_pool_size` to control memory pool in provider.
* Add `deb` install package & docker image for no_avx machines.
* Especially for cloud computing and virtual machines
* Automatically disable `avx` instructions in cmake when machine's CPU don't support `avx` instructions.
* Add Parallel NN api in trainer_config_helpers.
* Add `travis ci` for Github
Bug fixes:
* Several bugs in trainer_config_helpers. Also complete the unittest for trainer_config_helpers
* Check if PaddlePaddle is installed when unittest.
* Fix bugs in GTX series GPU
* Fix bug in MultinomialSampler
Also more documentation was written since last release.
# Release v0.8.0beta.0
PaddlePaddle v0.8.0beta.0 release. The install package is not stable yet and it's a pre-release version.
benchmark/.gitignore 0 → 100644
浏览文件 @ 021b3a44
paddle/image/logs
paddle/image/*.pyc
paddle/image/train.list
paddle/rnn/logs
paddle/rnn/*.pyc
paddle/rnn/imdb.pkl
caffe/image/logs
tensorflow/image/logs
tensorflow/rnn/logs
benchmark/README.md 0 → 100644
浏览文件 @ 021b3a44
# Benchmark
Machine:
- CPU: 12-core Intel(R) Xeon(R) CPU E5-2620 v2 @2.10GHz
- GPU: Tesla K40m
- cuDNN: v5.1
- system: Docker 1.12.1, all platforms are tested in docker environment.
Platforms:
- PaddlePaddle: paddledev/paddle:gpu-devel-v0.9.0a0
- Tensorflow: gcr.io/tensorflow/tensorflow:0.11.0rc0-gpu
- Caffe: kaixhin/cuda-caffe
Several convolutional neural networks and recurrent neural networks are used to test.
## Image
### Benchmark Model
AlexNet, GoogleNet and a small network used in Caffe.
- [AlexNet](https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet): but the group size is one.
- [GoogleNet](https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet): but remove loss1 and loss2 when testing benchmark.
- [SmallNet](https://github.com/BVLC/caffe/blob/master/examples/cifar10/cifar10\_quick\_train\_test.prototxt)
### Single-GPU
- AlexNet: input - 3 * 227 * 227, Time: ms/batch
| BatchSize | 64 | 128 | 256 | 512 |
|--------------|-----| -----| ------| -----|
| PaddlePaddle | 195 | 334 | 602 | 1629 |
| TensorFlow | 223 | 364 | 645 | 1235 |
| Caffe | 324 | 627 | 1232 | 2513 |
**Notation**
All platforms use cuDNN-v5.1. We see that caffe is slower in this experiment, because its workspace limit size of cuDNN-conv interface is 8 * 1024 * 1024, which is smaller in PaddlePaddle and TensorFlow. Note that Caffe will be faster if increasing the workspace limit size.
- GoogletNet: input - 3 * 224 * 224, Time: ms/batch
| BatchSize | 64 | 128 | 256 |
|--------------|-------| -------| --------|
| PaddlePaddle | 613 | 1149 | 2348 |
| TensorFlow | 644 | 1176 | 2219 |
| Caffe | 694 | 1364 | out of memory |
- SmallNet: input - 3 * 32 * 32, Time ms/batch
| BatchSize | 64 | 128 | 256 | 512 |
|--------------|--------| -------- | --------|---------|
| PaddlePaddle | 10.463 | 18.184 | 33.113 | 63.039 |
| TensorFlow | 9 | 15 | 28 | 59 |
| Caffe | 9.373 | 16.6606 | 31.4797 | 59.719 |
**Notation**
All the single-GPU experiments in caffe use `caffe time` to calculate elapsed time, which does not include parameter updating time. However, both PaddlePaddle and TensorFlow experiments contain the parameter updating time. As compared with the total time, this part is relatively little on single machine, we can ignore it.
In Tensorflow, they implement algorithm searching method instead of using the algorithm searching interface in cuDNN.
### Multi-GPU: 4 GPUs
- AlexNet, ms / batch
| total-BatchSize | 128 * 4 | 256 * 4 |
|------------------|----------| -----------|
| PaddlePaddle | 347 | 622 |
| TensorFlow | 377 | 675 |
| Caffe | 1229 | 2435 |
For example, if `total-BatchSize = 128 * 4`, the speedup ratio is calculated by
```
time_at_1gpu_batch_128 * 4 / time_at_4gpu_total_batch_512
= (334 * 4)/347
= 3.85
```
<img src="figs/alexnet-4gpu.png" width="420">
- GoogleNet, ms / batch
| total-BatchSize | 128 * 4 | 256 * 4 |
|-------------------|--------------| ----------- |
| PaddlePaddle | 1178 | 2367 |
| TensorFlow | 1210 | 2292 |
| Caffe | 2007 | out of memory |
<img src="figs/googlenet-4gpu.png" width="420">
## RNN
We use lstm network for text classfication to test benchmark.
### Dataset
- [IMDB](http://www.iro.umontreal.ca/~lisa/deep/data/imdb.pkl)
- Sequence length is 100. In fact, PaddlePaddle supports training with variable-length sequence, but TensorFlow needs to pad. Thus, we also pad sequence length to 100 in PaddlePaddle in order to compare.
- Dictionary size=30000
- Peephole connection is used in `lstmemory` by default in PaddlePaddle. It is also configured in TensorFlow.
### Single-GPU
#### LSTM in Text Classification
Testing `2 lstm layer + fc` network with different hidden size and batch size.
- Batch size = 64, ms / batch
| hidden_size | 256 | 512 | 1280 |
|--------------|-------| -------| --------|
| PaddlePaddle | 83 | 184 | 641 |
| TensorFlow | 175 | 280 | 818 |
- Batch size = 128, ms / batch
| hidden_size | 256 | 512 | 1280 |
|--------------|------- | -------| --------|
| PaddlePaddle | 110 | 261 | 1007 |
| TensorFlow | 181 | 361 | 1237 |
- Batch size = 256, ms / batch
| hidden_size | 256 | 512 | 1280 |
|--------------|-------| -------| --------|
| PaddlePaddle | 170 | 414 | 1655 |
| TensorFlow | 238 | 536 | 1905 |
<img src="figs/rnn_lstm_cls.png" width="600">
#### Seq2Seq
The benchmark of sequence-to-sequence network will be added later.
### Multi GPU: 4 GPUs
#### LSTM in Text Classification
- hidden_size = 256, ms / batch
| batch_size | 256 | 512 |
|--------------| -------| --------|
| PaddlePaddle | 90 | 118 |
| TensorFlow | 226 | 118 |
- hidden_size = 512, ms / batch
| batch_size | 256 | 512 |
|--------------| -------| --------|
| PaddlePaddle | 189 | 268 |
| TensorFlow | 297 | 383 |
<img src="figs/rnn_lstm_4gpus.png" width="420">
#### Seq2Seq
The benchmark of sequence-to-sequence network will be added later.
benchmark/caffe/image/alexnet.prototxt 0 → 100644
浏览文件 @ 021b3a44
name: "alexnet"
input: "data"
input_dim: 64
input_dim: 3
input_dim: 227
input_dim: 227
input: "label"
input_dim: 64
input_dim: 1
input_dim: 1
input_dim: 1
force_backward: true
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 96
kernel_size: 11
stride: 4
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "conv1"
top: "conv1"
}
layer {
name: "norm1"
type: "LRN"
bottom: "conv1"
top: "norm1"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "norm1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 256
pad: 2
kernel_size: 5
group: 1
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0.1
}
}
}
layer {
name: "relu2"
type: "ReLU"
bottom: "conv2"
top: "conv2"
}
layer {
name: "norm2"
type: "LRN"
bottom: "conv2"
top: "norm2"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "pool2"
type: "Pooling"
bottom: "norm2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "conv3"
type: "Convolution"
bottom: "pool2"
top: "conv3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "relu3"
type: "ReLU"
bottom: "conv3"
top: "conv3"
}
layer {
name: "conv4"
type: "Convolution"
bottom: "conv3"
top: "conv4"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
group: 1
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0.1
}
}
}
layer {
name: "relu4"
type: "ReLU"
bottom: "conv4"
top: "conv4"
}
layer {
name: "conv5"
type: "Convolution"
bottom: "conv4"
top: "conv5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 256
pad: 1
kernel_size: 3
group: 1
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0.1
}
}
}
layer {
name: "relu5"
type: "ReLU"
bottom: "conv5"
top: "conv5"
}
layer {
name: "pool5"
type: "Pooling"
bottom: "conv5"
top: "pool5"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "fc6"
type: "InnerProduct"
bottom: "pool5"
top: "fc6"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 4096
weight_filler {
type: "gaussian"
std: 0.005
}
bias_filler {
type: "constant"
value: 0.1
}
}
}
layer {
name: "relu6"
type: "ReLU"
bottom: "fc6"
top: "fc6"
}
layer {
name: "drop6"
type: "Dropout"
bottom: "fc6"
top: "fc6"
dropout_param {
dropout_ratio: 0.5
}
}
layer {
name: "fc7"
type: "InnerProduct"
bottom: "fc6"
top: "fc7"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 4096
weight_filler {
type: "gaussian"
std: 0.005
}
bias_filler {
type: "constant"
value: 0.1
}
}
}
layer {
name: "relu7"
type: "ReLU"
bottom: "fc7"
top: "fc7"
}
layer {
name: "drop7"
type: "Dropout"
bottom: "fc7"
top: "fc7"
dropout_param {
dropout_ratio: 0.5
}
}
layer {
name: "fc8"
type: "InnerProduct"
bottom: "fc7"
top: "fc8"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 1000
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "loss"
type: "SoftmaxWithLoss"
bottom: "fc8"
bottom: "label"
top: "loss"
}
benchmark/caffe/image/googlenet.prototxt 0 → 100644
浏览文件 @ 021b3a44
name: "googlenet"
input: "data"
input_dim: 128
input_dim: 3
input_dim: 224
input_dim: 224
input: "label"
input_dim: 128
input_dim: 1
input_dim: 1
input_dim: 1
layer {
name: "conv1/7x7_s2"
type: "Convolution"
bottom: "data"
top: "conv1/7x7_s2"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
pad: 3
kernel_size: 7
stride: 2
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv1/relu_7x7"
type: "ReLU"
bottom: "conv1/7x7_s2"
top: "conv1/7x7_s2"
}
layer {
name: "pool1/3x3_s2"
type: "Pooling"
bottom: "conv1/7x7_s2"
top: "pool1/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
#layer {
# name: "pool1/norm1"
# type: "LRN"
# bottom: "pool1/3x3_s2"
# top: "pool1/norm1"
# lrn_param {
# local_size: 5
# alpha: 0.0001
# beta: 0.75
# }
#}
layer {
name: "conv2/3x3_reduce"
type: "Convolution"
# bottom: "pool1/norm1"
bottom: "pool1/3x3_s2"
top: "conv2/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv2/relu_3x3_reduce"
type: "ReLU"
bottom: "conv2/3x3_reduce"
top: "conv2/3x3_reduce"
}
layer {
name: "conv2/3x3"
type: "Convolution"
bottom: "conv2/3x3_reduce"
top: "conv2/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 192
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv2/relu_3x3"
type: "ReLU"
bottom: "conv2/3x3"
top: "conv2/3x3"
}
#layer {
# name: "conv2/norm2"
# type: "LRN"
# bottom: "conv2/3x3"
# top: "conv2/norm2"
# lrn_param {
# local_size: 5
# alpha: 0.0001
# beta: 0.75
# }
#}
layer {
name: "pool2/3x3_s2"
type: "Pooling"
# bottom: "conv2/norm2"
bottom: "conv2/3x3"
top: "pool2/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "inception_3a/1x1"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_1x1"
type: "ReLU"
bottom: "inception_3a/1x1"
top: "inception_3a/1x1"
}
layer {
name: "inception_3a/3x3_reduce"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 96
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_3a/3x3_reduce"
top: "inception_3a/3x3_reduce"
}
layer {
name: "inception_3a/3x3"
type: "Convolution"
bottom: "inception_3a/3x3_reduce"
top: "inception_3a/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_3x3"
type: "ReLU"
bottom: "inception_3a/3x3"
top: "inception_3a/3x3"
}
layer {
name: "inception_3a/5x5_reduce"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 16
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_3a/5x5_reduce"
top: "inception_3a/5x5_reduce"
}
layer {
name: "inception_3a/5x5"
type: "Convolution"
bottom: "inception_3a/5x5_reduce"
top: "inception_3a/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 32
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_5x5"
type: "ReLU"
bottom: "inception_3a/5x5"
top: "inception_3a/5x5"
}
layer {
name: "inception_3a/pool"
type: "Pooling"
bottom: "pool2/3x3_s2"
top: "inception_3a/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_3a/pool_proj"
type: "Convolution"
bottom: "inception_3a/pool"
top: "inception_3a/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_pool_proj"
type: "ReLU"
bottom: "inception_3a/pool_proj"
top: "inception_3a/pool_proj"
}
layer {
name: "inception_3a/output"
type: "Concat"
bottom: "inception_3a/1x1"
bottom: "inception_3a/3x3"
bottom: "inception_3a/5x5"
bottom: "inception_3a/pool_proj"
top: "inception_3a/output"
}
layer {
name: "inception_3b/1x1"
type: "Convolution"
bottom: "inception_3a/output"
top: "inception_3b/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3b/relu_1x1"
type: "ReLU"
bottom: "inception_3b/1x1"
top: "inception_3b/1x1"
}
layer {
name: "inception_3b/3x3_reduce"
type: "Convolution"
bottom: "inception_3a/output"
top: "inception_3b/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3b/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_3b/3x3_reduce"
top: "inception_3b/3x3_reduce"
}
layer {
name: "inception_3b/3x3"
type: "Convolution"
bottom: "inception_3b/3x3_reduce"
top: "inception_3b/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 192
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3b/relu_3x3"
type: "ReLU"
bottom: "inception_3b/3x3"
top: "inception_3b/3x3"
}
layer {
name: "inception_3b/5x5_reduce"
type: "Convolution"
bottom: "inception_3a/output"
top: "inception_3b/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3b/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_3b/5x5_reduce"
top: "inception_3b/5x5_reduce"
}
layer {
name: "inception_3b/5x5"
type: "Convolution"
bottom: "inception_3b/5x5_reduce"
top: "inception_3b/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 96
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3b/relu_5x5"
type: "ReLU"
bottom: "inception_3b/5x5"
top: "inception_3b/5x5"
}
layer {
name: "inception_3b/pool"
type: "Pooling"
bottom: "inception_3a/output"
top: "inception_3b/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_3b/pool_proj"
type: "Convolution"
bottom: "inception_3b/pool"
top: "inception_3b/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3b/relu_pool_proj"
type: "ReLU"
bottom: "inception_3b/pool_proj"
top: "inception_3b/pool_proj"
}
layer {
name: "inception_3b/output"
type: "Concat"
bottom: "inception_3b/1x1"
bottom: "inception_3b/3x3"
bottom: "inception_3b/5x5"
bottom: "inception_3b/pool_proj"
top: "inception_3b/output"
}
layer {
name: "pool3/3x3_s2"
type: "Pooling"
bottom: "inception_3b/output"
top: "pool3/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "inception_4a/1x1"
type: "Convolution"
bottom: "pool3/3x3_s2"
top: "inception_4a/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 192
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4a/relu_1x1"
type: "ReLU"
bottom: "inception_4a/1x1"
top: "inception_4a/1x1"
}
layer {
name: "inception_4a/3x3_reduce"
type: "Convolution"
bottom: "pool3/3x3_s2"
top: "inception_4a/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 96
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4a/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_4a/3x3_reduce"
top: "inception_4a/3x3_reduce"
}
layer {
name: "inception_4a/3x3"
type: "Convolution"
bottom: "inception_4a/3x3_reduce"
top: "inception_4a/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 208
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4a/relu_3x3"
type: "ReLU"
bottom: "inception_4a/3x3"
top: "inception_4a/3x3"
}
layer {
name: "inception_4a/5x5_reduce"
type: "Convolution"
bottom: "pool3/3x3_s2"
top: "inception_4a/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 16
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4a/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_4a/5x5_reduce"
top: "inception_4a/5x5_reduce"
}
layer {
name: "inception_4a/5x5"
type: "Convolution"
bottom: "inception_4a/5x5_reduce"
top: "inception_4a/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 48
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4a/relu_5x5"
type: "ReLU"
bottom: "inception_4a/5x5"
top: "inception_4a/5x5"
}
layer {
name: "inception_4a/pool"
type: "Pooling"
bottom: "pool3/3x3_s2"
top: "inception_4a/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_4a/pool_proj"
type: "Convolution"
bottom: "inception_4a/pool"
top: "inception_4a/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4a/relu_pool_proj"
type: "ReLU"
bottom: "inception_4a/pool_proj"
top: "inception_4a/pool_proj"
}
layer {
name: "inception_4a/output"
type: "Concat"
bottom: "inception_4a/1x1"
bottom: "inception_4a/3x3"
bottom: "inception_4a/5x5"
bottom: "inception_4a/pool_proj"
top: "inception_4a/output"
}
#layer {
# name: "loss1/ave_pool"
# type: "Pooling"
# bottom: "inception_4a/output"
# top: "loss1/ave_pool"
# pooling_param {
# pool: AVE
# kernel_size: 5
# stride: 3
# }
#}
#layer {
# name: "loss1/conv"
# type: "Convolution"
# bottom: "loss1/ave_pool"
# top: "loss1/conv"
# param {
# lr_mult: 1
# decay_mult: 1
# }
# param {
# lr_mult: 2
# decay_mult: 0
# }
# convolution_param {
# num_output: 128
# kernel_size: 1
# weight_filler {
# type: "xavier"
# }
# bias_filler {
# type: "constant"
# value: 0.2
# }
# }
#}
#layer {
# name: "loss1/relu_conv"
# type: "ReLU"
# bottom: "loss1/conv"
# top: "loss1/conv"
#}
#layer {
# name: "loss1/fc"
# type: "InnerProduct"
# bottom: "loss1/conv"
# top: "loss1/fc"
# param {
# lr_mult: 1
# decay_mult: 1
# }
# param {
# lr_mult: 2
# decay_mult: 0
# }
# inner_product_param {
# num_output: 1024
# weight_filler {
# type: "xavier"
# }
# bias_filler {
# type: "constant"
# value: 0.2
# }
# }
#}
#layer {
# name: "loss1/relu_fc"
# type: "ReLU"
# bottom: "loss1/fc"
# top: "loss1/fc"
#}
#layer {
# name: "loss1/drop_fc"
# type: "Dropout"
# bottom: "loss1/fc"
# top: "loss1/fc"
# dropout_param {
# dropout_ratio: 0.7
# }
#}
#layer {
# name: "loss1/classifier"
# type: "InnerProduct"
# bottom: "loss1/fc"
# top: "loss1/classifier"
# param {
# lr_mult: 1
# decay_mult: 1
# }
# param {
# lr_mult: 2
# decay_mult: 0
# }
# inner_product_param {
# num_output: 1000
# weight_filler {
# type: "xavier"
# }
# bias_filler {
# type: "constant"
# value: 0
# }
# }
#}
#layer {
# name: "loss1/loss"
# type: "SoftmaxWithLoss"
# bottom: "loss1/classifier"
# bottom: "label"
# top: "loss1/loss1"
# loss_weight: 0.3
#}
layer {
name: "inception_4b/1x1"
type: "Convolution"
bottom: "inception_4a/output"
top: "inception_4b/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 160
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4b/relu_1x1"
type: "ReLU"
bottom: "inception_4b/1x1"
top: "inception_4b/1x1"
}
layer {
name: "inception_4b/3x3_reduce"
type: "Convolution"
bottom: "inception_4a/output"
top: "inception_4b/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 112
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4b/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_4b/3x3_reduce"
top: "inception_4b/3x3_reduce"
}
layer {
name: "inception_4b/3x3"
type: "Convolution"
bottom: "inception_4b/3x3_reduce"
top: "inception_4b/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 224
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4b/relu_3x3"
type: "ReLU"
bottom: "inception_4b/3x3"
top: "inception_4b/3x3"
}
layer {
name: "inception_4b/5x5_reduce"
type: "Convolution"
bottom: "inception_4a/output"
top: "inception_4b/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 24
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4b/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_4b/5x5_reduce"
top: "inception_4b/5x5_reduce"
}
layer {
name: "inception_4b/5x5"
type: "Convolution"
bottom: "inception_4b/5x5_reduce"
top: "inception_4b/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4b/relu_5x5"
type: "ReLU"
bottom: "inception_4b/5x5"
top: "inception_4b/5x5"
}
layer {
name: "inception_4b/pool"
type: "Pooling"
bottom: "inception_4a/output"
top: "inception_4b/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_4b/pool_proj"
type: "Convolution"
bottom: "inception_4b/pool"
top: "inception_4b/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4b/relu_pool_proj"
type: "ReLU"
bottom: "inception_4b/pool_proj"
top: "inception_4b/pool_proj"
}
layer {
name: "inception_4b/output"
type: "Concat"
bottom: "inception_4b/1x1"
bottom: "inception_4b/3x3"
bottom: "inception_4b/5x5"
bottom: "inception_4b/pool_proj"
top: "inception_4b/output"
}
layer {
name: "inception_4c/1x1"
type: "Convolution"
bottom: "inception_4b/output"
top: "inception_4c/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4c/relu_1x1"
type: "ReLU"
bottom: "inception_4c/1x1"
top: "inception_4c/1x1"
}
layer {
name: "inception_4c/3x3_reduce"
type: "Convolution"
bottom: "inception_4b/output"
top: "inception_4c/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4c/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_4c/3x3_reduce"
top: "inception_4c/3x3_reduce"
}
layer {
name: "inception_4c/3x3"
type: "Convolution"
bottom: "inception_4c/3x3_reduce"
top: "inception_4c/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 256
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4c/relu_3x3"
type: "ReLU"
bottom: "inception_4c/3x3"
top: "inception_4c/3x3"
}
layer {
name: "inception_4c/5x5_reduce"
type: "Convolution"
bottom: "inception_4b/output"
top: "inception_4c/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 24
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4c/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_4c/5x5_reduce"
top: "inception_4c/5x5_reduce"
}
layer {
name: "inception_4c/5x5"
type: "Convolution"
bottom: "inception_4c/5x5_reduce"
top: "inception_4c/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4c/relu_5x5"
type: "ReLU"
bottom: "inception_4c/5x5"
top: "inception_4c/5x5"
}
layer {
name: "inception_4c/pool"
type: "Pooling"
bottom: "inception_4b/output"
top: "inception_4c/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_4c/pool_proj"
type: "Convolution"
bottom: "inception_4c/pool"
top: "inception_4c/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4c/relu_pool_proj"
type: "ReLU"
bottom: "inception_4c/pool_proj"
top: "inception_4c/pool_proj"
}
layer {
name: "inception_4c/output"
type: "Concat"
bottom: "inception_4c/1x1"
bottom: "inception_4c/3x3"
bottom: "inception_4c/5x5"
bottom: "inception_4c/pool_proj"
top: "inception_4c/output"
}
layer {
name: "inception_4d/1x1"
type: "Convolution"
bottom: "inception_4c/output"
top: "inception_4d/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 112
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4d/relu_1x1"
type: "ReLU"
bottom: "inception_4d/1x1"
top: "inception_4d/1x1"
}
layer {
name: "inception_4d/3x3_reduce"
type: "Convolution"
bottom: "inception_4c/output"
top: "inception_4d/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 144
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4d/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_4d/3x3_reduce"
top: "inception_4d/3x3_reduce"
}
layer {
name: "inception_4d/3x3"
type: "Convolution"
bottom: "inception_4d/3x3_reduce"
top: "inception_4d/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 288
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4d/relu_3x3"
type: "ReLU"
bottom: "inception_4d/3x3"
top: "inception_4d/3x3"
}
layer {
name: "inception_4d/5x5_reduce"
type: "Convolution"
bottom: "inception_4c/output"
top: "inception_4d/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4d/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_4d/5x5_reduce"
top: "inception_4d/5x5_reduce"
}
layer {
name: "inception_4d/5x5"
type: "Convolution"
bottom: "inception_4d/5x5_reduce"
top: "inception_4d/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4d/relu_5x5"
type: "ReLU"
bottom: "inception_4d/5x5"
top: "inception_4d/5x5"
}
layer {
name: "inception_4d/pool"
type: "Pooling"
bottom: "inception_4c/output"
top: "inception_4d/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_4d/pool_proj"
type: "Convolution"
bottom: "inception_4d/pool"
top: "inception_4d/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 64
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4d/relu_pool_proj"
type: "ReLU"
bottom: "inception_4d/pool_proj"
top: "inception_4d/pool_proj"
}
layer {
name: "inception_4d/output"
type: "Concat"
bottom: "inception_4d/1x1"
bottom: "inception_4d/3x3"
bottom: "inception_4d/5x5"
bottom: "inception_4d/pool_proj"
top: "inception_4d/output"
}
#layer {
# name: "loss2/ave_pool"
# type: "Pooling"
# bottom: "inception_4d/output"
# top: "loss2/ave_pool"
# pooling_param {
# pool: AVE
# kernel_size: 5
# stride: 3
# }
#}
#layer {
# name: "loss2/conv"
# type: "Convolution"
# bottom: "loss2/ave_pool"
# top: "loss2/conv"
# param {
# lr_mult: 1
# decay_mult: 1
# }
# param {
# lr_mult: 2
# decay_mult: 0
# }
# convolution_param {
# num_output: 128
# kernel_size: 1
# weight_filler {
# type: "xavier"
# }
# bias_filler {
# type: "constant"
# value: 0.2
# }
# }
#}
#layer {
# name: "loss2/relu_conv"
# type: "ReLU"
# bottom: "loss2/conv"
# top: "loss2/conv"
#}
#layer {
# name: "loss2/fc"
# type: "InnerProduct"
# bottom: "loss2/conv"
# top: "loss2/fc"
# param {
# lr_mult: 1
# decay_mult: 1
# }
# param {
# lr_mult: 2
# decay_mult: 0
# }
# inner_product_param {
# num_output: 1024
# weight_filler {
# type: "xavier"
# }
# bias_filler {
# type: "constant"
# value: 0.2
# }
# }
#}
#layer {
# name: "loss2/relu_fc"
# type: "ReLU"
# bottom: "loss2/fc"
# top: "loss2/fc"
#}
#layer {
# name: "loss2/drop_fc"
# type: "Dropout"
# bottom: "loss2/fc"
# top: "loss2/fc"
# dropout_param {
# dropout_ratio: 0.7
# }
#}
#layer {
# name: "loss2/classifier"
# type: "InnerProduct"
# bottom: "loss2/fc"
# top: "loss2/classifier"
# param {
# lr_mult: 1
# decay_mult: 1
# }
# param {
# lr_mult: 2
# decay_mult: 0
# }
# inner_product_param {
# num_output: 1000
# weight_filler {
# type: "xavier"
# }
# bias_filler {
# type: "constant"
# value: 0
# }
# }
#}
#layer {
# name: "loss2/loss"
# type: "SoftmaxWithLoss"
# bottom: "loss2/classifier"
# bottom: "label"
# top: "loss2/loss1"
# loss_weight: 0.3
#}
layer {
name: "inception_4e/1x1"
type: "Convolution"
bottom: "inception_4d/output"
top: "inception_4e/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 256
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4e/relu_1x1"
type: "ReLU"
bottom: "inception_4e/1x1"
top: "inception_4e/1x1"
}
layer {
name: "inception_4e/3x3_reduce"
type: "Convolution"
bottom: "inception_4d/output"
top: "inception_4e/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 160
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4e/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_4e/3x3_reduce"
top: "inception_4e/3x3_reduce"
}
layer {
name: "inception_4e/3x3"
type: "Convolution"
bottom: "inception_4e/3x3_reduce"
top: "inception_4e/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 320
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4e/relu_3x3"
type: "ReLU"
bottom: "inception_4e/3x3"
top: "inception_4e/3x3"
}
layer {
name: "inception_4e/5x5_reduce"
type: "Convolution"
bottom: "inception_4d/output"
top: "inception_4e/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4e/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_4e/5x5_reduce"
top: "inception_4e/5x5_reduce"
}
layer {
name: "inception_4e/5x5"
type: "Convolution"
bottom: "inception_4e/5x5_reduce"
top: "inception_4e/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4e/relu_5x5"
type: "ReLU"
bottom: "inception_4e/5x5"
top: "inception_4e/5x5"
}
layer {
name: "inception_4e/pool"
type: "Pooling"
bottom: "inception_4d/output"
top: "inception_4e/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_4e/pool_proj"
type: "Convolution"
bottom: "inception_4e/pool"
top: "inception_4e/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_4e/relu_pool_proj"
type: "ReLU"
bottom: "inception_4e/pool_proj"
top: "inception_4e/pool_proj"
}
layer {
name: "inception_4e/output"
type: "Concat"
bottom: "inception_4e/1x1"
bottom: "inception_4e/3x3"
bottom: "inception_4e/5x5"
bottom: "inception_4e/pool_proj"
top: "inception_4e/output"
}
layer {
name: "pool4/3x3_s2"
type: "Pooling"
bottom: "inception_4e/output"
top: "pool4/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "inception_5a/1x1"
type: "Convolution"
bottom: "pool4/3x3_s2"
top: "inception_5a/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 256
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5a/relu_1x1"
type: "ReLU"
bottom: "inception_5a/1x1"
top: "inception_5a/1x1"
}
layer {
name: "inception_5a/3x3_reduce"
type: "Convolution"
bottom: "pool4/3x3_s2"
top: "inception_5a/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 160
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5a/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_5a/3x3_reduce"
top: "inception_5a/3x3_reduce"
}
layer {
name: "inception_5a/3x3"
type: "Convolution"
bottom: "inception_5a/3x3_reduce"
top: "inception_5a/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 320
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5a/relu_3x3"
type: "ReLU"
bottom: "inception_5a/3x3"
top: "inception_5a/3x3"
}
layer {
name: "inception_5a/5x5_reduce"
type: "Convolution"
bottom: "pool4/3x3_s2"
top: "inception_5a/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 32
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5a/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_5a/5x5_reduce"
top: "inception_5a/5x5_reduce"
}
layer {
name: "inception_5a/5x5"
type: "Convolution"
bottom: "inception_5a/5x5_reduce"
top: "inception_5a/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5a/relu_5x5"
type: "ReLU"
bottom: "inception_5a/5x5"
top: "inception_5a/5x5"
}
layer {
name: "inception_5a/pool"
type: "Pooling"
bottom: "pool4/3x3_s2"
top: "inception_5a/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_5a/pool_proj"
type: "Convolution"
bottom: "inception_5a/pool"
top: "inception_5a/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5a/relu_pool_proj"
type: "ReLU"
bottom: "inception_5a/pool_proj"
top: "inception_5a/pool_proj"
}
layer {
name: "inception_5a/output"
type: "Concat"
bottom: "inception_5a/1x1"
bottom: "inception_5a/3x3"
bottom: "inception_5a/5x5"
bottom: "inception_5a/pool_proj"
top: "inception_5a/output"
}
layer {
name: "inception_5b/1x1"
type: "Convolution"
bottom: "inception_5a/output"
top: "inception_5b/1x1"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 384
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5b/relu_1x1"
type: "ReLU"
bottom: "inception_5b/1x1"
top: "inception_5b/1x1"
}
layer {
name: "inception_5b/3x3_reduce"
type: "Convolution"
bottom: "inception_5a/output"
top: "inception_5b/3x3_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 192
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5b/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_5b/3x3_reduce"
top: "inception_5b/3x3_reduce"
}
layer {
name: "inception_5b/3x3"
type: "Convolution"
bottom: "inception_5b/3x3_reduce"
top: "inception_5b/3x3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 384
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5b/relu_3x3"
type: "ReLU"
bottom: "inception_5b/3x3"
top: "inception_5b/3x3"
}
layer {
name: "inception_5b/5x5_reduce"
type: "Convolution"
bottom: "inception_5a/output"
top: "inception_5b/5x5_reduce"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 48
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5b/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_5b/5x5_reduce"
top: "inception_5b/5x5_reduce"
}
layer {
name: "inception_5b/5x5"
type: "Convolution"
bottom: "inception_5b/5x5_reduce"
top: "inception_5b/5x5"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5b/relu_5x5"
type: "ReLU"
bottom: "inception_5b/5x5"
top: "inception_5b/5x5"
}
layer {
name: "inception_5b/pool"
type: "Pooling"
bottom: "inception_5a/output"
top: "inception_5b/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_5b/pool_proj"
type: "Convolution"
bottom: "inception_5b/pool"
top: "inception_5b/pool_proj"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 128
kernel_size: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_5b/relu_pool_proj"
type: "ReLU"
bottom: "inception_5b/pool_proj"
top: "inception_5b/pool_proj"
}
layer {
name: "inception_5b/output"
type: "Concat"
bottom: "inception_5b/1x1"
bottom: "inception_5b/3x3"
bottom: "inception_5b/5x5"
bottom: "inception_5b/pool_proj"
top: "inception_5b/output"
}
layer {
name: "pool5/7x7_s1"
type: "Pooling"
bottom: "inception_5b/output"
top: "pool5/7x7_s1"
pooling_param {
pool: AVE
kernel_size: 7
stride: 1
}
}
layer {
name: "pool5/drop_7x7_s1"
type: "Dropout"
bottom: "pool5/7x7_s1"
top: "pool5/7x7_s1"
dropout_param {
dropout_ratio: 0.4
}
}
layer {
name: "loss3/classifier"
type: "InnerProduct"
bottom: "pool5/7x7_s1"
top: "loss3/classifier"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 1000
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "loss3/loss3"
type: "SoftmaxWithLoss"
bottom: "loss3/classifier"
bottom: "label"
top: "loss3/loss3"
loss_weight: 1
}
benchmark/caffe/image/run.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function test() {
cfg=$1
batch=$2
prefix=$3
sed -i "/input: \"data\"/{n;s/^input_dim.*/input_dim: $batch/g}" $cfg
sed -i "/input: \"label\"/{n;s/^input_dim.*/input_dim: $batch/g}" $cfg
caffe time --model=$cfg --iterations=50 --gpu 0 > logs/$prefix-1gpu-batch${batch}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
# alexnet
test alexnet.prototxt 64 alexnet
test alexnet.prototxt 128 alexnet
test alexnet.prototxt 256 alexnet
test alexnet.prototxt 512 alexnet
# googlenet
test googlenet.prototxt 64 googlenet
test googlenet.prototxt 128 googlenet
# small net
test smallnet_mnist_cifar.prototxt 64 smallnet
test smallnet_mnist_cifar.prototxt 128 smallnet
test smallnet_mnist_cifar.prototxt 256 smallnet
test smallnet_mnist_cifar.prototxt 512 smallnet
benchmark/caffe/image/run_multi.sh 0 → 100755
浏览文件 @ 021b3a44
#!/bin/bash
set -e
function test() {
cfg=$1
batch=$2
prefix=$3
batch_per_gpu=`expr ${batch} / 4`
sed -i "/input: \"data\"/{n;s/^input_dim.*/input_dim: ${batch_per_gpu}/g}" $cfg
sed -i "/input: \"label\"/{n;s/^input_dim.*/input_dim: ${batch_per_gpu}/g}" $cfg
sed -i "1c\net : \"${cfg}\"" solver.prototxt
caffe train --solver=solver.prototxt -gpu 0,1,2,3 > logs/${prefix}-4gpu-batch${batch}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
# alexnet
test alexnet.prototxt 512 alexnet
test alexnet.prototxt 1024 alexnet
# googlnet
test googlenet.prototxt 512 googlenet
benchmark/caffe/image/smallnet_mnist_cifar.prototxt 0 → 100644
浏览文件 @ 021b3a44
name: "mnist/cifar"
input: "data"
input_dim: 128
input_dim: 3
input_dim: 32
input_dim: 32
input: "label"
input_dim: 128
input_dim: 1
input_dim: 1
input_dim: 1
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 32
pad: 2
kernel_size: 5
stride: 1
weight_filler {
type: "gaussian"
std: 0.0001
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "pool1"
top: "pool1"
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 32
pad: 2
kernel_size: 5
stride: 1
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu2"
type: "ReLU"
bottom: "conv2"
top: "conv2"
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: AVE
kernel_size: 3
stride: 2
}
}
layer {
name: "conv3"
type: "Convolution"
bottom: "pool2"
top: "conv3"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 64
pad: 2
kernel_size: 5
stride: 1
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu3"
type: "ReLU"
bottom: "conv3"
top: "conv3"
}
layer {
name: "pool3"
type: "Pooling"
bottom: "conv3"
top: "pool3"
pooling_param {
pool: AVE
kernel_size: 3
stride: 2
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool3"
top: "ip1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 64
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 10
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "accuracy"
type: "Accuracy"
bottom: "ip2"
bottom: "label"
top: "accuracy"
include {
phase: TEST
}
}
layer {
name: "loss"
type: "SoftmaxWithLoss"
bottom: "ip2"
bottom: "label"
top: "loss"
}
benchmark/caffe/image/solver.prototxt 0 → 100644
浏览文件 @ 021b3a44
net: "alexnet.prototxt"
base_lr: 0.01
lr_policy: "fixed"
display: 20
max_iter: 200
momentum: 0.9
weight_decay: 0.0005
snapshot: 10000
snapshot_prefix: "models/caffe_alexnet_train"
solver_mode: GPU
benchmark/paddle/image/alexnet.py 0 → 100644
浏览文件 @ 021b3a44
#!/usr/bin/env python
from paddle.trainer_config_helpers import *
height = 227
width = 227
num_class = 1000
batch_size = get_config_arg('batch_size', int, 128)
args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
define_py_data_sources2(
"train.list", None, module="provider", obj="process", args=args)
settings(
batch_size=batch_size,
learning_rate=0.01 / batch_size,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * batch_size))
# conv1
net = data_layer('data', size=height * width * 3)
net = img_conv_layer(
input=net,
filter_size=11,
num_channels=3,
num_filters=96,
stride=4,
padding=1)
net = img_cmrnorm_layer(input=net, size=5, scale=0.0001, power=0.75)
net = img_pool_layer(input=net, pool_size=3, stride=2)
# conv2
net = img_conv_layer(
input=net, filter_size=5, num_filters=256, stride=1, padding=2, groups=1)
net = img_cmrnorm_layer(input=net, size=5, scale=0.0001, power=0.75)
net = img_pool_layer(input=net, pool_size=3, stride=2)
# conv3
net = img_conv_layer(
input=net, filter_size=3, num_filters=384, stride=1, padding=1)
# conv4
net = img_conv_layer(
input=net, filter_size=3, num_filters=384, stride=1, padding=1, groups=1)
# conv5
net = img_conv_layer(
input=net, filter_size=3, num_filters=256, stride=1, padding=1, groups=1)
net = img_pool_layer(input=net, pool_size=3, stride=2)
net = fc_layer(
input=net,
size=4096,
act=ReluActivation(),
layer_attr=ExtraAttr(drop_rate=0.5))
net = fc_layer(
input=net,
size=4096,
act=ReluActivation(),
layer_attr=ExtraAttr(drop_rate=0.5))
net = fc_layer(input=net, size=1000, act=SoftmaxActivation())
lab = data_layer('label', num_class)
loss = cross_entropy(input=net, label=lab)
outputs(loss)
benchmark/paddle/image/googlenet.py 0 → 100644
浏览文件 @ 021b3a44
#!/usr/bin/env python
from paddle.trainer_config_helpers import *
height = 224
width = 224
num_class = 1000
batch_size = get_config_arg('batch_size', int, 128)
args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
define_py_data_sources2(
"train.list", None, module="provider", obj="process", args=args)
settings(
batch_size=batch_size,
learning_rate=0.01 / batch_size,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * batch_size))
def inception2(name, input, channels, \
filter1,
filter3R, filter3,
filter5R, filter5,
proj):
conv1 = name + '_1'
conv3r = name + '_3r'
conv3 = name + '_3'
conv5r = name + '_5r'
conv5 = name + '_5'
maxpool = name + '_max'
convproj = name + '_proj'
cov1 = img_conv_layer(
name=conv1,
input=input,
filter_size=1,
num_channels=channels,
num_filters=filter1,
stride=1,
padding=0)
cov3r = img_conv_layer(
name=conv3r,
input=input,
filter_size=1,
num_channels=channels,
num_filters=filter3R,
stride=1,
padding=0)
cov3 = img_conv_layer(
name=conv3,
input=cov3r,
filter_size=3,
num_filters=filter3,
stride=1,
padding=1)
cov5r = img_conv_layer(
name=conv5r,
input=input,
filter_size=1,
num_channels=channels,
num_filters=filter5R,
stride=1,
padding=0)
cov5 = img_conv_layer(
name=conv5,
input=cov5r,
filter_size=5,
num_filters=filter5,
stride=1,
padding=2)
pool1 = img_pool_layer(
name=maxpool,
input=input,
pool_size=3,
num_channels=channels,
stride=1,
padding=1)
covprj = img_conv_layer(
name=convproj,
input=pool1,
filter_size=1,
num_filters=proj,
stride=1,
padding=0)
cat = concat_layer(name=name, input=[cov1, cov3, cov5, covprj])
return cat
def inception(name, input, channels, \
filter1,
filter3R, filter3,
filter5R, filter5,
proj):
cov1 = conv_projection(
input=input,
filter_size=1,
num_channels=channels,
num_filters=filter1,
stride=1,
padding=0)
cov3r = img_conv_layer(
name=name + '_3r',
input=input,
filter_size=1,
num_channels=channels,
num_filters=filter3R,
stride=1,
padding=0)
cov3 = conv_projection(
input=cov3r, filter_size=3, num_filters=filter3, stride=1, padding=1)
cov5r = img_conv_layer(
name=name + '_5r',
input=input,
filter_size=1,
num_channels=channels,
num_filters=filter5R,
stride=1,
padding=0)
cov5 = conv_projection(
input=cov5r, filter_size=5, num_filters=filter5, stride=1, padding=2)
pool1 = img_pool_layer(
name=name + '_max',
input=input,
pool_size=3,
num_channels=channels,
stride=1,
padding=1)
covprj = conv_projection(
input=pool1, filter_size=1, num_filters=proj, stride=1, padding=0)
cat = concat_layer(
name=name,
input=[cov1, cov3, cov5, covprj],
bias_attr=True,
act=ReluActivation())
return cat
lab = data_layer(name="label", size=1000)
data = data_layer(name="input", size=3 * height * width)
# stage 1
conv1 = img_conv_layer(
name="conv1",
input=data,
filter_size=7,
num_channels=3,
num_filters=64,
stride=2,
padding=3)
pool1 = img_pool_layer(
name="pool1", input=conv1, pool_size=3, num_channels=64, stride=2)
# stage 2
conv2_1 = img_conv_layer(
name="conv2_1",
input=pool1,
filter_size=1,
num_filters=64,
stride=1,
padding=0)
conv2_2 = img_conv_layer(
name="conv2_2",
input=conv2_1,
filter_size=3,
num_filters=192,
stride=1,
padding=1)
pool2 = img_pool_layer(
name="pool2", input=conv2_2, pool_size=3, num_channels=192, stride=2)
# stage 3
ince3a = inception("ince3a", pool2, 192, 64, 96, 128, 16, 32, 32)
ince3b = inception("ince3b", ince3a, 256, 128, 128, 192, 32, 96, 64)
pool3 = img_pool_layer(
name="pool3", input=ince3b, num_channels=480, pool_size=3, stride=2)
# stage 4
ince4a = inception("ince4a", pool3, 480, 192, 96, 208, 16, 48, 64)
ince4b = inception("ince4b", ince4a, 512, 160, 112, 224, 24, 64, 64)
ince4c = inception("ince4c", ince4b, 512, 128, 128, 256, 24, 64, 64)
ince4d = inception("ince4d", ince4c, 512, 112, 144, 288, 32, 64, 64)
ince4e = inception("ince4e", ince4d, 528, 256, 160, 320, 32, 128, 128)
pool4 = img_pool_layer(
name="pool4", input=ince4e, num_channels=832, pool_size=3, stride=2)
# stage 5
ince5a = inception("ince5a", pool4, 832, 256, 160, 320, 32, 128, 128)
ince5b = inception("ince5b", ince5a, 832, 384, 192, 384, 48, 128, 128)
pool5 = img_pool_layer(
name="pool5",
input=ince5b,
num_channels=1024,
pool_size=7,
stride=7,
pool_type=AvgPooling())
# We remove loss1 and loss2 for all system when testing benchmark
# output 1
# pool_o1 = img_pool_layer(name="pool_o1", input=ince4a, num_channels=512, pool_size=5, stride=3, pool_type=AvgPooling())
# conv_o1 = img_conv_layer(name="conv_o1", input=pool_o1, filter_size=1, num_filters=128, stride=1, padding=0)
# fc_o1 = fc_layer(name="fc_o1", input=conv_o1, size=1024, layer_attr=ExtraAttr(drop_rate=0.7), act=ReluActivation())
# out1 = fc_layer(name="output1", input=fc_o1, size=1000, act=SoftmaxActivation())
# loss1 = cross_entropy(name='loss1', input=out1, label=lab, coeff=0.3)
# output 2
#pool_o2 = img_pool_layer(name="pool_o2", input=ince4d, num_channels=528, pool_size=5, stride=3, pool_type=AvgPooling())
#conv_o2 = img_conv_layer(name="conv_o2", input=pool_o2, filter_size=1, num_filters=128, stride=1, padding=0)
#fc_o2 = fc_layer(name="fc_o2", input=conv_o2, size=1024, layer_attr=ExtraAttr(drop_rate=0.7), act=ReluActivation())
#out2 = fc_layer(name="output2", input=fc_o2, size=1000, act=SoftmaxActivation())
#loss2 = cross_entropy(name='loss2', input=out2, label=lab, coeff=0.3)
# output 3
dropout = dropout_layer(name="dropout", input=pool5, dropout_rate=0.4)
out3 = fc_layer(
name="output3", input=dropout, size=1000, act=SoftmaxActivation())
loss3 = cross_entropy(name='loss3', input=out3, label=lab)
outputs(loss3)
benchmark/paddle/image/provider.py 0 → 100644
浏览文件 @ 021b3a44
import io, os
import random
import numpy as np
from paddle.trainer.PyDataProvider2 import *
def initHook(settings, height, width, color, num_class, **kwargs):
settings.height = height
settings.width = width
settings.color = color
settings.num_class = num_class
if settings.color:
settings.data_size = settings.height * settings.width * 3
else:
settings.data_size = settings.height * settings.width
settings.slots = [dense_vector(settings.data_size), integer_value(1)]
@provider(
init_hook=initHook, min_pool_size=-1, cache=CacheType.CACHE_PASS_IN_MEM)
def process(settings, file_list):
for i in xrange(1024):
img = np.random.rand(1, settings.data_size).reshape(-1, 1).flatten()
lab = random.randint(0, settings.num_class)
yield img.astype('float32'), int(lab)
benchmark/paddle/image/run.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function train() {
cfg=$1
thread=$2
bz=$3
args="batch_size=$3"
prefix=$4
paddle train --job=time \
--config=$cfg \
--use_gpu=True \
--trainer_count=$thread \
--log_period=10 \
--test_period=100 \
--config_args=$args \
> logs/$prefix-${thread}gpu-$bz.log 2>&1
}
if [ ! -d "train.list" ]; then
echo " " > train.list
fi
if [ ! -d "logs" ]; then
mkdir logs
fi
#========single-gpu=========#
# alexnet
train alexnet.py 1 64 alexnet
train alexnet.py 1 128 alexnet
train alexnet.py 1 256 alexnet
train alexnet.py 1 512 alexnet
# googlenet
train googlenet.py 1 64 googlenet
train googlenet.py 1 128 googlenet
train googlenet.py 1 256 googlenet
# smallnet
train smallnet_mnist_cifar.py 1 64 smallnet
train smallnet_mnist_cifar.py 1 128 smallnet
train smallnet_mnist_cifar.py 1 256 smallnet
train smallnet_mnist_cifar.py 1 512 smallnet
############################
#========multi-gpus=========#
train alexnet.py 4 512 alexnet
train alexnet.py 4 1024 alexnet
train googlenet.py 4 512 googlenet
train googlenet.py 4 1024 googlenet
benchmark/paddle/image/smallnet_mnist_cifar.py 0 → 100644
浏览文件 @ 021b3a44
#!/usr/bin/env python
from paddle.trainer_config_helpers import *
height = 32
width = 32
num_class = 10
batch_size = get_config_arg('batch_size', int, 128)
args = {'height': height, 'width': width, 'color': True, 'num_class': num_class}
define_py_data_sources2(
"train.list", None, module="provider", obj="process", args=args)
settings(
batch_size=batch_size,
learning_rate=0.01 / batch_size,
learning_method=MomentumOptimizer(0.9),
regularization=L2Regularization(0.0005 * batch_size))
# conv1
net = data_layer('data', size=height * width * 3)
net = img_conv_layer(
input=net,
filter_size=5,
num_channels=3,
num_filters=32,
stride=1,
padding=2)
net = img_pool_layer(input=net, pool_size=3, stride=2, padding=1)
# conv2
net = img_conv_layer(
input=net, filter_size=5, num_filters=32, stride=1, padding=2)
net = img_pool_layer(
input=net, pool_size=3, stride=2, padding=1, pool_type=AvgPooling())
# conv3
net = img_conv_layer(
input=net, filter_size=3, num_filters=64, stride=1, padding=1)
net = img_pool_layer(
input=net, pool_size=3, stride=2, padding=1, pool_type=AvgPooling())
net = fc_layer(input=net, size=64, act=ReluActivation())
net = fc_layer(input=net, size=10, act=SoftmaxActivation())
lab = data_layer('label', num_class)
loss = classification_cost(input=net, label=lab)
outputs(loss)
benchmark/paddle/rnn/imdb.py 0 → 100755
浏览文件 @ 021b3a44
from __future__ import print_function
import six.moves.cPickle as pickle
import gzip
import os
import numpy
def get_dataset_file(dataset, default_dataset, origin):
data_dir, data_file = os.path.split(dataset)
if (not os.path.isfile(dataset)) and data_file == default_dataset:
from six.moves import urllib
print('Downloading data from %s' % origin)
urllib.request.urlretrieve(origin, dataset)
return dataset
def create_data(path="imdb.pkl"):
if (not os.path.isfile('imdb.train.pkl')):
path = get_dataset_file(
path, "imdb.pkl",
"http://www.iro.umontreal.ca/~lisa/deep/data/imdb.pkl")
if path.endswith(".gz"):
f = gzip.open(path, 'rb')
else:
f = open(path, 'rb')
train_set = pickle.load(f)
test_set = pickle.load(f)
f.close()
pickle.dump(train_set, open('imdb.train.pkl', 'wb'))
pickle.dump(test_set, open('imdb.test.pkl', 'wb'))
if (not os.path.isfile('train.list')):
file('train.list', 'w').write('imdb.train.pkl\n')
def main():
create_data('imdb.pkl')
if __name__ == "__main__":
main()
benchmark/paddle/rnn/provider.py 0 → 100644
浏览文件 @ 021b3a44
import io, os
import random
import numpy as np
import six.moves.cPickle as pickle
from paddle.trainer.PyDataProvider2 import *
def remove_unk(x, n_words):
return [[1 if w >= n_words else w for w in sen] for sen in x]
# ==============================================================
# tensorflow uses fixed length, but PaddlePaddle can process
# variable-length. Padding is used in benchmark in order to
# compare with other platform.
# ==============================================================
def pad_sequences(sequences,
maxlen=None,
dtype='int32',
padding='post',
truncating='post',
value=0.):
lengths = [len(s) for s in sequences]
nb_samples = len(sequences)
if maxlen is None:
maxlen = np.max(lengths)
x = (np.ones((nb_samples, maxlen)) * value).astype(dtype)
for idx, s in enumerate(sequences):
if len(s) == 0:
continue # empty list was found
if truncating == 'pre':
trunc = s[-maxlen:]
elif truncating == 'post':
trunc = s[:maxlen]
else:
raise ValueError("Truncating type '%s' not understood" % padding)
if padding == 'post':
x[idx, :len(trunc)] = trunc
elif padding == 'pre':
x[idx, -len(trunc):] = trunc
else:
raise ValueError("Padding type '%s' not understood" % padding)
return x
def initHook(settings, vocab_size, pad_seq, maxlen, **kwargs):
settings.vocab_size = vocab_size
settings.pad_seq = pad_seq
settings.maxlen = maxlen
settings.input_types = [
integer_value_sequence(vocab_size), integer_value(2)
]
@provider(
init_hook=initHook, min_pool_size=-1, cache=CacheType.CACHE_PASS_IN_MEM)
def process(settings, file):
f = open(file, 'rb')
train_set = pickle.load(f)
f.close()
x, y = train_set
# remove unk, namely remove the words out of dictionary
x = remove_unk(x, settings.vocab_size)
if settings.pad_seq:
x = pad_sequences(x, maxlen=settings.maxlen, value=0.)
for i in range(len(y)):
yield map(int, x[i]), int(y[i])
benchmark/paddle/rnn/rnn.py 0 → 100755
浏览文件 @ 021b3a44
#!/usr/bin/env python
from paddle.trainer_config_helpers import *
import imdb
num_class = 2
vocab_size = 30000
fixedlen = 100
batch_size = get_config_arg('batch_size', int, 128)
lstm_num = get_config_arg('lstm_num', int, 1)
hidden_size = get_config_arg('hidden_size', int, 128)
# whether to pad sequence into fixed length
pad_seq = get_config_arg('pad_seq', bool, True)
imdb.create_data('imdb.pkl')
args = {'vocab_size': vocab_size, 'pad_seq': pad_seq, 'maxlen': fixedlen}
define_py_data_sources2(
"train.list", None, module="provider", obj="process", args=args)
settings(
batch_size=batch_size,
learning_rate=2e-3,
learning_method=AdamOptimizer(),
regularization=L2Regularization(8e-4),
gradient_clipping_threshold=25)
net = data_layer('data', size=vocab_size)
net = embedding_layer(input=net, size=128)
for i in xrange(lstm_num):
net = simple_lstm(input=net, size=hidden_size)
net = last_seq(input=net)
net = fc_layer(input=net, size=2, act=SoftmaxActivation())
lab = data_layer('label', num_class)
loss = classification_cost(input=net, label=lab)
outputs(loss)
benchmark/paddle/rnn/run.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function train() {
cfg=$1
thread=$2
args="lstm_num=${3},seq_pad=${4},hidden_size=${5},batch_size=${6}"
paddle train --job=time \
--config=$cfg \
--use_gpu=1 \
--trainer_count=$thread \
--log_period=10 \
--test_period=100 \
--num_passes=1 \
--feed_data=1 \
--config_args=$args \
>logs/rnn-pad${4}-${thread}gpu-lstm${3}-batch${6}-hid${5}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
## padding, single gpu
#-----config--gpu--lstm_num--padding--hidden_size--batch_size
## lstm_num=2, batch_size=64
train rnn.py 1 2 1 256 64
train rnn.py 1 2 1 512 64
train rnn.py 1 2 1 1280 64
## lstm_num=2, batch_size=128
train rnn.py 1 2 1 256 128
train rnn.py 1 2 1 512 128
train rnn.py 1 2 1 1280 128
## lstm_num=4, batch_size=256
train rnn.py 1 2 1 256 256
train rnn.py 1 2 1 512 256
train rnn.py 1 2 1 1280 256
#==================multi gpus=====================#
# hidden_size=256, lstm_num=2, different batch size
train rnn.py 4 2 1 256 128
train rnn.py 4 2 1 256 256
train rnn.py 4 2 1 256 512
# hidden_size=512, lstm_num=4, different batch size
train rnn.py 4 2 1 512 128
train rnn.py 4 2 1 512 256
train rnn.py 4 2 1 512 512
benchmark/tensorflow/image/alexnet.py 0 → 100644
浏览文件 @ 021b3a44
from six.moves import xrange # pylint: disable=redefined-builtin
from datetime import datetime
import math
import time
import tensorflow.python.platform
import tensorflow as tf
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 128, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('forward_only', False,
"""Only run the forward pass.""")
tf.app.flags.DEFINE_boolean('forward_backward_only', False,
"""Only run the forward-forward pass.""")
tf.app.flags.DEFINE_string('data_format', 'NCHW',
"""The data format for Convnet operations.
Can be either NHWC or NCHW.
""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
def _conv(name, inpOp, nIn, nOut, kH, kW, dH, dW, padType, wd=0.0005):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [kH, kW, nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
if wd is not None and wd > 0:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
if FLAGS.data_format == 'NCHW':
strides = [1, 1, dH, dW]
else:
strides = [1, dH, dW, 1]
conv = tf.nn.conv2d(
inpOp,
kernel,
strides,
padding=padType,
data_format=FLAGS.data_format)
biases = tf.get_variable(
name=name + '_b',
shape=[nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32)
bias = tf.reshape(
tf.nn.bias_add(
conv, biases, data_format=FLAGS.data_format),
conv.get_shape())
conv1 = tf.nn.relu(bias, name=scope)
return conv1
def _affine(name, inpOp, nIn, nOut, wd=0.0005, act=True, drop=None):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
if wd is not None and wd > 0:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
biases = tf.get_variable(
name + '_b', [nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32,
trainable=True)
affine1 = tf.nn.relu_layer(inpOp, kernel, biases, name=name) if act else \
tf.matmul(inpOp, kernel) + biases
output = tf.nn.dropout(affine1, drop) if drop else affine1
return output
def _mpool(name, inpOp, kH, kW, dH, dW):
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.max_pool(
inpOp,
ksize=ksize,
strides=strides,
padding='VALID',
data_format=FLAGS.data_format,
name=name)
def _norm(name, l_input, lsize=4):
return tf.nn.lrn(l_input,
lsize,
bias=1.0,
alpha=0.001 / 9.0,
beta=0.75,
name=name)
def loss(logits, labels):
labels = tf.cast(labels, tf.int64)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
logits, labels, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
# The total loss is defined as the cross entropy loss plus all of the weight
# decay terms (L2 loss).
return tf.add_n(tf.get_collection('losses'), name='total_loss')
def get_incoming_shape(incoming):
""" Returns the incoming data shape """
if isinstance(incoming, tf.Tensor):
return incoming.get_shape().as_list()
elif type(incoming) in [np.array, list, tuple]:
return np.shape(incoming)
else:
raise Exception("Invalid incoming layer.")
def inference(images):
conv1 = _conv('conv1', images, 3, 96, 11, 11, 4, 4, 'VALID')
pool1 = _mpool('pool1', conv1, 3, 3, 2, 2)
norm1 = _norm('norm1', pool1, lsize=5)
conv2 = _conv('conv2', norm1, 96, 256, 5, 5, 1, 1, 'SAME')
pool2 = _mpool('pool2', conv2, 3, 3, 2, 2)
norm2 = _norm('norm2', pool2, lsize=5)
conv3 = _conv('conv3', norm2, 256, 384, 3, 3, 1, 1, 'SAME')
conv4 = _conv('conv4', conv3, 384, 384, 3, 3, 1, 1, 'SAME')
conv5 = _conv('conv5', conv4, 384, 256, 3, 3, 1, 1, 'SAME')
pool5 = _mpool('pool5', conv5, 3, 3, 2, 2)
resh1 = tf.reshape(pool5, [-1, 256 * 6 * 6])
affn1 = _affine('fc6', resh1, 256 * 6 * 6, 4096, 0.5)
affn2 = _affine('fc7', affn1, 4096, 4096, 0.5)
affn3 = _affine('fc8', affn2, 4096, 1000, wd=None, act=False) # last fc
return affn3
def time_tensorflow_run(session, target, info_string):
num_steps_burn_in = 10
total_duration = 0.0
total_duration_squared = 0.0
if not isinstance(target, list):
target = [target]
target_op = tf.group(*target)
for i in xrange(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
_ = session.run(target_op)
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
print('%s: step %d, duration = %.3f' %
(datetime.now(), i - num_steps_burn_in, duration))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: %s across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), info_string, FLAGS.num_batches, mn, sd))
def _add_loss_summaries(total_loss):
"""
Generates moving average for all losses and associated summaries for
visualizing the performance of the network.
Args:
total_loss: Total loss from loss().
Returns:
loss_averages_op: op for generating moving averages of losses.
"""
# Compute the moving average of all individual losses and the total loss.
loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
losses = tf.get_collection('losses')
loss_averages_op = loss_averages.apply(losses + [total_loss])
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
tf.scalar_summary(l.op.name + ' (raw)', l)
tf.scalar_summary(l.op.name, loss_averages.average(l))
return loss_averages_op
def run_benchmark():
with tf.Graph().as_default():
with tf.device('/gpu:0'):
# Generate some dummy images.
image_size = 224
# Note that our padding definition is slightly different the cuda-convnet.
# In order to force the model to start with the same activations sizes,
# we add 3 to the image_size and employ VALID padding above.
if FLAGS.data_format == 'NCHW':
image_shape = [
FLAGS.batch_size, 3, image_size + 3, image_size + 3
]
else:
image_shape = [
FLAGS.batch_size, image_size + 3, image_size + 3, 3
]
images = tf.get_variable(
'image',
image_shape,
initializer=tf.truncated_normal_initializer(
stddev=0.1, dtype=tf.float32),
dtype=tf.float32,
trainable=False)
labels = tf.get_variable(
'label', [FLAGS.batch_size],
initializer=tf.constant_initializer(1),
dtype=tf.int32,
trainable=False)
# Build a Graph that computes the logits predictions from the
# inference model.
last_layer = inference(images)
objective = loss(last_layer, labels)
# Compute the gradient with respect to all the parameters.
# Compute gradients.
# opt = tf.train.GradientDescentOptimizer(0.001)
opt = tf.train.MomentumOptimizer(0.001, 0.9)
grads = opt.compute_gradients(objective)
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(
0.0, dtype=tf.float32),
trainable=False,
dtype=tf.float32)
apply_gradient_op = opt.apply_gradients(
grads, global_step=global_step)
# Track the moving averages of all trainable variables.
variable_averages = tf.train.ExponentialMovingAverage(0.9,
global_step)
variables_averages_op = variable_averages.apply(
tf.trainable_variables())
# Build an initialization operation.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
run_forward = True
run_forward_backward = True
if FLAGS.forward_only and FLAGS.forward_backward_only:
raise ValueError("Cannot specify --forward_only and "
"--forward_backward_only at the same time.")
if FLAGS.forward_only:
run_forward_backward = False
elif FLAGS.forward_backward_only:
run_forward = False
if run_forward:
time_tensorflow_run(sess, last_layer, "Forward")
if run_forward_backward:
with tf.control_dependencies(
[apply_gradient_op, variables_averages_op]):
train_op = tf.no_op(name='train')
time_tensorflow_run(sess, [train_op, objective],
"Forward-backward")
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/image/alexnet_multi_gpu.py 0 → 100644
浏览文件 @ 021b3a44
from six.moves import xrange # pylint: disable=redefined-builtin
from datetime import datetime
import math
import re
import time
import tensorflow.python.platform
import tensorflow as tf
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 64, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_string('data_format', 'NCHW',
"""The data format for Convnet operations.
Can be either NHWC or NCHW.
""")
tf.app.flags.DEFINE_string('train_dir', '/train_model',
"""Directory where to write event logs """
"""and checkpoint.""")
tf.app.flags.DEFINE_integer('num_gpus', 4, """How many GPUs to use.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 50000
NUM_EPOCHS_PER_DECAY = 50
INITIAL_LEARNING_RATE = 0.1
LEARNING_RATE_DECAY_FACTOR = 0.1
TOWER_NAME = 'tower'
def _conv(name, inpOp, nIn, nOut, kH, kW, dH, dW, padType, wd=0.005):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [kH, kW, nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
if wd is not None:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
if FLAGS.data_format == 'NCHW':
strides = [1, 1, dH, dW]
else:
strides = [1, dH, dW, 1]
conv = tf.nn.conv2d(
inpOp,
kernel,
strides,
padding=padType,
data_format=FLAGS.data_format)
biases = tf.get_variable(
name=name + '_b',
shape=[nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32)
bias = tf.reshape(
tf.nn.bias_add(
conv, biases, data_format=FLAGS.data_format),
conv.get_shape())
conv1 = tf.nn.relu(bias, name=scope)
return conv1
def _affine(name, inpOp, nIn, nOut, wd=0.005, act=True):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
if wd is not None:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
biases = tf.get_variable(
name + '_b', [nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32,
trainable=True)
affine1 = tf.nn.relu_layer(inpOp, kernel, biases, name=name) if act else \
tf.matmul(inpOp, kernel) + biases
return affine1
def _mpool(name, inpOp, kH, kW, dH, dW):
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.max_pool(
inpOp,
ksize=ksize,
strides=strides,
padding='VALID',
data_format=FLAGS.data_format,
name=name)
def _norm(name, l_input, lsize=4):
return tf.nn.lrn(l_input,
lsize,
bias=1.0,
alpha=0.001 / 9.0,
beta=0.75,
name=name)
def loss(logits, labels):
labels = tf.cast(labels, tf.int64)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
logits, labels, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
# The total loss is defined as the cross entropy loss plus all of the weight
# decay terms (L2 loss).
return tf.add_n(tf.get_collection('losses'), name='total_loss')
def get_incoming_shape(incoming):
""" Returns the incoming data shape """
if isinstance(incoming, tf.Tensor):
return incoming.get_shape().as_list()
elif type(incoming) in [np.array, list, tuple]:
return np.shape(incoming)
else:
raise Exception("Invalid incoming layer.")
def inference(images):
conv1 = _conv('conv1', images, 3, 96, 11, 11, 4, 4, 'VALID')
pool1 = _mpool('pool1', conv1, 3, 3, 2, 2)
norm1 = _norm('norm1', pool1, lsize=5)
conv2 = _conv('conv2', norm1, 96, 256, 5, 5, 1, 1, 'SAME')
pool2 = _mpool('pool2', conv2, 3, 3, 2, 2)
norm2 = _norm('norm2', pool2, lsize=5)
conv3 = _conv('conv3', norm2, 256, 384, 3, 3, 1, 1, 'SAME')
conv4 = _conv('conv4', conv3, 384, 384, 3, 3, 1, 1, 'SAME')
conv5 = _conv('conv5', conv4, 384, 256, 3, 3, 1, 1, 'SAME')
pool5 = _mpool('pool5', conv5, 3, 3, 2, 2)
resh1 = tf.reshape(pool5, [-1, 256 * 6 * 6])
affn1 = _affine('fc6', resh1, 256 * 6 * 6, 4096)
affn2 = _affine('fc7', affn1, 4096, 4096)
affn3 = _affine('fc8', affn2, 4096, 1000, wd=None, act=False) # last fc
return affn3
def tower_loss(scope):
"""Calculate the total loss on a single tower running the model.
Args:
scope: unique prefix string identifying the tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
image_size = 224
if FLAGS.data_format == 'NCHW':
image_shape = [FLAGS.batch_size, 3, image_size + 3, image_size + 3]
else:
image_shape = [FLAGS.batch_size, image_size + 3, image_size + 3, 3]
images = tf.get_variable(
'image',
image_shape,
initializer=tf.truncated_normal_initializer(
stddev=0.1, dtype=tf.float32),
dtype=tf.float32,
trainable=False)
labels = tf.get_variable(
'label', [FLAGS.batch_size],
initializer=tf.constant_initializer(1),
dtype=tf.int32,
trainable=False)
# Build a Graph that computes the logits predictions from the
# inference model.
last_layer = inference(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = loss(last_layer, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Compute the moving average of all individual losses and the total loss.
loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
loss_averages_op = loss_averages.apply(losses + [total_loss])
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % TOWER_NAME, '', l.op.name)
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
tf.scalar_summary(loss_name + ' (raw)', l)
tf.scalar_summary(loss_name, loss_averages.average(l))
with tf.control_dependencies([loss_averages_op]):
total_loss = tf.identity(total_loss)
return total_loss
def average_gradients(tower_grads):
"""Calculate the average gradient for each shared variable across all towers.
Note that this function provides a synchronization point across all towers.
Args:
tower_grads: List of lists of (gradient, variable) tuples. The outer list
is over individual gradients. The inner list is over the gradient
calculation for each tower.
Returns:
List of pairs of (gradient, variable) where the gradient has been averaged
across all towers.
"""
average_grads = []
for grad_and_vars in zip(*tower_grads):
# Note that each grad_and_vars looks like the following:
# ((grad0_gpu0, var0_gpu0), ... , (grad0_gpuN, var0_gpuN))
grads = []
for g, _ in grad_and_vars:
# Add 0 dimension to the gradients to represent the tower.
expanded_g = tf.expand_dims(g, 0)
# Append on a 'tower' dimension which we will average over below.
grads.append(expanded_g)
# Average over the 'tower' dimension.
grad = tf.concat(0, grads)
grad = tf.reduce_mean(grad, 0)
# Keep in mind that the Variables are redundant because they are shared
# across towers. So .. we will just return the first tower's pointer to
# the Variable.
v = grad_and_vars[0][1]
grad_and_var = (grad, v)
average_grads.append(grad_and_var)
return average_grads
def time_tensorflow_run(session, target):
num_steps_burn_in = 50
total_duration = 0.0
total_duration_squared = 0.0
for i in xrange(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
_, loss_value = session.run(target)
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
num_examples_per_step = FLAGS.batch_size * FLAGS.num_gpus
examples_per_sec = num_examples_per_step / duration
sec_per_batch = duration
format_str = (
'%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch batch_size = %d)')
print(format_str %
(datetime.now(), i - num_steps_burn_in, loss_value,
duration, sec_per_batch, num_examples_per_step))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: FwdBwd across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), FLAGS.num_batches, mn, sd))
def run_benchmark():
with tf.Graph().as_default(), tf.device('/cpu:0'):
# Create a variable to count the number of train() calls. This equals the
# number of batches processed * FLAGS.num_gpus.
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(0),
trainable=False)
# Calculate the learning rate schedule.
num_batches_per_epoch = (NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN /
FLAGS.batch_size)
decay_steps = int(num_batches_per_epoch * NUM_EPOCHS_PER_DECAY)
# Decay the learning rate exponentially based on the number of steps.
lr = tf.train.exponential_decay(
INITIAL_LEARNING_RATE,
global_step,
decay_steps,
LEARNING_RATE_DECAY_FACTOR,
staircase=True)
# Create an optimizer that performs gradient descent.
opt = tf.train.MomentumOptimizer(lr, 0.9)
# Calculate the gradients for each model tower.
tower_grads = []
for i in xrange(FLAGS.num_gpus):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (TOWER_NAME, i)) as scope:
# Calculate the loss for one tower of the model. This function
# constructs the entire model but shares the variables across
# all towers.
loss = tower_loss(scope)
# Reuse variables for the next tower.
tf.get_variable_scope().reuse_variables()
# Retain the summaries from the final tower.
summaries = tf.get_collection(tf.GraphKeys.SUMMARIES, scope)
# Calculate the gradients for the batch of data on this tower.
grads = opt.compute_gradients(loss)
# Keep track of the gradients across all towers.
tower_grads.append(grads)
# We must calculate the mean of each gradient. Note that this is the
# synchronization point across all towers.
grads = average_gradients(tower_grads)
# Apply the gradients to adjust the shared variables.
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
# Group all updates to into a single train op.
train_op = tf.group(apply_gradient_op)
# Build an initialization operation.
init = tf.initialize_all_variables()
# Start running operations on the Graph. allow_soft_placement must be set to
# True to build towers on GPU, as some of the ops do not have GPU
# implementations.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
time_tensorflow_run(sess, [train_op, loss])
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/image/googlenet.py 0 → 100644
浏览文件 @ 021b3a44
from six.moves import xrange
from datetime import datetime
import math
import time
import tensorflow.python.platform
import tensorflow as tf
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 128, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('forward_only', False,
"""Only run the forward pass.""")
tf.app.flags.DEFINE_boolean('forward_backward_only', False,
"""Only run the forward-forward pass.""")
tf.app.flags.DEFINE_string('data_format', 'NCHW',
"""The data format for Convnet operations.
Can be either NHWC or NCHW.
""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
parameters = []
conv_counter = 1
pool_counter = 1
affine_counter = 1
def _conv(inpOp, nIn, nOut, kH, kW, dH, dW, padType, wd=0.0005):
global conv_counter
global parameters
name = 'conv' + str(conv_counter)
conv_counter += 1
with tf.name_scope(name) as scope:
kernel = tf.Variable(
tf.truncated_normal(
[kH, kW, nIn, nOut], dtype=tf.float32, stddev=1e-1),
name='weights')
if wd is not None and wd > 0:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
if FLAGS.data_format == 'NCHW':
strides = [1, 1, dH, dW]
else:
strides = [1, dH, dW, 1]
conv = tf.nn.conv2d(
inpOp,
kernel,
strides,
padding=padType,
data_format=FLAGS.data_format)
biases = tf.Variable(
tf.constant(
0.0, shape=[nOut], dtype=tf.float32),
trainable=True,
name='biases')
bias = tf.reshape(
tf.nn.bias_add(
conv, biases, data_format=FLAGS.data_format),
conv.get_shape())
conv1 = tf.nn.relu(bias, name=scope)
parameters += [kernel, biases]
return conv1
def _affine(inpOp, nIn, nOut, act=True, wd=0.0005):
global affine_counter
global parameters
name = 'affine' + str(affine_counter)
affine_counter += 1
with tf.name_scope(name) as scope:
kernel = tf.Variable(
tf.truncated_normal(
[nIn, nOut], dtype=tf.float32, stddev=1e-1),
name='weights')
if wd is not None and wd > 0:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
biases = tf.Variable(
tf.constant(
0.0, shape=[nOut], dtype=tf.float32),
trainable=True,
name='biases')
affine1 = tf.nn.relu_layer(
inpOp, kernel, biases,
name=name) if act else tf.matmul(inpOp, kernel) + biases
parameters += [kernel, biases]
return affine1
def _mpool(inpOp, kH, kW, dH, dW, padding):
global pool_counter
global parameters
name = 'pool' + str(pool_counter)
pool_counter += 1
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.max_pool(
inpOp,
ksize=ksize,
strides=strides,
padding=padding,
data_format=FLAGS.data_format,
name=name)
def _apool(inpOp, kH, kW, dH, dW, padding):
global pool_counter
global parameters
name = 'pool' + str(pool_counter)
pool_counter += 1
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.avg_pool(
inpOp,
ksize=ksize,
strides=strides,
padding=padding,
data_format=FLAGS.data_format,
name=name)
def _inception(inp, inSize, o1s, o2s1, o2s2, o3s1, o3s2, o4s1, o4s2):
conv1 = _conv(inp, inSize, o1s, 1, 1, 1, 1, 'VALID')
conv3_ = _conv(inp, inSize, o2s1, 1, 1, 1, 1, 'VALID')
conv3 = _conv(conv3_, o2s1, o2s2, 3, 3, 1, 1, 'SAME')
conv5_ = _conv(inp, inSize, o3s1, 1, 1, 1, 1, 'VALID')
conv5 = _conv(conv5_, o3s1, o3s2, 5, 5, 1, 1, 'SAME')
pool_ = _mpool(inp, o4s1, o4s1, 1, 1, 'SAME')
pool = _conv(pool_, inSize, o4s2, 1, 1, 1, 1, 'VALID')
if FLAGS.data_format == 'NCHW':
channel_dim = 1
else:
channel_dim = 3
incept = tf.concat(channel_dim, [conv1, conv3, conv5, pool])
return incept
def loss(logits, labels):
batch_size = tf.size(labels)
labels = tf.expand_dims(labels, 1)
indices = tf.expand_dims(tf.range(0, batch_size, 1), 1)
concated = tf.concat(1, [indices, labels])
onehot_labels = tf.sparse_to_dense(concated,
tf.pack([batch_size, 1000]), 1.0, 0.0)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits, onehot_labels, name='xentropy')
loss = tf.reduce_mean(cross_entropy, name='xentropy_mean')
return loss
def inference(images):
# stage 1
conv1 = _conv(images, 3, 64, 7, 7, 2, 2, 'SAME')
pool1 = _mpool(conv1, 3, 3, 2, 2, 'SAME')
# stage 2
conv2 = _conv(pool1, 64, 64, 1, 1, 1, 1, 'VALID')
conv3 = _conv(conv2, 64, 192, 3, 3, 1, 1, 'SAME')
pool3 = _mpool(conv3, 3, 3, 2, 2, 'SAME')
# stage 3
incept3a = _inception(pool3, 192, 64, 96, 128, 16, 32, 3, 32)
incept3b = _inception(incept3a, 256, 128, 128, 192, 32, 96, 3, 64)
pool4 = _mpool(incept3b, 3, 3, 2, 2, 'SAME')
# stage 4
incept4a = _inception(pool4, 480, 192, 96, 208, 16, 48, 3, 64)
incept4b = _inception(incept4a, 512, 160, 112, 224, 24, 64, 3, 64)
incept4c = _inception(incept4b, 512, 128, 128, 256, 24, 64, 3, 64)
incept4d = _inception(incept4c, 512, 112, 144, 288, 32, 64, 3, 64)
incept4e = _inception(incept4d, 528, 256, 160, 320, 32, 128, 3, 128)
pool5 = _mpool(incept4e, 3, 3, 2, 2, 'SAME')
# stage 5
incept5a = _inception(pool5, 832, 256, 160, 320, 32, 128, 3, 128)
incept5b = _inception(incept5a, 832, 384, 192, 384, 48, 128, 3, 128)
pool6 = _apool(incept5b, 7, 7, 1, 1, 'VALID')
# output 1
resh1 = tf.reshape(pool6, [-1, 1024])
drop = tf.nn.dropout(resh1, 0.4)
affn1 = _affine(resh1, 1024, 1000, act=False)
return affn1
def time_tensorflow_run(session, target, info_string):
num_steps_burn_in = 10
total_duration = 0.0
total_duration_squared = 0.0
if not isinstance(target, list):
target = [target]
target_op = tf.group(*target)
for i in range(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
_ = session.run(target_op)
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
print('%s: step %d, duration = %.3f' %
(datetime.now(), i - num_steps_burn_in, duration))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: %s across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), info_string, FLAGS.num_batches, mn, sd))
def run_benchmark():
global parameters
with tf.Graph().as_default():
# Generate some dummy images.
image_size = 224
if FLAGS.data_format == 'NCHW':
image_shape = [FLAGS.batch_size, 3, image_size, image_size]
else:
image_shape = [FLAGS.batch_size, image_size, image_size, 3]
images = tf.get_variable(
'image',
image_shape,
initializer=tf.truncated_normal_initializer(
stddev=0.1, dtype=tf.float32),
dtype=tf.float32,
trainable=False)
labels = tf.get_variable(
'label', [FLAGS.batch_size],
initializer=tf.constant_initializer(1),
dtype=tf.int32,
trainable=False)
# Build a Graph that computes the logits predictions from the
# inference model.
last_layer = inference(images)
objective = loss(last_layer, labels)
# Compute gradients.
# opt = tf.train.GradientDescentOptimizer(0.001)
opt = tf.train.MomentumOptimizer(0.001, 0.9)
grads = opt.compute_gradients(objective)
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(
0.0, dtype=tf.float32),
trainable=False,
dtype=tf.float32)
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
# Track the moving averages of all trainable variables.
variable_averages = tf.train.ExponentialMovingAverage(0.9, global_step)
variables_averages_op = variable_averages.apply(tf.trainable_variables(
))
# Build an initialization operation.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
run_forward = True
run_forward_backward = True
if FLAGS.forward_only and FLAGS.forward_backward_only:
raise ValueError("Cannot specify --forward_only and "
"--forward_backward_only at the same time.")
if FLAGS.forward_only:
run_forward_backward = False
elif FLAGS.forward_backward_only:
run_forward = False
if run_forward:
# Run the forward benchmark.
time_tensorflow_run(sess, last_layer, "Forward")
if run_forward_backward:
with tf.control_dependencies(
[apply_gradient_op, variables_averages_op]):
train_op = tf.no_op(name='train')
time_tensorflow_run(sess, [train_op, objective], "Forward-backward")
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/image/googlenet_multi_gpu.py 0 → 100644
浏览文件 @ 021b3a44
from six.moves import xrange # pylint: disable=redefined-builtin
from datetime import datetime
import math
import re
import time
import tensorflow.python.platform
import tensorflow as tf
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 64, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_string('data_format', 'NCHW',
"""The data format for Convnet operations.
Can be either NHWC or NCHW.
""")
tf.app.flags.DEFINE_string('train_dir', '/train_model',
"""Directory where to write event logs """
"""and checkpoint.""")
tf.app.flags.DEFINE_integer('num_gpus', 4, """How many GPUs to use.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 50000
NUM_EPOCHS_PER_DECAY = 50
INITIAL_LEARNING_RATE = 0.1
LEARNING_RATE_DECAY_FACTOR = 0.1
TOWER_NAME = 'tower'
def _conv(name, inpOp, nIn, nOut, kH, kW, dH, dW, padType, wd=0.005):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [kH, kW, nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
if wd is not None:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
if FLAGS.data_format == 'NCHW':
strides = [1, 1, dH, dW]
else:
strides = [1, dH, dW, 1]
conv = tf.nn.conv2d(
inpOp,
kernel,
strides,
padding=padType,
data_format=FLAGS.data_format)
biases = tf.get_variable(
name=name + '_b',
shape=[nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32)
bias = tf.reshape(
tf.nn.bias_add(
conv, biases, data_format=FLAGS.data_format),
conv.get_shape())
conv1 = tf.nn.relu(bias, name=scope)
return conv1
def _affine(name, inpOp, nIn, nOut, wd=0.005, act=True):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
if wd is not None:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
biases = tf.get_variable(
name + '_b', [nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32,
trainable=True)
affine1 = tf.nn.relu_layer(inpOp, kernel, biases, name=name) if act else \
tf.matmul(inpOp, kernel) + biases
return affine1
def _mpool(name, inpOp, kH, kW, dH, dW, padding):
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.max_pool(
inpOp,
ksize=ksize,
strides=strides,
padding=padding,
data_format=FLAGS.data_format,
name=name)
def _apool(name, inpOp, kH, kW, dH, dW, padding):
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.avg_pool(
inpOp,
ksize=ksize,
strides=strides,
padding=padding,
data_format=FLAGS.data_format,
name=name)
def loss(logits, labels):
labels = tf.cast(labels, tf.int64)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
logits, labels, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
# The total loss is defined as the cross entropy loss plus all of the weight
# decay terms (L2 loss).
return tf.add_n(tf.get_collection('losses'), name='total_loss')
def get_incoming_shape(incoming):
""" Returns the incoming data shape """
if isinstance(incoming, tf.Tensor):
return incoming.get_shape().as_list()
elif type(incoming) in [np.array, list, tuple]:
return np.shape(incoming)
else:
raise Exception("Invalid incoming layer.")
def _inception(name, inp, inSize, o1s, o2s1, o2s2, o3s1, o3s2, o4s1, o4s2):
conv1 = _conv(name + '_1', inp, inSize, o1s, 1, 1, 1, 1, 'VALID')
conv3_ = _conv(name + '_3r', inp, inSize, o2s1, 1, 1, 1, 1, 'VALID')
conv3 = _conv(name + '_3', conv3_, o2s1, o2s2, 3, 3, 1, 1, 'SAME')
conv5_ = _conv(name + '_5r', inp, inSize, o3s1, 1, 1, 1, 1, 'VALID')
conv5 = _conv(name + '5', conv5_, o3s1, o3s2, 5, 5, 1, 1, 'SAME')
pool_ = _mpool(name + 'pool', inp, o4s1, o4s1, 1, 1, 'SAME')
pool = _conv(name + 'proj', pool_, inSize, o4s2, 1, 1, 1, 1, 'VALID')
if FLAGS.data_format == 'NCHW':
channel_dim = 1
else:
channel_dim = 3
incept = tf.concat(channel_dim, [conv1, conv3, conv5, pool])
return incept
def inference(images):
# stage 1
conv1 = _conv('conv1', images, 3, 64, 7, 7, 2, 2, 'SAME')
pool1 = _mpool('pool1', conv1, 3, 3, 2, 2, 'SAME')
# stage 2
conv2 = _conv('conv2', pool1, 64, 64, 1, 1, 1, 1, 'VALID')
conv3 = _conv('conv3', conv2, 64, 192, 3, 3, 1, 1, 'SAME')
pool3 = _mpool('pool3', conv3, 3, 3, 2, 2, 'SAME')
# stage 3
incept3a = _inception('ince3a', pool3, 192, 64, 96, 128, 16, 32, 3, 32)
incept3b = _inception('ince3b', incept3a, 256, 128, 128, 192, 32, 96, 3, 64)
pool4 = _mpool('pool4', incept3b, 3, 3, 2, 2, 'SAME')
# stage 4
incept4a = _inception('ince4a', pool4, 480, 192, 96, 208, 16, 48, 3, 64)
incept4b = _inception('ince4b', incept4a, 512, 160, 112, 224, 24, 64, 3, 64)
incept4c = _inception('ince4c', incept4b, 512, 128, 128, 256, 24, 64, 3, 64)
incept4d = _inception('ince4d', incept4c, 512, 112, 144, 288, 32, 64, 3, 64)
incept4e = _inception('ince4e', incept4d, 528, 256, 160, 320, 32, 128, 3,
128)
pool5 = _mpool('pool5', incept4e, 3, 3, 2, 2, 'SAME')
# stage 5
incept5a = _inception('ince5a', pool5, 832, 256, 160, 320, 32, 128, 3, 128)
incept5b = _inception('ince5b', incept5a, 832, 384, 192, 384, 48, 128, 3,
128)
pool6 = _apool('pool6', incept5b, 7, 7, 1, 1, 'VALID')
# output 1
resh1 = tf.reshape(pool6, [-1, 1024])
drop = tf.nn.dropout(resh1, 0.4)
affn1 = _affine('fc_out', resh1, 1024, 1000, act=False)
return affn1
def tower_loss(scope):
"""Calculate the total loss on a single tower running the model.
Args:
scope: unique prefix string identifying the tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
image_size = 224
if FLAGS.data_format == 'NCHW':
image_shape = [FLAGS.batch_size, 3, image_size, image_size]
else:
image_shape = [FLAGS.batch_size, image_size, image_size, 3]
images = tf.get_variable(
'image',
image_shape,
initializer=tf.truncated_normal_initializer(
stddev=0.1, dtype=tf.float32),
dtype=tf.float32,
trainable=False)
labels = tf.get_variable(
'label', [FLAGS.batch_size],
initializer=tf.constant_initializer(1),
dtype=tf.int32,
trainable=False)
# Build a Graph that computes the logits predictions from the
# inference model.
last_layer = inference(images)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
_ = loss(last_layer, labels)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Compute the moving average of all individual losses and the total loss.
loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
loss_averages_op = loss_averages.apply(losses + [total_loss])
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % TOWER_NAME, '', l.op.name)
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
tf.scalar_summary(loss_name + ' (raw)', l)
tf.scalar_summary(loss_name, loss_averages.average(l))
with tf.control_dependencies([loss_averages_op]):
total_loss = tf.identity(total_loss)
return total_loss
def average_gradients(tower_grads):
"""Calculate the average gradient for each shared variable across all towers.
Note that this function provides a synchronization point across all towers.
Args:
tower_grads: List of lists of (gradient, variable) tuples. The outer list
is over individual gradients. The inner list is over the gradient
calculation for each tower.
Returns:
List of pairs of (gradient, variable) where the gradient has been averaged
across all towers.
"""
average_grads = []
for grad_and_vars in zip(*tower_grads):
# Note that each grad_and_vars looks like the following:
# ((grad0_gpu0, var0_gpu0), ... , (grad0_gpuN, var0_gpuN))
grads = []
for g, _ in grad_and_vars:
# Add 0 dimension to the gradients to represent the tower.
expanded_g = tf.expand_dims(g, 0)
# Append on a 'tower' dimension which we will average over below.
grads.append(expanded_g)
# Average over the 'tower' dimension.
grad = tf.concat(0, grads)
grad = tf.reduce_mean(grad, 0)
# Keep in mind that the Variables are redundant because they are shared
# across towers. So .. we will just return the first tower's pointer to
# the Variable.
v = grad_and_vars[0][1]
grad_and_var = (grad, v)
average_grads.append(grad_and_var)
return average_grads
def time_tensorflow_run(session, target):
num_steps_burn_in = 50
total_duration = 0.0
total_duration_squared = 0.0
for i in xrange(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
_, loss_value = session.run(target)
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
num_examples_per_step = FLAGS.batch_size * FLAGS.num_gpus
examples_per_sec = num_examples_per_step / duration
sec_per_batch = duration
format_str = (
'%s: step %d, loss = %.2f (%.1f examples/sec; %.3f '
'sec/batch batch_size = %d)')
print(format_str %
(datetime.now(), i - num_steps_burn_in, loss_value,
duration, sec_per_batch, num_examples_per_step))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: FwdBwd across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), FLAGS.num_batches, mn, sd))
def run_benchmark():
with tf.Graph().as_default(), tf.device('/cpu:0'):
# Create a variable to count the number of train() calls. This equals the
# number of batches processed * FLAGS.num_gpus.
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(0),
trainable=False)
# Calculate the learning rate schedule.
num_batches_per_epoch = (NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN /
FLAGS.batch_size)
decay_steps = int(num_batches_per_epoch * NUM_EPOCHS_PER_DECAY)
# Decay the learning rate exponentially based on the number of steps.
lr = tf.train.exponential_decay(
INITIAL_LEARNING_RATE,
global_step,
decay_steps,
LEARNING_RATE_DECAY_FACTOR,
staircase=True)
# Create an optimizer that performs gradient descent.
opt = tf.train.MomentumOptimizer(lr, 0.9)
# Calculate the gradients for each model tower.
tower_grads = []
for i in xrange(FLAGS.num_gpus):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (TOWER_NAME, i)) as scope:
# Calculate the loss for one tower of the model. This function
# constructs the entire model but shares the variables across
# all towers.
loss = tower_loss(scope)
# Reuse variables for the next tower.
tf.get_variable_scope().reuse_variables()
# Retain the summaries from the final tower.
summaries = tf.get_collection(tf.GraphKeys.SUMMARIES, scope)
# Calculate the gradients for the batch of data on this tower.
grads = opt.compute_gradients(loss)
# Keep track of the gradients across all towers.
tower_grads.append(grads)
# We must calculate the mean of each gradient. Note that this is the
# synchronization point across all towers.
grads = average_gradients(tower_grads)
# Apply the gradients to adjust the shared variables.
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
# Group all updates to into a single train op.
train_op = tf.group(apply_gradient_op)
# Build an initialization operation.
init = tf.initialize_all_variables()
# Start running operations on the Graph. allow_soft_placement must be set to
# True to build towers on GPU, as some of the ops do not have GPU
# implementations.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
time_tensorflow_run(sess, [train_op, loss])
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/image/run.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function test() {
cfg=$1
batch_size=$2
prefix=$3
python $cfg --batch_size=$batch_size > logs/${prefix}-1gpu-${batch_size}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
# alexnet
test alexnet.py 64 alexnet
test alexnet.py 128 alexnet
test alexnet.py 256 alexnet
test alexnet.py 512 alexnet
# googlenet
test googlenet.py 64 googlenet
test googlenet.py 128 googlenet
# smallnet
test smallnet_mnist_cifar.py 64 smallnet
test smallnet_mnist_cifar.py 128 smallnet
test smallnet_mnist_cifar.py 256 smallnet
test smallnet_mnist_cifar.py 512 smallnet
benchmark/tensorflow/image/run_multi.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function test() {
cfg=$1
num_gpu=$2
batch_size=$3
batch_per_gpu=`expr ${batch_size} / ${num_gpu}`
prefix=$4
python $cfg --num_gpus=$num_gpu --batch_size=${batch_per_gpu} > logs/${prefix}-4gpu-${batch_size}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
# alexnet
test alexnet_multi_gpu.py 4 512 alexnet
test alexnet_multi_gpu.py 4 1024 alexnet
# googlenet
test googlenet_multi_gpu.py 4 512 alexnet
test googlenet_multi_gpu.py 4 1024 alexnet
benchmark/tensorflow/image/smallnet_mnist_cifar.py 0 → 100644
浏览文件 @ 021b3a44
from six.moves import xrange # pylint: disable=redefined-builtin
from datetime import datetime
import math
import time
import tensorflow.python.platform
import tensorflow as tf
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 128, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('forward_only', False,
"""Only run the forward pass.""")
tf.app.flags.DEFINE_boolean('forward_backward_only', False,
"""Only run the forward-forward pass.""")
tf.app.flags.DEFINE_string('data_format', 'NCHW',
"""The data format for Convnet operations.
Can be either NHWC or NCHW.
""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
parameters = []
conv_counter = 1
pool_counter = 1
affine_counter = 1
def _conv(inpOp, nIn, nOut, kH, kW, dH, dW, padType, wd=0.005, act=True):
global conv_counter
global parameters
name = 'conv' + str(conv_counter)
conv_counter += 1
with tf.name_scope(name) as scope:
kernel = tf.Variable(
tf.truncated_normal(
[kH, kW, nIn, nOut], dtype=tf.float32, stddev=1e-1),
name='weights')
if wd is not None:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
if FLAGS.data_format == 'NCHW':
strides = [1, 1, dH, dW]
else:
strides = [1, dH, dW, 1]
conv = tf.nn.conv2d(
inpOp,
kernel,
strides,
padding=padType,
data_format=FLAGS.data_format)
biases = tf.Variable(
tf.constant(
0.0, shape=[nOut], dtype=tf.float32),
trainable=True,
name='biases')
bias = tf.reshape(
tf.nn.bias_add(
conv, biases, data_format=FLAGS.data_format),
conv.get_shape())
conv1 = tf.nn.relu(bias, name=scope) if act else bias
parameters += [kernel, biases]
return conv1
def _affine(inpOp, nIn, nOut, wd=None, act=True):
global affine_counter
global parameters
name = 'affine' + str(affine_counter)
affine_counter += 1
with tf.name_scope(name) as scope:
kernel = tf.Variable(
tf.truncated_normal(
[nIn, nOut], dtype=tf.float32, stddev=1e-1),
name='weights')
if wd is not None:
weight_decay = tf.mul(tf.nn.l2_loss(kernel), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
biases = tf.Variable(
tf.constant(
0.0, shape=[nOut], dtype=tf.float32),
trainable=True,
name='biases')
affine1 = tf.nn.relu_layer(
inpOp, kernel, biases,
name=name) if act else tf.matmul(inpOp, kernel) + biases
parameters += [kernel, biases]
return affine1
def _mpool(inpOp, kH, kW, dH, dW, padding):
global pool_counter
global parameters
name = 'pool' + str(pool_counter)
pool_counter += 1
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.max_pool(
inpOp,
ksize=ksize,
strides=strides,
padding=padding,
data_format=FLAGS.data_format,
name=name)
def _apool(inpOp, kH, kW, dH, dW, padding):
global pool_counter
global parameters
name = 'pool' + str(pool_counter)
pool_counter += 1
if FLAGS.data_format == 'NCHW':
ksize = [1, 1, kH, kW]
strides = [1, 1, dH, dW]
else:
ksize = [1, kH, kW, 1]
strides = [1, dH, dW, 1]
return tf.nn.avg_pool(
inpOp,
ksize=ksize,
strides=strides,
padding=padding,
data_format=FLAGS.data_format,
name=name)
def _norm(name, l_input, lsize=4):
return tf.nn.lrn(l_input,
lsize,
bias=1.0,
alpha=0.001 / 9.0,
beta=0.75,
name=name)
def loss(logits, labels):
batch_size = tf.size(labels)
labels = tf.expand_dims(labels, 1)
indices = tf.expand_dims(tf.range(0, batch_size, 1), 1)
concated = tf.concat(1, [indices, labels])
onehot_labels = tf.sparse_to_dense(concated,
tf.pack([batch_size, 10]), 1.0, 0.0)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits, onehot_labels, name='xentropy')
loss = tf.reduce_mean(cross_entropy, name='xentropy_mean')
return loss
def get_incoming_shape(incoming):
""" Returns the incoming data shape """
if isinstance(incoming, tf.Tensor):
return incoming.get_shape().as_list()
elif type(incoming) in [np.array, list, tuple]:
return np.shape(incoming)
else:
raise Exception("Invalid incoming layer.")
def inference(images):
conv1 = _conv(images, 3, 32, 5, 5, 1, 1, 'SAME')
pool1 = _mpool(conv1, 3, 3, 2, 2, 'SAME')
conv2 = _conv(pool1, 32, 32, 5, 5, 1, 1, 'SAME')
pool2 = _apool(conv2, 3, 3, 2, 2, 'SAME')
conv3 = _conv(pool2, 32, 64, 5, 5, 1, 1, 'SAME')
pool3 = _apool(conv3, 3, 3, 2, 2, 'SAME')
resh1 = tf.reshape(pool3, [-1, 64 * 4 * 4])
affn1 = _affine(resh1, 64 * 4 * 4, 64)
affn2 = _affine(affn1, 64, 10, act=False)
print('conv1:', get_incoming_shape(conv1))
print('pool1:', get_incoming_shape(pool1))
print('conv2:', get_incoming_shape(conv2))
print('pool2:', get_incoming_shape(pool2))
print('conv3:', get_incoming_shape(conv3))
print('pool3:', get_incoming_shape(pool3))
return affn2
def time_tensorflow_run(session, target, info_string):
num_steps_burn_in = 10
total_duration = 0.0
total_duration_squared = 0.0
if not isinstance(target, list):
target = [target]
target_op = tf.group(*target)
for i in xrange(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
_ = session.run(target_op)
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
print('%s: step %d, duration = %.3f' %
(datetime.now(), i - num_steps_burn_in, duration))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: %s across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), info_string, FLAGS.num_batches, mn, sd))
def run_benchmark():
global parameters
with tf.Graph().as_default():
# Generate some dummy images.
image_size = 32
# Note that our padding definition is slightly different the cuda-convnet.
# In order to force the model to start with the same activations sizes,
# we add 3 to the image_size and employ VALID padding above.
if FLAGS.data_format == 'NCHW':
image_shape = [FLAGS.batch_size, 3, image_size, image_size]
else:
image_shape = [FLAGS.batch_size, image_size, image_size, 3]
images = tf.get_variable(
'image',
image_shape,
initializer=tf.truncated_normal_initializer(
stddev=0.1, dtype=tf.float32),
dtype=tf.float32,
trainable=False)
labels = tf.get_variable(
'label', [FLAGS.batch_size],
initializer=tf.constant_initializer(1),
dtype=tf.int32,
trainable=False)
# Build a Graph that computes the logits predictions from the
# inference model.
last_layer = inference(images)
objective = loss(last_layer, labels)
# Compute gradients.
opt = tf.train.MomentumOptimizer(0.001, 0.9)
grads = opt.compute_gradients(objective)
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(
0.0, dtype=tf.float32),
trainable=False,
dtype=tf.float32)
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
# Track the moving averages of all trainable variables.
variable_averages = tf.train.ExponentialMovingAverage(0.9, global_step)
variables_averages_op = variable_averages.apply(tf.trainable_variables(
))
# Build an initialization operation.
init = tf.initialize_all_variables()
# Start running operations on the Graph.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
run_forward = True
run_forward_backward = True
if FLAGS.forward_only and FLAGS.forward_backward_only:
raise ValueError("Cannot specify --forward_only and "
"--forward_backward_only at the same time.")
if FLAGS.forward_only:
run_forward_backward = False
elif FLAGS.forward_backward_only:
run_forward = False
if run_forward:
# Run the forward benchmark.
time_tensorflow_run(sess, last_layer, "Forward")
if run_forward_backward:
with tf.control_dependencies(
[apply_gradient_op, variables_averages_op]):
train_op = tf.no_op(name='train')
time_tensorflow_run(sess, [train_op, objective], "Forward-backward")
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/rnn/README.md 0 → 100644
浏览文件 @ 021b3a44
You also should install tflearn:
```bash
pip install -r requirements.txt
```
benchmark/tensorflow/rnn/reader.py 0 → 100755
浏览文件 @ 021b3a44
import os.path
import io
import numpy as np
import tensorflow as tf
# tflearn
import tflearn
from tflearn.data_utils import to_categorical, pad_sequences
from tflearn.datasets import imdb
FLAGS = tf.app.flags.FLAGS
class DataSet(object):
def __init__(self, data, labels):
assert data.shape[0] == labels.shape[0], (
'data.shape: %s labels.shape: %s' % (data.shape, labels.shape))
self._num_examples = data.shape[0]
self._data = data
self._labels = labels
self._epochs_completed = 0
self._index_in_epoch = 0
@property
def data(self):
return self._data
@property
def labels(self):
return self._labels
@property
def num_examples(self):
return self._num_examples
@property
def epochs_completed(self):
return self._epochs_completed
def next_batch(self, batch_size):
assert batch_size <= self._num_examples
start = self._index_in_epoch
self._index_in_epoch += batch_size
if self._index_in_epoch > self._num_examples:
# Finished epoch
self._epochs_completed += 1
# Shuffle the data
perm = np.arange(self._num_examples)
np.random.shuffle(perm)
self._data = self._data[perm]
self._labels = self._labels[perm]
# Start next epoch
start = 0
self._index_in_epoch = batch_size
end = self._index_in_epoch
return self._data[start:end], self._labels[start:end]
def create_datasets(file_path, vocab_size=30000, val_fraction=0.0):
# IMDB Dataset loading
train, test, _ = imdb.load_data(
path=file_path,
n_words=vocab_size,
valid_portion=val_fraction,
sort_by_len=False)
trainX, trainY = train
testX, testY = test
# Data preprocessing
# Sequence padding
trainX = pad_sequences(trainX, maxlen=FLAGS.max_len, value=0.)
testX = pad_sequences(testX, maxlen=FLAGS.max_len, value=0.)
# Converting labels to binary vectors
trainY = to_categorical(trainY, nb_classes=2)
testY = to_categorical(testY, nb_classes=2)
train_dataset = DataSet(trainX, trainY)
return train_dataset
def main():
create_datasets('imdb.pkl')
if __name__ == "__main__":
main()
benchmark/tensorflow/rnn/rnn.py 0 → 100755
浏览文件 @ 021b3a44
#!/usr/bin/env python
from six.moves import xrange # pylint: disable=redefined-builtin
import math
import time
import numpy as np
from datetime import datetime
import reader
import tensorflow as tf
from tensorflow.python.ops import rnn
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 128, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('num_layers', 1, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('max_len', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('forward_only', False,
"""Only run the forward pass.""")
tf.app.flags.DEFINE_boolean('forward_backward_only', False,
"""Only run the forward-forward pass.""")
tf.app.flags.DEFINE_integer('hidden_size', 128, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('emb_size', 128, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
VOCAB_SIZE = 30000
NUM_CLASS = 2
def get_feed_dict(x_data, y_data=None):
feed_dict = {}
if y_data is not None:
feed_dict[y_input] = y_data
for i in xrange(x_data.shape[0]):
feed_dict[x_input[i]] = x_data[i, :, :]
return feed_dict
def get_incoming_shape(incoming):
""" Returns the incoming data shape """
if isinstance(incoming, tf.Tensor):
return incoming.get_shape().as_list()
elif type(incoming) in [np.array, list, tuple]:
return np.shape(incoming)
else:
raise Exception("Invalid incoming layer.")
# Note input * W is done in LSTMCell,
# which is different from PaddlePaddle
def single_lstm(name,
incoming,
n_units,
use_peepholes=True,
return_seq=False,
return_state=False):
with tf.name_scope(name) as scope:
cell = tf.nn.rnn_cell.LSTMCell(n_units, use_peepholes=use_peepholes)
output, _cell_state = rnn.rnn(cell, incoming, dtype=tf.float32)
out = output if return_seq else output[-1]
return (out, _cell_state) if return_state else out
def lstm(name,
incoming,
n_units,
use_peepholes=True,
return_seq=False,
return_state=False,
num_layers=1):
with tf.name_scope(name) as scope:
lstm_cell = tf.nn.rnn_cell.LSTMCell(
n_units, use_peepholes=use_peepholes)
cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * num_layers)
initial_state = cell.zero_state(FLAGS.batch_size, dtype=tf.float32)
if not isinstance(incoming, list):
# if the input is embeding, the Tensor shape : [None, time_step, emb_size]
incoming = [
tf.squeeze(input_, [1])
for input_ in tf.split(1, FLAGS.max_len, incoming)
]
outputs, state = tf.nn.rnn(cell,
incoming,
initial_state=initial_state,
dtype=tf.float32)
out = outputs if return_seq else outputs[-1]
return (out, _cell_state) if return_state else out
def embedding(name, incoming, vocab_size, emb_size):
with tf.name_scope(name) as scope:
#with tf.device("/cpu:0"):
embedding = tf.get_variable(
name + '_emb', [vocab_size, emb_size], dtype=tf.float32)
out = tf.nn.embedding_lookup(embedding, incoming)
return out
def fc(name, inpOp, nIn, nOut, act=True):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
biases = tf.get_variable(
name + '_b', [nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32,
trainable=True)
net = tf.nn.relu_layer(inpOp, kernel, biases, name=name) if act else \
tf.matmul(inpOp, kernel) + biases
return net
def inference(seq):
net = embedding('emb', seq, VOCAB_SIZE, FLAGS.emb_size)
print "emb:", get_incoming_shape(net)
net = lstm('lstm', net, FLAGS.hidden_size, num_layers=FLAGS.num_layers)
print "lstm:", get_incoming_shape(net)
net = fc('fc1', net, FLAGS.hidden_size, 2)
return net
def loss(logits, labels):
# one label index for one sample
labels = tf.cast(labels, tf.float32)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits, labels, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
return tf.add_n(tf.get_collection('losses'), name='total_loss')
def time_tensorflow_run(session, target, x_input, y_input, info_string):
num_steps_burn_in = 50
total_duration = 0.0
total_duration_squared = 0.0
if not isinstance(target, list):
target = [target]
target_op = tf.group(*target)
train_dataset = reader.create_datasets("imdb.pkl", VOCAB_SIZE)
for i in xrange(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
data, label = train_dataset.next_batch(FLAGS.batch_size)
_ = session.run(target_op, feed_dict={x_input: data, y_input: label})
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
print('%s: step %d, duration = %.3f' %
(datetime.now(), i - num_steps_burn_in, duration))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: %s across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), info_string, FLAGS.num_batches, mn, sd))
def run_benchmark():
with tf.Graph().as_default():
global_step = 0
with tf.device('/cpu:0'):
global_step = tf.Variable(0, trainable=False)
with tf.device('/gpu:0'):
#x_input = tf.placeholder(tf.int32, [None, FLAGS.max_len], name="x_input")
#y_input = tf.placeholder(tf.int32, [None, NUM_CLASS], name="y_input")
x_input = tf.placeholder(
tf.int32, [FLAGS.batch_size, FLAGS.max_len], name="x_input")
y_input = tf.placeholder(
tf.int32, [FLAGS.batch_size, NUM_CLASS], name="y_input")
# Generate some dummy sequnce.
last_layer = inference(x_input)
objective = loss(last_layer, y_input)
opt = tf.train.AdamOptimizer(0.001)
grads = opt.compute_gradients(objective)
apply_gradient_op = opt.apply_gradients(
grads, global_step=global_step)
init = tf.initialize_all_variables()
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
run_forward = True
run_forward_backward = True
if FLAGS.forward_only and FLAGS.forward_backward_only:
raise ValueError("Cannot specify --forward_only and "
"--forward_backward_only at the same time.")
if FLAGS.forward_only:
run_forward_backward = False
elif FLAGS.forward_backward_only:
run_forward = False
if run_forward:
time_tensorflow_run(sess, last_layer, x_input, y_input,
"Forward")
if run_forward_backward:
with tf.control_dependencies([apply_gradient_op]):
train_op = tf.no_op(name='train')
time_tensorflow_run(sess, [train_op, objective], x_input,
y_input, "Forward-backward")
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/rnn/rnn_multi_gpu.py 0 → 100755
浏览文件 @ 021b3a44
#!/usr/bin/env python
from six.moves import xrange # pylint: disable=redefined-builtin
import re
import math
import time
import numpy as np
from datetime import datetime
import reader
import tensorflow as tf
from tensorflow.python.ops import rnn
FLAGS = tf.app.flags.FLAGS
tf.app.flags.DEFINE_integer('batch_size', 64, """Batch size.""")
tf.app.flags.DEFINE_integer('num_batches', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('num_layers', 1, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('max_len', 100, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('hidden_size', 128, """Number of batches to run.""")
tf.app.flags.DEFINE_integer('emb_size', 64, """Number of batches to run.""")
tf.app.flags.DEFINE_boolean('log_device_placement', False,
"""Whether to log device placement.""")
tf.app.flags.DEFINE_integer('num_gpus', 4, """How many GPUs to use.""")
VOCAB_SIZE = 30000
NUM_CLASS = 2
NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN = 50000
NUM_EPOCHS_PER_DECAY = 50
INITIAL_LEARNING_RATE = 0.1
LEARNING_RATE_DECAY_FACTOR = 0.1
TOWER_NAME = 'tower'
train_dataset = reader.create_datasets("imdb.pkl", VOCAB_SIZE)
def get_incoming_shape(incoming):
""" Returns the incoming data shape """
if isinstance(incoming, tf.Tensor):
return incoming.get_shape().as_list()
elif type(incoming) in [np.array, list, tuple]:
return np.shape(incoming)
else:
raise Exception("Invalid incoming layer.")
# Note input * W is done in LSTMCell,
# which is different from PaddlePaddle
def single_lstm(name,
incoming,
n_units,
use_peepholes=True,
return_seq=False,
return_state=False):
with tf.name_scope(name) as scope:
cell = tf.nn.rnn_cell.LSTMCell(n_units, use_peepholes=use_peepholes)
output, _cell_state = rnn.rnn(cell, incoming, dtype=tf.float32)
out = output if return_seq else output[-1]
return (out, _cell_state) if return_state else out
def lstm(name,
incoming,
n_units,
use_peepholes=True,
return_seq=False,
return_state=False,
num_layers=1):
with tf.name_scope(name) as scope:
lstm_cell = tf.nn.rnn_cell.LSTMCell(
n_units, use_peepholes=use_peepholes)
cell = tf.nn.rnn_cell.MultiRNNCell([lstm_cell] * num_layers)
initial_state = cell.zero_state(FLAGS.batch_size, dtype=tf.float32)
if not isinstance(incoming, list):
# if the input is embeding, the Tensor shape : [None, time_step, emb_size]
incoming = [
tf.squeeze(input_, [1])
for input_ in tf.split(1, FLAGS.max_len, incoming)
]
outputs, state = tf.nn.rnn(cell,
incoming,
initial_state=initial_state,
dtype=tf.float32)
out = outputs if return_seq else outputs[-1]
return (out, _cell_state) if return_state else out
def embedding(name, incoming, vocab_size, emb_size):
with tf.name_scope(name) as scope:
#with tf.device("/cpu:0"):
embedding = tf.get_variable(
name + '_emb', [vocab_size, emb_size], dtype=tf.float32)
out = tf.nn.embedding_lookup(embedding, incoming)
return out
def fc(name, inpOp, nIn, nOut, act=True):
with tf.name_scope(name) as scope:
kernel = tf.get_variable(
name + '_w', [nIn, nOut],
initializer=tf.truncated_normal_initializer(
stddev=0.01, dtype=tf.float32),
dtype=tf.float32)
biases = tf.get_variable(
name + '_b', [nOut],
initializer=tf.constant_initializer(
value=0.0, dtype=tf.float32),
dtype=tf.float32,
trainable=True)
net = tf.nn.relu_layer(inpOp, kernel, biases, name=name) if act else \
tf.matmul(inpOp, kernel) + biases
return net
def inference(seq):
net = embedding('emb', seq, VOCAB_SIZE, FLAGS.emb_size)
print "emb:", get_incoming_shape(net)
net = lstm('lstm', net, FLAGS.hidden_size, num_layers=FLAGS.num_layers)
print "lstm:", get_incoming_shape(net)
net = fc('fc1', net, FLAGS.hidden_size, 2)
return net
def loss(logits, labels):
# one label index for one sample
#labels = tf.cast(labels, tf.int64)
# cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
# logits, labels, name='cross_entropy_per_example')
labels = tf.cast(labels, tf.float32)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(
logits, labels, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
return tf.add_n(tf.get_collection('losses'), name='total_loss')
def tower_loss(scope):
"""Calculate the total loss on a single tower running the model.
Args:
scope: unique prefix string identifying the tower, e.g. 'tower_0'
Returns:
Tensor of shape [] containing the total loss for a batch of data
"""
data, label = train_dataset.next_batch(FLAGS.batch_size)
# Build a Graph that computes the logits predictions from the
# inference model.
last_layer = inference(data)
# Build the portion of the Graph calculating the losses. Note that we will
# assemble the total_loss using a custom function below.
#_ = loss(last_layer, label)
_ = loss(last_layer, label)
# Assemble all of the losses for the current tower only.
losses = tf.get_collection('losses', scope)
# Calculate the total loss for the current tower.
total_loss = tf.add_n(losses, name='total_loss')
# Compute the moving average of all individual losses and the total loss.
loss_averages = tf.train.ExponentialMovingAverage(0.9, name='avg')
loss_averages_op = loss_averages.apply(losses + [total_loss])
# Attach a scalar summary to all individual losses and the total loss; do the
# same for the averaged version of the losses.
for l in losses + [total_loss]:
# Remove 'tower_[0-9]/' from the name in case this is a multi-GPU training
# session. This helps the clarity of presentation on tensorboard.
loss_name = re.sub('%s_[0-9]*/' % TOWER_NAME, '', l.op.name)
# Name each loss as '(raw)' and name the moving average version of the loss
# as the original loss name.
tf.scalar_summary(loss_name + ' (raw)', l)
#tf.scalar_summary(loss_name, loss_averages.average(l))
with tf.control_dependencies([loss_averages_op]):
total_loss = tf.identity(total_loss)
return total_loss
def average_gradients(tower_grads):
"""Calculate the average gradient for each shared variable across all towers.
Note that this function provides a synchronization point across all towers.
Args:
tower_grads: List of lists of (gradient, variable) tuples. The outer list
is over individual gradients. The inner list is over the gradient
calculation for each tower.
Returns:
List of pairs of (gradient, variable) where the gradient has been averaged
across all towers.
"""
average_grads = []
for grad_and_vars in zip(*tower_grads):
# Note that each grad_and_vars looks like the following:
# ((grad0_gpu0, var0_gpu0), ... , (grad0_gpuN, var0_gpuN))
grads = []
for g, _ in grad_and_vars:
# Add 0 dimension to the gradients to represent the tower.
expanded_g = tf.expand_dims(g, 0)
# Append on a 'tower' dimension which we will average over below.
grads.append(expanded_g)
# Average over the 'tower' dimension.
grad = tf.concat(0, grads)
grad = tf.reduce_mean(grad, 0)
# Keep in mind that the Variables are redundant because they are shared
# across towers. So .. we will just return the first tower's pointer to
# the Variable.
v = grad_and_vars[0][1]
grad_and_var = (grad, v)
average_grads.append(grad_and_var)
return average_grads
def time_tensorflow_run(session, target):
num_steps_burn_in = 80
total_duration = 0.0
total_duration_squared = 0.0
for i in xrange(FLAGS.num_batches + num_steps_burn_in):
start_time = time.time()
_ = session.run(target, feed_dict={x_input: data, y_input: label})
_, loss_value = session.run(target)
duration = time.time() - start_time
if i > num_steps_burn_in:
if not i % 10:
num_examples_per_step = FLAGS.batch_size * FLAGS.num_gpus
examples_per_sec = num_examples_per_step / duration
# sec_per_batch = duration / FLAGS.num_gpus
sec_per_batch = duration
format_str = (
'%s: step %d, loss= %.2f (%.1f examples/sec; %.3f '
'sec/batch batch_size= %d)')
print(format_str %
(datetime.now(), i - num_steps_burn_in, loss_value,
duration, sec_per_batch, num_examples_per_step))
total_duration += duration
total_duration_squared += duration * duration
mn = total_duration / FLAGS.num_batches
vr = total_duration_squared / FLAGS.num_batches - mn * mn
sd = math.sqrt(vr)
print('%s: FwdBwd across %d steps, %.3f +/- %.3f sec / batch' %
(datetime.now(), FLAGS.num_batches, mn, sd))
def run_benchmark():
with tf.Graph().as_default(), tf.device('/cpu:0'):
# Create a variable to count the number of train() calls. This equals the
# number of batches processed * FLAGS.num_gpus.
global_step = tf.get_variable(
'global_step', [],
initializer=tf.constant_initializer(0),
trainable=False)
# Calculate the learning rate schedule.
num_batches_per_epoch = (NUM_EXAMPLES_PER_EPOCH_FOR_TRAIN /
FLAGS.batch_size)
decay_steps = int(num_batches_per_epoch * NUM_EPOCHS_PER_DECAY)
# Create an optimizer that performs gradient descent.
opt = tf.train.AdamOptimizer(0.001)
#train_dataset = reader.create_datasets("imdb.pkl", VOCAB_SIZE)
# Calculate the gradients for each model tower.
tower_grads = []
for i in xrange(FLAGS.num_gpus):
with tf.device('/gpu:%d' % i):
with tf.name_scope('%s_%d' % (TOWER_NAME, i)) as scope:
# Calculate the loss for one tower of the model. This function
# constructs the entire model but shares the variables across
# all towers.
loss = tower_loss(scope)
# Reuse variables for the next tower.
tf.get_variable_scope().reuse_variables()
# Retain the summaries from the final tower.
# summaries = tf.get_collection(tf.GraphKeys.SUMMARIES, scope)
# Calculate the gradients for the batch of data on this tower.
grads = opt.compute_gradients(loss)
# Keep track of the gradients across all towers.
tower_grads.append(grads)
# We must calculate the mean of each gradient. Note that this is the
# synchronization point across all towers.
grads = average_gradients(tower_grads)
# Apply the gradients to adjust the shared variables.
apply_gradient_op = opt.apply_gradients(grads, global_step=global_step)
# Group all updates to into a single train op.
train_op = tf.group(apply_gradient_op)
# Build an initialization operation.
init = tf.initialize_all_variables()
# Start running operations on the Graph. allow_soft_placement must be set to
# True to build towers on GPU, as some of the ops do not have GPU
# implementations.
sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True,
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
time_tensorflow_run(sess, [train_op, loss])
def main(_):
run_benchmark()
if __name__ == '__main__':
tf.app.run()
benchmark/tensorflow/rnn/run.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function test() {
lstm_num=$1
batch_size=$2
hid_size=$3
prefix=$4
python rnn.py --num_layers=${lstm_num} --batch_size=$batch_size \
--hidden_size=${hid_size} \
--forward_backward_only=1 \
> logs/1gpu-${lstm_num}lstm-batch${batch_size}-hid${hid_size}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
#--lstm_num--batch_size--hidden_size--#
test 2 64 256
test 2 64 512
test 2 64 1280
test 2 128 256
test 2 128 512
test 2 128 1280
test 2 256 256
test 2 256 512
test 2 256 1280
benchmark/tensorflow/rnn/run_multi.sh 0 → 100755
浏览文件 @ 021b3a44
set -e
function test() {
num_gpu=$1
lstm_num=$2
hid_size=$3
batch_per_gpu=`expr ${batch_size} / ${num_gpu}`
batch_size=$4
python rnn_multi_gpu.py --num_layers=${lstm_num} --batch_size=$batch_per_gpu \
--num_gpus=${num_gpu} \
--hidden_size=${hid_size} \
--forward_backward_only=1 \
> logs/${num_gpu}gpu-${lstm_num}lstm-hid${hid_size}-batch${batch_size}.log 2>&1
}
if [ ! -d "logs" ]; then
mkdir logs
fi
#--num_gpus--lstm_num--hiddne_size--batch_size--#
test 4 2 256 128
test 4 2 256 256
test 4 2 256 512
test 4 2 512 128
test 4 2 512 256
test 4 2 512 512
cmake/version.cmake 0 → 100644
浏览文件 @ 021b3a44
# Get the latest git tag.
set(PADDLE_VERSION $ENV{PADDLE_VERSION})
set(tmp_version "HEAD")
while ("${PADDLE_VERSION}" STREQUAL "")
execute_process(
COMMAND ${GIT_EXECUTABLE} describe --tags --abbrev=0 ${tmp_version}
WORKING_DIRECTORY ${PROJ_ROOT}
OUTPUT_VARIABLE GIT_TAG_NAME
RESULT_VARIABLE GIT_RESULT
ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
if (NOT ${GIT_RESULT})
# Check the tag is a correct version
if (${GIT_TAG_NAME} MATCHES "v[0-9]+\\.[0-9]+\\.[0-9]+(\\.(a|b|rc)\\.[0-9]+)?")
string(REPLACE "v" "" PADDLE_VERSION ${GIT_TAG_NAME})
else() # otherwise, get the previous git tag name.
set(tmp_version "${GIT_TAG_NAME}~1")
endif()
else()
set(PADDLE_VERSION "0.0.0")
message(WARNING "Cannot add paddle version from git tag")
endif()
endwhile()
message(STATUS "Paddle version is ${PADDLE_VERSION}")
demo/gan/.gitignore 0 → 100644
浏览文件 @ 021b3a44
output/
uniform_params/
cifar_params/
mnist_params/
*.png
.pydevproject
.project
*.log
*.pyc
data/mnist_data/
data/cifar-10-batches-py/
demo/gan/README.md 0 → 100644
浏览文件 @ 021b3a44
# 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).
$python gan_trainer.py -d cifar --use_gpu 1
The generated images will be stored in ./cifar_samples/
The corresponding models will be stored in ./cifar_params/
\ No newline at end of file
demo/gan/data/download_cifar.sh 0 → 100755
浏览文件 @ 021b3a44
# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
tar zxf cifar-10-python.tar.gz
rm cifar-10-python.tar.gz
demo/gan/data/get_mnist_data.sh 0 → 100644
浏览文件 @ 021b3a44
#!/usr/bin/env sh
# This script downloads the mnist data and unzips it.
set -e
DIR="$( 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
demo/gan/gan_conf.py 0 → 100644
浏览文件 @ 021b3a44
# Copyright (c) 2016 Baidu, 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 *
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))
demo/gan/gan_conf_image.py 0 → 100644
浏览文件 @ 021b3a44
# Copyright (c) 2016 Baidu, 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 *
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 = "_conv"
else:
nameApx = "_convt"
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))
demo/gan/gan_trainer.py 0 → 100644
浏览文件 @ 021b3a44
# Copyright (c) 2016 Baidu, 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.
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()
demo/quick_start/data/README.md 0 → 100644
浏览文件 @ 021b3a44
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`.
demo/quick_start/data/get_data.sh
浏览文件 @ 021b3a44
...@@ -17,14 +17,11 @@ set -e ...@@ -17,14 +17,11 @@ set -e
DIR="$( cd "$(dirname "$0")" ; pwd -P )" DIR="$( cd "$(dirname "$0")" ; pwd -P )"
cd $DIR cd $DIR
echo "Downloading Amazon Electronics reviews data..." # Download the preprocessed data
# http://jmcauley.ucsd.edu/data/amazon/ wget http://paddlepaddle.bj.bcebos.com/demo/quick_start_preprocessed_data/preprocessed_data.tar.gz
wget http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz
echo "Downloading mosesdecoder..." # Extract package
#https://github.com/moses-smt/mosesdecoder tar zxvf preprocessed_data.tar.gz
wget https://github.com/moses-smt/mosesdecoder/archive/master.zip
unzip master.zip # Remove compressed package
rm master.zip rm preprocessed_data.tar.gz
echo "Done."
demo/quick_start/data/pred.txt 已删除 100644 → 0
浏览文件 @ 4f1bf30d
the device is cute , but that &apos;s just about all that &apos;s good. the specs are what you &apos;d expect : it &apos;s a wifi mic , with some noise filter options. the app has the option to upload your baby &apos;s name and photo , which is a cutesy touch. but the app is otherwise unstable and useless unless you upgrade for $ 60 / year.set up involves downloading the app , turning on the mic , switching your phone to the wifi network of the mic , telling the app your wifi settings , switching your wifi back to your home router. the app is then directly connected to your mic.the app is adware ! the main screen says &quot; cry notifications on / off : upgrade to evoz premium and receive a text message of email when your baby is crying &quot; .but the adware points out an important limitation , this monitor is only intended to be used from your home network. if you want to access it remotely , get a webcam. this app would make a lot more sense of the premium features were included with the hardware .
don &apos;t be fooled by my one star rating. if there was a zero , i would have selected it. this product was a waste of my money.it has never worked like the company said it supposed to. i only have one device , an iphone 4gs. after charging the the iphone mid way , the i.sound portable power max 16,000 mah is completely drained. the led light no longer lit up. when plugging the isound portable power max into a wall outlet to charge , it would charge for about 20-30 minutes and then all four battery led indicator lit up showing a full charge. i would leave it on to charge for the full 8 hours or more but each time with the same result upon using. don &apos;t buy this thing. put your money to good use elsewhere .
demo/quick_start/preprocess.sh → demo/quick_start/data/proc_from_raw_data/get_data.sh
浏览文件 @ 021b3a44
...@@ -16,10 +16,26 @@ ...@@ -16,10 +16,26 @@
# 1. size of pos : neg = 1:1. # 1. size of pos : neg = 1:1.
# 2. size of testing set = min(25k, len(all_data) * 0.1), others is traning set. # 2. size of testing set = min(25k, len(all_data) * 0.1), others is traning set.
# 3. distinct train set and test set. # 3. distinct train set and test set.
# 4. build dict
set -e set -e
DIR="$( cd "$(dirname "$0")" ; pwd -P )"
cd $DIR
# Download data
echo "Downloading Amazon Electronics reviews data..."
# http://jmcauley.ucsd.edu/data/amazon/
wget http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz
echo "Downloading mosesdecoder..."
# https://github.com/moses-smt/mosesdecoder
wget https://github.com/moses-smt/mosesdecoder/archive/master.zip
unzip master.zip
rm master.zip
##################
# Preprocess data
echo "Preprocess data..."
export LC_ALL=C export LC_ALL=C
UNAME_STR=`uname` UNAME_STR=`uname`
...@@ -29,11 +45,11 @@ else ...@@ -29,11 +45,11 @@ else
SHUF_PROG='gshuf' SHUF_PROG='gshuf'
fi fi
mkdir -p data/tmp mkdir -p tmp
python preprocess.py -i data/reviews_Electronics_5.json.gz python preprocess.py -i reviews_Electronics_5.json.gz
# uniq and shuffle # uniq and shuffle
cd data/tmp cd tmp
echo 'uniq and shuffle...' echo 'Uniq and shuffle...'
cat pos_*|sort|uniq|${SHUF_PROG}> pos.shuffed cat pos_*|sort|uniq|${SHUF_PROG}> pos.shuffed
cat neg_*|sort|uniq|${SHUF_PROG}> neg.shuffed cat neg_*|sort|uniq|${SHUF_PROG}> neg.shuffed
...@@ -53,11 +69,11 @@ cat train.pos train.neg | ${SHUF_PROG} >../train.txt ...@@ -53,11 +69,11 @@ cat train.pos train.neg | ${SHUF_PROG} >../train.txt
cat test.pos test.neg | ${SHUF_PROG} >../test.txt cat test.pos test.neg | ${SHUF_PROG} >../test.txt
cd - cd -
echo 'data/train.txt' > data/train.list echo 'train.txt' > train.list
echo 'data/test.txt' > data/test.list echo 'test.txt' > test.list
# use 30k dict # use 30k dict
rm -rf data/tmp rm -rf tmp
mv data/dict.txt data/dict_all.txt mv dict.txt dict_all.txt
cat data/dict_all.txt | head -n 30001 > data/dict.txt cat dict_all.txt | head -n 30001 > dict.txt
echo 'preprocess finished' echo 'Done.'
demo/quick_start/preprocess.py → demo/quick_start/data/proc_from_raw_data/preprocess.py
浏览文件 @ 021b3a44
...@@ -14,7 +14,7 @@ ...@@ -14,7 +14,7 @@
# See the License for the specific language governing permissions and # See the License for the specific language governing permissions and
# limitations under the License. # limitations under the License.
""" """
1. (remove HTML before or not)tokensizing 1. Tokenize the words and punctuation
2. pos sample : rating score 5; neg sample: rating score 1-2. 2. pos sample : rating score 5; neg sample: rating score 1-2.
Usage: Usage:
...@@ -76,7 +76,11 @@ def tokenize(sentences): ...@@ -76,7 +76,11 @@ def tokenize(sentences):
sentences : a list of input sentences. sentences : a list of input sentences.
return: a list of processed text. return: a list of processed text.
""" """
dir = './data/mosesdecoder-master/scripts/tokenizer/tokenizer.perl' 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', '-'] tokenizer_cmd = [dir, '-l', 'en', '-q', '-']
assert isinstance(sentences, list) assert isinstance(sentences, list)
text = "\n".join(sentences) text = "\n".join(sentences)
...@@ -104,7 +108,7 @@ def tokenize_batch(id): ...@@ -104,7 +108,7 @@ def tokenize_batch(id):
num_batch, instance, pre_fix = parse_queue.get() num_batch, instance, pre_fix = parse_queue.get()
if num_batch == -1: ### parse_queue finished if num_batch == -1: ### parse_queue finished
tokenize_queue.put((-1, None, None)) tokenize_queue.put((-1, None, None))
sys.stderr.write("tokenize theread %s finish\n" % (id)) sys.stderr.write("Thread %s finish\n" % (id))
break break
tokenize_instance = tokenize(instance) tokenize_instance = tokenize(instance)
tokenize_queue.put((num_batch, tokenize_instance, pre_fix)) tokenize_queue.put((num_batch, tokenize_instance, pre_fix))
......
demo/semantic_role_labeling/data/get_data.sh
浏览文件 @ 021b3a44
...@@ -14,10 +14,10 @@ ...@@ -14,10 +14,10 @@
# limitations under the License. # limitations under the License.
set -e set -e
wget http://www.cs.upc.edu/~srlconll/conll05st-tests.tar.gz wget http://www.cs.upc.edu/~srlconll/conll05st-tests.tar.gz
wget https://www.googledrive.com/host/0B7Q8d52jqeI9ejh6Q1RpMTFQT1k/semantic_role_labeling/verbDict.txt --no-check-certificate wget http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/verbDict.txt
wget https://www.googledrive.com/host/0B7Q8d52jqeI9ejh6Q1RpMTFQT1k/semantic_role_labeling/targetDict.txt --no-check-certificate wget http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/targetDict.txt
wget https://www.googledrive.com/host/0B7Q8d52jqeI9ejh6Q1RpMTFQT1k/semantic_role_labeling/wordDict.txt --no-check-certificate wget http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/wordDict.txt
wget https://www.googledrive.com/host/0B7Q8d52jqeI9ejh6Q1RpMTFQT1k/semantic_role_labeling/emb --no-check-certificate wget http://paddlepaddle.bj.bcebos.com/demo/srl_dict_and_embedding/emb
tar -xzvf conll05st-tests.tar.gz tar -xzvf conll05st-tests.tar.gz
rm conll05st-tests.tar.gz rm conll05st-tests.tar.gz
cp ./conll05st-release/test.wsj/words/test.wsj.words.gz . cp ./conll05st-release/test.wsj/words/test.wsj.words.gz .
......
doc/CMakeLists.txt
浏览文件 @ 021b3a44
...@@ -15,25 +15,11 @@ set(SPHINX_CACHE_DIR "${CMAKE_CURRENT_BINARY_DIR}/_doctrees") ...@@ -15,25 +15,11 @@ set(SPHINX_CACHE_DIR "${CMAKE_CURRENT_BINARY_DIR}/_doctrees")
# HTML output directory # HTML output directory
set(SPHINX_HTML_DIR "${CMAKE_CURRENT_BINARY_DIR}/html") set(SPHINX_HTML_DIR "${CMAKE_CURRENT_BINARY_DIR}/html")
set(PADDLE_DOXYGEN_OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/doxygen_xml")
configure_file( configure_file(
"${CMAKE_CURRENT_SOURCE_DIR}/conf.py.in" "${CMAKE_CURRENT_SOURCE_DIR}/conf.py.in"
"${BINARY_BUILD_DIR}/conf.py" "${BINARY_BUILD_DIR}/conf.py"
@ONLY) @ONLY)
configure_file(
"${CMAKE_CURRENT_SOURCE_DIR}/Doxyfile.in"
"${CMAKE_CURRENT_BINARY_DIR}/Doxyfile"
@ONLY
)
add_custom_target(paddle_doxygen_docs ALL
${DOXYGEN_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/Doxyfile
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
)
sphinx_add_target(paddle_docs sphinx_add_target(paddle_docs
html html
${BINARY_BUILD_DIR} ${BINARY_BUILD_DIR}
...@@ -42,5 +28,4 @@ sphinx_add_target(paddle_docs ...@@ -42,5 +28,4 @@ sphinx_add_target(paddle_docs
${SPHINX_HTML_DIR}) ${SPHINX_HTML_DIR})
add_dependencies(paddle_docs add_dependencies(paddle_docs
gen_proto_py gen_proto_py)
paddle_doxygen_docs)
doc/Doxyfile.in 已删除 100644 → 0
浏览文件 @ 4f1bf30d
# Doxyfile 1.8.10
# This file describes the settings to be used by the documentation system
# doxygen (www.doxygen.org) for a project.
#
# All text after a double hash (##) is considered a comment and is placed in
# front of the TAG it is preceding.
#
# All text after a single hash (#) is considered a comment and will be ignored.
# The format is:
# TAG = value [value, ...]
# For lists, items can also be appended using:
# TAG += value [value, ...]
# Values that contain spaces should be placed between quotes (\" \").
#---------------------------------------------------------------------------
# Project related configuration options
#---------------------------------------------------------------------------
# This tag specifies the encoding used for all characters in the config file
# that follow. The default is UTF-8 which is also the encoding used for all text
# before the first occurrence of this tag. Doxygen uses libiconv (or the iconv
# built into libc) for the transcoding. See http://www.gnu.org/software/libiconv
# for the list of possible encodings.
# The default value is: UTF-8.
DOXYFILE_ENCODING = UTF-8
# The PROJECT_NAME tag is a single word (or a sequence of words surrounded by
# double-quotes, unless you are using Doxywizard) that should identify the
# project for which the documentation is generated. This name is used in the
# title of most generated pages and in a few other places.
# The default value is: My Project.
PROJECT_NAME = "paddle"
# The PROJECT_NUMBER tag can be used to enter a project or revision number. This
# could be handy for archiving the generated documentation or if some version
# control system is used.
PROJECT_NUMBER = 1.0.0
# Using the PROJECT_BRIEF tag one can provide an optional one line description
# for a project that appears at the top of each page and should give viewer a
# quick idea about the purpose of the project. Keep the description short.
PROJECT_BRIEF =
# With the PROJECT_LOGO tag one can specify a logo or an icon that is included
# in the documentation. The maximum height of the logo should not exceed 55
# pixels and the maximum width should not exceed 200 pixels. Doxygen will copy
# the logo to the output directory.
PROJECT_LOGO =
# The OUTPUT_DIRECTORY tag is used to specify the (relative or absolute) path
# into which the generated documentation will be written. If a relative path is
# entered, it will be relative to the location where doxygen was started. If
# left blank the current directory will be used.
OUTPUT_DIRECTORY = @PADDLE_DOXYGEN_OUTPUT@
# If the CREATE_SUBDIRS tag is set to YES then doxygen will create 4096 sub-
# directories (in 2 levels) under the output directory of each output format and
# will distribute the generated files over these directories. Enabling this
# option can be useful when feeding doxygen a huge amount of source files, where
# putting all generated files in the same directory would otherwise causes
# performance problems for the file system.
# The default value is: NO.
CREATE_SUBDIRS = NO
# If the ALLOW_UNICODE_NAMES tag is set to YES, doxygen will allow non-ASCII
# characters to appear in the names of generated files. If set to NO, non-ASCII
# characters will be escaped, for example _xE3_x81_x84 will be used for Unicode
# U+3044.
# The default value is: NO.
ALLOW_UNICODE_NAMES = NO
# The OUTPUT_LANGUAGE tag is used to specify the language in which all
# documentation generated by doxygen is written. Doxygen will use this
# information to generate all constant output in the proper language.
# Possible values are: Afrikaans, Arabic, Armenian, Brazilian, Catalan, Chinese,
# Chinese-Traditional, Croatian, Czech, Danish, Dutch, English (United States),
# Esperanto, Farsi (Persian), Finnish, French, German, Greek, Hungarian,
# Indonesian, Italian, Japanese, Japanese-en (Japanese with English messages),
# Korean, Korean-en (Korean with English messages), Latvian, Lithuanian,
# Macedonian, Norwegian, Persian (Farsi), Polish, Portuguese, Romanian, Russian,
# Serbian, Serbian-Cyrillic, Slovak, Slovene, Spanish, Swedish, Turkish,
# Ukrainian and Vietnamese.
# The default value is: English.
OUTPUT_LANGUAGE = English
# If the BRIEF_MEMBER_DESC tag is set to YES, doxygen will include brief member
# descriptions after the members that are listed in the file and class
# documentation (similar to Javadoc). Set to NO to disable this.
# The default value is: YES.
BRIEF_MEMBER_DESC = YES
# If the REPEAT_BRIEF tag is set to YES, doxygen will prepend the brief
# description of a member or function before the detailed description
#
# Note: If both HIDE_UNDOC_MEMBERS and BRIEF_MEMBER_DESC are set to NO, the
# brief descriptions will be completely suppressed.
# The default value is: YES.
REPEAT_BRIEF = YES
# This tag implements a quasi-intelligent brief description abbreviator that is
# used to form the text in various listings. Each string in this list, if found
# as the leading text of the brief description, will be stripped from the text
# and the result, after processing the whole list, is used as the annotated
# text. Otherwise, the brief description is used as-is. If left blank, the
# following values are used ($name is automatically replaced with the name of
# the entity):The $name class, The $name widget, The $name file, is, provides,
# specifies, contains, represents, a, an and the.
ABBREVIATE_BRIEF =
# If the ALWAYS_DETAILED_SEC and REPEAT_BRIEF tags are both set to YES then
# doxygen will generate a detailed section even if there is only a brief
# description.
# The default value is: NO.
ALWAYS_DETAILED_SEC = NO
# If the INLINE_INHERITED_MEMB tag is set to YES, doxygen will show all
# inherited members of a class in the documentation of that class as if those
# members were ordinary class members. Constructors, destructors and assignment
# operators of the base classes will not be shown.
# The default value is: NO.
INLINE_INHERITED_MEMB = NO
# If the FULL_PATH_NAMES tag is set to YES, doxygen will prepend the full path
# before files name in the file list and in the header files. If set to NO the
# shortest path that makes the file name unique will be used
# The default value is: YES.
FULL_PATH_NAMES = YES
# The STRIP_FROM_PATH tag can be used to strip a user-defined part of the path.
# Stripping is only done if one of the specified strings matches the left-hand
# part of the path. The tag can be used to show relative paths in the file list.
# If left blank the directory from which doxygen is run is used as the path to
# strip.
#
# Note that you can specify absolute paths here, but also relative paths, which
# will be relative from the directory where doxygen is started.
# This tag requires that the tag FULL_PATH_NAMES is set to YES.
STRIP_FROM_PATH =
# The STRIP_FROM_INC_PATH tag can be used to strip a user-defined part of the
# path mentioned in the documentation of a class, which tells the reader which
# header file to include in order to use a class. If left blank only the name of
# the header file containing the class definition is used. Otherwise one should
# specify the list of include paths that are normally passed to the compiler
# using the -I flag.
STRIP_FROM_INC_PATH =
# If the SHORT_NAMES tag is set to YES, doxygen will generate much shorter (but
# less readable) file names. This can be useful is your file systems doesn't
# support long names like on DOS, Mac, or CD-ROM.
# The default value is: NO.
SHORT_NAMES = NO
# If the JAVADOC_AUTOBRIEF tag is set to YES then doxygen will interpret the
# first line (until the first dot) of a Javadoc-style comment as the brief
# description. If set to NO, the Javadoc-style will behave just like regular Qt-
# style comments (thus requiring an explicit @brief command for a brief
# description.)
# The default value is: NO.
JAVADOC_AUTOBRIEF = NO
# If the QT_AUTOBRIEF tag is set to YES then doxygen will interpret the first
# line (until the first dot) of a Qt-style comment as the brief description. If
# set to NO, the Qt-style will behave just like regular Qt-style comments (thus
# requiring an explicit \brief command for a brief description.)
# The default value is: NO.
QT_AUTOBRIEF = NO
# The MULTILINE_CPP_IS_BRIEF tag can be set to YES to make doxygen treat a
# multi-line C++ special comment block (i.e. a block of //! or /// comments) as
# a brief description. This used to be the default behavior. The new default is
# to treat a multi-line C++ comment block as a detailed description. Set this
# tag to YES if you prefer the old behavior instead.
#
# Note that setting this tag to YES also means that rational rose comments are
# not recognized any more.
# The default value is: NO.
MULTILINE_CPP_IS_BRIEF = NO
# If the INHERIT_DOCS tag is set to YES then an undocumented member inherits the
# documentation from any documented member that it re-implements.
# The default value is: YES.
INHERIT_DOCS = YES
# If the SEPARATE_MEMBER_PAGES tag is set to YES then doxygen will produce a new
# page for each member. If set to NO, the documentation of a member will be part
# of the file/class/namespace that contains it.
# The default value is: NO.
SEPARATE_MEMBER_PAGES = NO
# The TAB_SIZE tag can be used to set the number of spaces in a tab. Doxygen
# uses this value to replace tabs by spaces in code fragments.
# Minimum value: 1, maximum value: 16, default value: 4.
TAB_SIZE = 2
# This tag can be used to specify a number of aliases that act as commands in
# the documentation. An alias has the form:
# name=value
# For example adding
# "sideeffect=@par Side Effects:\n"
# will allow you to put the command \sideeffect (or @sideeffect) in the
# documentation, which will result in a user-defined paragraph with heading
# "Side Effects:". You can put \n's in the value part of an alias to insert
# newlines.
ALIASES =
# This tag can be used to specify a number of word-keyword mappings (TCL only).
# A mapping has the form "name=value". For example adding "class=itcl::class"
# will allow you to use the command class in the itcl::class meaning.
TCL_SUBST =
# Set the OPTIMIZE_OUTPUT_FOR_C tag to YES if your project consists of C sources
# only. Doxygen will then generate output that is more tailored for C. For
# instance, some of the names that are used will be different. The list of all
# members will be omitted, etc.
# The default value is: NO.
OPTIMIZE_OUTPUT_FOR_C = NO
# Set the OPTIMIZE_OUTPUT_JAVA tag to YES if your project consists of Java or
# Python sources only. Doxygen will then generate output that is more tailored
# for that language. For instance, namespaces will be presented as packages,
# qualified scopes will look different, etc.
# The default value is: NO.
OPTIMIZE_OUTPUT_JAVA = NO
# Set the OPTIMIZE_FOR_FORTRAN tag to YES if your project consists of Fortran
# sources. Doxygen will then generate output that is tailored for Fortran.
# The default value is: NO.
OPTIMIZE_FOR_FORTRAN = NO
# Set the OPTIMIZE_OUTPUT_VHDL tag to YES if your project consists of VHDL
# sources. Doxygen will then generate output that is tailored for VHDL.
# The default value is: NO.
OPTIMIZE_OUTPUT_VHDL = NO
# Doxygen selects the parser to use depending on the extension of the files it
# parses. With this tag you can assign which parser to use for a given
# extension. Doxygen has a built-in mapping, but you can override or extend it
# using this tag. The format is ext=language, where ext is a file extension, and
# language is one of the parsers supported by doxygen: IDL, Java, Javascript,
# C#, C, C++, D, PHP, Objective-C, Python, Fortran (fixed format Fortran:
# FortranFixed, free formatted Fortran: FortranFree, unknown formatted Fortran:
# Fortran. In the later case the parser tries to guess whether the code is fixed
# or free formatted code, this is the default for Fortran type files), VHDL. For
# instance to make doxygen treat .inc files as Fortran files (default is PHP),
# and .f files as C (default is Fortran), use: inc=Fortran f=C.
#
# Note: For files without extension you can use no_extension as a placeholder.
#
# Note that for custom extensions you also need to set FILE_PATTERNS otherwise
# the files are not read by doxygen.
EXTENSION_MAPPING =
# If the MARKDOWN_SUPPORT tag is enabled then doxygen pre-processes all comments
# according to the Markdown format, which allows for more readable
# documentation. See http://daringfireball.net/projects/markdown/ for details.
# The output of markdown processing is further processed by doxygen, so you can
# mix doxygen, HTML, and XML commands with Markdown formatting. Disable only in
# case of backward compatibilities issues.
# The default value is: YES.
MARKDOWN_SUPPORT = YES
# When enabled doxygen tries to link words that correspond to documented
# classes, or namespaces to their corresponding documentation. Such a link can
# be prevented in individual cases by putting a % sign in front of the word or
# globally by setting AUTOLINK_SUPPORT to NO.
# The default value is: YES.
AUTOLINK_SUPPORT = YES
# If you use STL classes (i.e. std::string, std::vector, etc.) but do not want
# to include (a tag file for) the STL sources as input, then you should set this
# tag to YES in order to let doxygen match functions declarations and
# definitions whose arguments contain STL classes (e.g. func(std::string);
# versus func(std::string) {}). This also make the inheritance and collaboration
# diagrams that involve STL classes more complete and accurate.
# The default value is: NO.
BUILTIN_STL_SUPPORT = YES
# If you use Microsoft's C++/CLI language, you should set this option to YES to
# enable parsing support.
# The default value is: NO.
CPP_CLI_SUPPORT = NO
# Set the SIP_SUPPORT tag to YES if your project consists of sip (see:
# http://www.riverbankcomputing.co.uk/software/sip/intro) sources only. Doxygen
# will parse them like normal C++ but will assume all classes use public instead
# of private inheritance when no explicit protection keyword is present.
# The default value is: NO.
SIP_SUPPORT = NO
# For Microsoft's IDL there are propget and propput attributes to indicate
# getter and setter methods for a property. Setting this option to YES will make
# doxygen to replace the get and set methods by a property in the documentation.
# This will only work if the methods are indeed getting or setting a simple
# type. If this is not the case, or you want to show the methods anyway, you
# should set this option to NO.
# The default value is: YES.
IDL_PROPERTY_SUPPORT = YES
# If member grouping is used in the documentation and the DISTRIBUTE_GROUP_DOC
# tag is set to YES then doxygen will reuse the documentation of the first
# member in the group (if any) for the other members of the group. By default
# all members of a group must be documented explicitly.
# The default value is: NO.
DISTRIBUTE_GROUP_DOC = NO
# If one adds a struct or class to a group and this option is enabled, then also
# any nested class or struct is added to the same group. By default this option
# is disabled and one has to add nested compounds explicitly via \ingroup.
# The default value is: NO.
GROUP_NESTED_COMPOUNDS = NO
# Set the SUBGROUPING tag to YES to allow class member groups of the same type
# (for instance a group of public functions) to be put as a subgroup of that
# type (e.g. under the Public Functions section). Set it to NO to prevent
# subgrouping. Alternatively, this can be done per class using the
# \nosubgrouping command.
# The default value is: YES.
SUBGROUPING = YES
# When the INLINE_GROUPED_CLASSES tag is set to YES, classes, structs and unions
# are shown inside the group in which they are included (e.g. using \ingroup)
# instead of on a separate page (for HTML and Man pages) or section (for LaTeX
# and RTF).
#
# Note that this feature does not work in combination with
# SEPARATE_MEMBER_PAGES.
# The default value is: NO.
INLINE_GROUPED_CLASSES = NO
# When the INLINE_SIMPLE_STRUCTS tag is set to YES, structs, classes, and unions
# with only public data fields or simple typedef fields will be shown inline in
# the documentation of the scope in which they are defined (i.e. file,
# namespace, or group documentation), provided this scope is documented. If set
# to NO, structs, classes, and unions are shown on a separate page (for HTML and
# Man pages) or section (for LaTeX and RTF).
# The default value is: NO.
INLINE_SIMPLE_STRUCTS = NO
# When TYPEDEF_HIDES_STRUCT tag is enabled, a typedef of a struct, union, or
# enum is documented as struct, union, or enum with the name of the typedef. So
# typedef struct TypeS {} TypeT, will appear in the documentation as a struct
# with name TypeT. When disabled the typedef will appear as a member of a file,
# namespace, or class. And the struct will be named TypeS. This can typically be
# useful for C code in case the coding convention dictates that all compound
# types are typedef'ed and only the typedef is referenced, never the tag name.
# The default value is: NO.
TYPEDEF_HIDES_STRUCT = NO
# The size of the symbol lookup cache can be set using LOOKUP_CACHE_SIZE. This
# cache is used to resolve symbols given their name and scope. Since this can be
# an expensive process and often the same symbol appears multiple times in the
# code, doxygen keeps a cache of pre-resolved symbols. If the cache is too small
# doxygen will become slower. If the cache is too large, memory is wasted. The
# cache size is given by this formula: 2^(16+LOOKUP_CACHE_SIZE). The valid range
# is 0..9, the default is 0, corresponding to a cache size of 2^16=65536
# symbols. At the end of a run doxygen will report the cache usage and suggest
# the optimal cache size from a speed point of view.
# Minimum value: 0, maximum value: 9, default value: 0.
LOOKUP_CACHE_SIZE = 0
#---------------------------------------------------------------------------
# Build related configuration options
#---------------------------------------------------------------------------
# If the EXTRACT_ALL tag is set to YES, doxygen will assume all entities in
# documentation are documented, even if no documentation was available. Private
# class members and static file members will be hidden unless the
# EXTRACT_PRIVATE respectively EXTRACT_STATIC tags are set to YES.
# Note: This will also disable the warnings about undocumented members that are
# normally produced when WARNINGS is set to YES.
# The default value is: NO.
EXTRACT_ALL = NO
# If the EXTRACT_PRIVATE tag is set to YES, all private members of a class will
# be included in the documentation.
# The default value is: NO.
EXTRACT_PRIVATE = NO
# If the EXTRACT_PACKAGE tag is set to YES, all members with package or internal
# scope will be included in the documentation.
# The default value is: NO.
EXTRACT_PACKAGE = NO
# If the EXTRACT_STATIC tag is set to YES, all static members of a file will be
# included in the documentation.
# The default value is: NO.
EXTRACT_STATIC = NO
# If the EXTRACT_LOCAL_CLASSES tag is set to YES, classes (and structs) defined
# locally in source files will be included in the documentation. If set to NO,
# only classes defined in header files are included. Does not have any effect
# for Java sources.
# The default value is: YES.
EXTRACT_LOCAL_CLASSES = YES
# This flag is only useful for Objective-C code. If set to YES, local methods,
# which are defined in the implementation section but not in the interface are
# included in the documentation. If set to NO, only methods in the interface are
# included.
# The default value is: NO.
EXTRACT_LOCAL_METHODS = NO
# If this flag is set to YES, the members of anonymous namespaces will be
# extracted and appear in the documentation as a namespace called
# 'anonymous_namespace{file}', where file will be replaced with the base name of
# the file that contains the anonymous namespace. By default anonymous namespace
# are hidden.
# The default value is: NO.
EXTRACT_ANON_NSPACES = NO
# If the HIDE_UNDOC_MEMBERS tag is set to YES, doxygen will hide all
# undocumented members inside documented classes or files. If set to NO these
# members will be included in the various overviews, but no documentation
# section is generated. This option has no effect if EXTRACT_ALL is enabled.
# The default value is: NO.
HIDE_UNDOC_MEMBERS = NO
# If the HIDE_UNDOC_CLASSES tag is set to YES, doxygen will hide all
# undocumented classes that are normally visible in the class hierarchy. If set
# to NO, these classes will be included in the various overviews. This option
# has no effect if EXTRACT_ALL is enabled.
# The default value is: NO.
HIDE_UNDOC_CLASSES = NO
# If the HIDE_FRIEND_COMPOUNDS tag is set to YES, doxygen will hide all friend
# (class|struct|union) declarations. If set to NO, these declarations will be
# included in the documentation.
# The default value is: NO.
HIDE_FRIEND_COMPOUNDS = NO
# If the HIDE_IN_BODY_DOCS tag is set to YES, doxygen will hide any
# documentation blocks found inside the body of a function. If set to NO, these
# blocks will be appended to the function's detailed documentation block.
# The default value is: NO.
HIDE_IN_BODY_DOCS = NO
# The INTERNAL_DOCS tag determines if documentation that is typed after a
# \internal command is included. If the tag is set to NO then the documentation
# will be excluded. Set it to YES to include the internal documentation.
# The default value is: NO.
INTERNAL_DOCS = NO
# If the CASE_SENSE_NAMES tag is set to NO then doxygen will only generate file
# names in lower-case letters. If set to YES, upper-case letters are also
# allowed. This is useful if you have classes or files whose names only differ
# in case and if your file system supports case sensitive file names. Windows
# and Mac users are advised to set this option to NO.
# The default value is: system dependent.
CASE_SENSE_NAMES = YES
# If the HIDE_SCOPE_NAMES tag is set to NO then doxygen will show members with
# their full class and namespace scopes in the documentation. If set to YES, the
# scope will be hidden.
# The default value is: NO.
HIDE_SCOPE_NAMES = NO
# If the HIDE_COMPOUND_REFERENCE tag is set to NO (default) then doxygen will
# append additional text to a page's title, such as Class Reference. If set to
# YES the compound reference will be hidden.
# The default value is: NO.
HIDE_COMPOUND_REFERENCE= NO
# If the SHOW_INCLUDE_FILES tag is set to YES then doxygen will put a list of
# the files that are included by a file in the documentation of that file.
# The default value is: YES.
SHOW_INCLUDE_FILES = NO
# If the SHOW_GROUPED_MEMB_INC tag is set to YES then Doxygen will add for each
# grouped member an include statement to the documentation, telling the reader
# which file to include in order to use the member.
# The default value is: NO.
SHOW_GROUPED_MEMB_INC = NO
# If the FORCE_LOCAL_INCLUDES tag is set to YES then doxygen will list include
# files with double quotes in the documentation rather than with sharp brackets.
# The default value is: NO.
FORCE_LOCAL_INCLUDES = NO
# If the INLINE_INFO tag is set to YES then a tag [inline] is inserted in the
# documentation for inline members.
# The default value is: YES.
INLINE_INFO = YES
# If the SORT_MEMBER_DOCS tag is set to YES then doxygen will sort the
# (detailed) documentation of file and class members alphabetically by member
# name. If set to NO, the members will appear in declaration order.
# The default value is: YES.
SORT_MEMBER_DOCS = YES
# If the SORT_BRIEF_DOCS tag is set to YES then doxygen will sort the brief
# descriptions of file, namespace and class members alphabetically by member
# name. If set to NO, the members will appear in declaration order. Note that
# this will also influence the order of the classes in the class list.
# The default value is: NO.
SORT_BRIEF_DOCS = NO
# If the SORT_MEMBERS_CTORS_1ST tag is set to YES then doxygen will sort the
# (brief and detailed) documentation of class members so that constructors and
# destructors are listed first. If set to NO the constructors will appear in the
# respective orders defined by SORT_BRIEF_DOCS and SORT_MEMBER_DOCS.
# Note: If SORT_BRIEF_DOCS is set to NO this option is ignored for sorting brief
# member documentation.
# Note: If SORT_MEMBER_DOCS is set to NO this option is ignored for sorting
# detailed member documentation.
# The default value is: NO.
SORT_MEMBERS_CTORS_1ST = NO
# If the SORT_GROUP_NAMES tag is set to YES then doxygen will sort the hierarchy
# of group names into alphabetical order. If set to NO the group names will
# appear in their defined order.
# The default value is: NO.
SORT_GROUP_NAMES = NO
# If the SORT_BY_SCOPE_NAME tag is set to YES, the class list will be sorted by
# fully-qualified names, including namespaces. If set to NO, the class list will
# be sorted only by class name, not including the namespace part.
# Note: This option is not very useful if HIDE_SCOPE_NAMES is set to YES.
# Note: This option applies only to the class list, not to the alphabetical
# list.
# The default value is: NO.
SORT_BY_SCOPE_NAME = NO
# If the STRICT_PROTO_MATCHING option is enabled and doxygen fails to do proper
# type resolution of all parameters of a function it will reject a match between
# the prototype and the implementation of a member function even if there is
# only one candidate or it is obvious which candidate to choose by doing a
# simple string match. By disabling STRICT_PROTO_MATCHING doxygen will still
# accept a match between prototype and implementation in such cases.
# The default value is: NO.
STRICT_PROTO_MATCHING = NO
# The GENERATE_TODOLIST tag can be used to enable (YES) or disable (NO) the todo
# list. This list is created by putting \todo commands in the documentation.
# The default value is: YES.
GENERATE_TODOLIST = YES
# The GENERATE_TESTLIST tag can be used to enable (YES) or disable (NO) the test
# list. This list is created by putting \test commands in the documentation.
# The default value is: YES.
GENERATE_TESTLIST = YES
# The GENERATE_BUGLIST tag can be used to enable (YES) or disable (NO) the bug
# list. This list is created by putting \bug commands in the documentation.
# The default value is: YES.
GENERATE_BUGLIST = YES
# The GENERATE_DEPRECATEDLIST tag can be used to enable (YES) or disable (NO)
# the deprecated list. This list is created by putting \deprecated commands in
# the documentation.
# The default value is: YES.
GENERATE_DEPRECATEDLIST= YES
# The ENABLED_SECTIONS tag can be used to enable conditional documentation
# sections, marked by \if <section_label> ... \endif and \cond <section_label>
# ... \endcond blocks.
ENABLED_SECTIONS =
# The MAX_INITIALIZER_LINES tag determines the maximum number of lines that the
# initial value of a variable or macro / define can have for it to appear in the
# documentation. If the initializer consists of more lines than specified here
# it will be hidden. Use a value of 0 to hide initializers completely. The
# appearance of the value of individual variables and macros / defines can be
# controlled using \showinitializer or \hideinitializer command in the
# documentation regardless of this setting.
# Minimum value: 0, maximum value: 10000, default value: 30.
MAX_INITIALIZER_LINES = 30
# Set the SHOW_USED_FILES tag to NO to disable the list of files generated at
# the bottom of the documentation of classes and structs. If set to YES, the
# list will mention the files that were used to generate the documentation.
# The default value is: YES.
SHOW_USED_FILES = YES
# Set the SHOW_FILES tag to NO to disable the generation of the Files page. This
# will remove the Files entry from the Quick Index and from the Folder Tree View
# (if specified).
# The default value is: YES.
SHOW_FILES = YES
# Set the SHOW_NAMESPACES tag to NO to disable the generation of the Namespaces
# page. This will remove the Namespaces entry from the Quick Index and from the
# Folder Tree View (if specified).
# The default value is: YES.
SHOW_NAMESPACES = YES
# The FILE_VERSION_FILTER tag can be used to specify a program or script that
# doxygen should invoke to get the current version for each file (typically from
# the version control system). Doxygen will invoke the program by executing (via
# popen()) the command command input-file, where command is the value of the
# FILE_VERSION_FILTER tag, and input-file is the name of an input file provided
# by doxygen. Whatever the program writes to standard output is used as the file
# version. For an example see the documentation.
FILE_VERSION_FILTER =
# The LAYOUT_FILE tag can be used to specify a layout file which will be parsed
# by doxygen. The layout file controls the global structure of the generated
# output files in an output format independent way. To create the layout file
# that represents doxygen's defaults, run doxygen with the -l option. You can
# optionally specify a file name after the option, if omitted DoxygenLayout.xml
# will be used as the name of the layout file.
#
# Note that if you run doxygen from a directory containing a file called
# DoxygenLayout.xml, doxygen will parse it automatically even if the LAYOUT_FILE
# tag is left empty.
LAYOUT_FILE =
# The CITE_BIB_FILES tag can be used to specify one or more bib files containing
# the reference definitions. This must be a list of .bib files. The .bib
# extension is automatically appended if omitted. This requires the bibtex tool
# to be installed. See also http://en.wikipedia.org/wiki/BibTeX for more info.
# For LaTeX the style of the bibliography can be controlled using
# LATEX_BIB_STYLE. To use this feature you need bibtex and perl available in the
# search path. See also \cite for info how to create references.
CITE_BIB_FILES =
#---------------------------------------------------------------------------
# Configuration options related to warning and progress messages
#---------------------------------------------------------------------------
# The QUIET tag can be used to turn on/off the messages that are generated to
# standard output by doxygen. If QUIET is set to YES this implies that the
# messages are off.
# The default value is: NO.
QUIET = NO
# The WARNINGS tag can be used to turn on/off the warning messages that are
# generated to standard error (stderr) by doxygen. If WARNINGS is set to YES
# this implies that the warnings are on.
#
# Tip: Turn warnings on while writing the documentation.
# The default value is: YES.
WARNINGS = YES
# If the WARN_IF_UNDOCUMENTED tag is set to YES then doxygen will generate
# warnings for undocumented members. If EXTRACT_ALL is set to YES then this flag
# will automatically be disabled.
# The default value is: YES.
WARN_IF_UNDOCUMENTED = NO
# If the WARN_IF_DOC_ERROR tag is set to YES, doxygen will generate warnings for
# potential errors in the documentation, such as not documenting some parameters
# in a documented function, or documenting parameters that don't exist or using
# markup commands wrongly.
# The default value is: YES.
WARN_IF_DOC_ERROR = YES
# This WARN_NO_PARAMDOC option can be enabled to get warnings for functions that
# are documented, but have no documentation for their parameters or return
# value. If set to NO, doxygen will only warn about wrong or incomplete
# parameter documentation, but not about the absence of documentation.
# The default value is: NO.
WARN_NO_PARAMDOC = NO
# The WARN_FORMAT tag determines the format of the warning messages that doxygen
# can produce. The string should contain the $file, $line, and $text tags, which
# will be replaced by the file and line number from which the warning originated
# and the warning text. Optionally the format may contain $version, which will
# be replaced by the version of the file (if it could be obtained via
# FILE_VERSION_FILTER)
# The default value is: $file:$line: $text.
WARN_FORMAT = "$file:$line: $text"
# The WARN_LOGFILE tag can be used to specify a file to which warning and error
# messages should be written. If left blank the output is written to standard
# error (stderr).
WARN_LOGFILE =
#---------------------------------------------------------------------------
# Configuration options related to the input files
#---------------------------------------------------------------------------
# The INPUT tag is used to specify the files and/or directories that contain
# documented source files. You may enter file names like myfile.cpp or
# directories like /usr/src/myproject. Separate the files or directories with
# spaces. See also FILE_PATTERNS and EXTENSION_MAPPING
# Note: If this tag is empty the current directory is searched.
INPUT = @PROJ_ROOT@/paddle
# This tag can be used to specify the character encoding of the source files
# that doxygen parses. Internally doxygen uses the UTF-8 encoding. Doxygen uses
# libiconv (or the iconv built into libc) for the transcoding. See the libiconv
# documentation (see: http://www.gnu.org/software/libiconv) for the list of
# possible encodings.
# The default value is: UTF-8.
INPUT_ENCODING = UTF-8
# If the value of the INPUT tag contains directories, you can use the
# FILE_PATTERNS tag to specify one or more wildcard patterns (like *.cpp and
# *.h) to filter out the source-files in the directories.
#
# Note that for custom extensions or not directly supported extensions you also
# need to set EXTENSION_MAPPING for the extension otherwise the files are not
# read by doxygen.
#
# If left blank the following patterns are tested:*.c, *.cc, *.cxx, *.cpp,
# *.c++, *.java, *.ii, *.ixx, *.ipp, *.i++, *.inl, *.idl, *.ddl, *.odl, *.h,
# *.hh, *.hxx, *.hpp, *.h++, *.cs, *.d, *.php, *.php4, *.php5, *.phtml, *.inc,
# *.m, *.markdown, *.md, *.mm, *.dox, *.py, *.f90, *.f, *.for, *.tcl, *.vhd,
# *.vhdl, *.ucf, *.qsf, *.as and *.js.
FILE_PATTERNS = *.c *.cc *.cpp *.cu *.h *.hpp *.cuh *.ph
# The RECURSIVE tag can be used to specify whether or not subdirectories should
# be searched for input files as well.
# The default value is: NO.
RECURSIVE = YES
# The EXCLUDE tag can be used to specify files and/or directories that should be
# excluded from the INPUT source files. This way you can easily exclude a
# subdirectory from a directory tree whose root is specified with the INPUT tag.
#
# Note that relative paths are relative to the directory from which doxygen is
# run.
EXCLUDE =
# The EXCLUDE_SYMLINKS tag can be used to select whether or not files or
# directories that are symbolic links (a Unix file system feature) are excluded
# from the input.
# The default value is: NO.
EXCLUDE_SYMLINKS = NO
# If the value of the INPUT tag contains directories, you can use the
# EXCLUDE_PATTERNS tag to specify one or more wildcard patterns to exclude
# certain files from those directories.
#
# Note that the wildcards are matched against the file with absolute path, so to
# exclude all test directories for example use the pattern */test/*
EXCLUDE_PATTERNS = */x86_64-scm-linux-gnu/* */internals/* */mkl/* */test/* */tests/* */platform/*
# The EXCLUDE_SYMBOLS tag can be used to specify one or more symbol names
# (namespaces, classes, functions, etc.) that should be excluded from the
# output. The symbol name can be a fully qualified name, a word, or if the
# wildcard * is used, a substring. Examples: ANamespace, AClass,
# AClass::ANamespace, ANamespace::*Test
#
# Note that the wildcards are matched against the file with absolute path, so to
# exclude all test directories use the pattern */test/*
EXCLUDE_SYMBOLS =
# The EXAMPLE_PATH tag can be used to specify one or more files or directories
# that contain example code fragments that are included (see the \include
# command).
EXAMPLE_PATH =
# If the value of the EXAMPLE_PATH tag contains directories, you can use the
# EXAMPLE_PATTERNS tag to specify one or more wildcard pattern (like *.cpp and
# *.h) to filter out the source-files in the directories. If left blank all
# files are included.
EXAMPLE_PATTERNS =
# If the EXAMPLE_RECURSIVE tag is set to YES then subdirectories will be
# searched for input files to be used with the \include or \dontinclude commands
# irrespective of the value of the RECURSIVE tag.
# The default value is: NO.
EXAMPLE_RECURSIVE = NO
# The IMAGE_PATH tag can be used to specify one or more files or directories
# that contain images that are to be included in the documentation (see the
# \image command).
IMAGE_PATH =
# The INPUT_FILTER tag can be used to specify a program that doxygen should
# invoke to filter for each input file. Doxygen will invoke the filter program
# by executing (via popen()) the command:
#
# <filter> <input-file>
#
# where <filter> is the value of the INPUT_FILTER tag, and <input-file> is the
# name of an input file. Doxygen will then use the output that the filter
# program writes to standard output. If FILTER_PATTERNS is specified, this tag
# will be ignored.
#
# Note that the filter must not add or remove lines; it is applied before the
# code is scanned, but not when the output code is generated. If lines are added
# or removed, the anchors will not be placed correctly.
INPUT_FILTER =
# The FILTER_PATTERNS tag can be used to specify filters on a per file pattern
# basis. Doxygen will compare the file name with each pattern and apply the
# filter if there is a match. The filters are a list of the form: pattern=filter
# (like *.cpp=my_cpp_filter). See INPUT_FILTER for further information on how
# filters are used. If the FILTER_PATTERNS tag is empty or if none of the
# patterns match the file name, INPUT_FILTER is applied.
FILTER_PATTERNS =
# If the FILTER_SOURCE_FILES tag is set to YES, the input filter (if set using
# INPUT_FILTER) will also be used to filter the input files that are used for
# producing the source files to browse (i.e. when SOURCE_BROWSER is set to YES).
# The default value is: NO.
FILTER_SOURCE_FILES = NO
# The FILTER_SOURCE_PATTERNS tag can be used to specify source filters per file
# pattern. A pattern will override the setting for FILTER_PATTERN (if any) and
# it is also possible to disable source filtering for a specific pattern using
# *.ext= (so without naming a filter).
# This tag requires that the tag FILTER_SOURCE_FILES is set to YES.
FILTER_SOURCE_PATTERNS =
# If the USE_MDFILE_AS_MAINPAGE tag refers to the name of a markdown file that
# is part of the input, its contents will be placed on the main page
# (index.html). This can be useful if you have a project on for instance GitHub
# and want to reuse the introduction page also for the doxygen output.
USE_MDFILE_AS_MAINPAGE =
#---------------------------------------------------------------------------
# Configuration options related to source browsing
#---------------------------------------------------------------------------
# If the SOURCE_BROWSER tag is set to YES then a list of source files will be
# generated. Documented entities will be cross-referenced with these sources.
#
# Note: To get rid of all source code in the generated output, make sure that
# also VERBATIM_HEADERS is set to NO.
# The default value is: NO.
SOURCE_BROWSER = NO
# Setting the INLINE_SOURCES tag to YES will include the body of functions,
# classes and enums directly into the documentation.
# The default value is: NO.
INLINE_SOURCES = NO
# Setting the STRIP_CODE_COMMENTS tag to YES will instruct doxygen to hide any
# special comment blocks from generated source code fragments. Normal C, C++ and
# Fortran comments will always remain visible.
# The default value is: YES.
STRIP_CODE_COMMENTS = YES
# If the REFERENCED_BY_RELATION tag is set to YES then for each documented
# function all documented functions referencing it will be listed.
# The default value is: NO.
REFERENCED_BY_RELATION = NO
# If the REFERENCES_RELATION tag is set to YES then for each documented function
# all documented entities called/used by that function will be listed.
# The default value is: NO.
REFERENCES_RELATION = NO
# If the REFERENCES_LINK_SOURCE tag is set to YES and SOURCE_BROWSER tag is set
# to YES then the hyperlinks from functions in REFERENCES_RELATION and
# REFERENCED_BY_RELATION lists will link to the source code. Otherwise they will
# link to the documentation.
# The default value is: YES.
REFERENCES_LINK_SOURCE = YES
# If SOURCE_TOOLTIPS is enabled (the default) then hovering a hyperlink in the
# source code will show a tooltip with additional information such as prototype,
# brief description and links to the definition and documentation. Since this
# will make the HTML file larger and loading of large files a bit slower, you
# can opt to disable this feature.
# The default value is: YES.
# This tag requires that the tag SOURCE_BROWSER is set to YES.
SOURCE_TOOLTIPS = YES
# If the USE_HTAGS tag is set to YES then the references to source code will
# point to the HTML generated by the htags(1) tool instead of doxygen built-in
# source browser. The htags tool is part of GNU's global source tagging system
# (see http://www.gnu.org/software/global/global.html). You will need version
# 4.8.6 or higher.
#
# To use it do the following:
# - Install the latest version of global
# - Enable SOURCE_BROWSER and USE_HTAGS in the config file
# - Make sure the INPUT points to the root of the source tree
# - Run doxygen as normal
#
# Doxygen will invoke htags (and that will in turn invoke gtags), so these
# tools must be available from the command line (i.e. in the search path).
#
# The result: instead of the source browser generated by doxygen, the links to
# source code will now point to the output of htags.
# The default value is: NO.
# This tag requires that the tag SOURCE_BROWSER is set to YES.
USE_HTAGS = NO
# If the VERBATIM_HEADERS tag is set the YES then doxygen will generate a
# verbatim copy of the header file for each class for which an include is
# specified. Set to NO to disable this.
# See also: Section \class.
# The default value is: YES.
VERBATIM_HEADERS = YES
#---------------------------------------------------------------------------
# Configuration options related to the alphabetical class index
#---------------------------------------------------------------------------
# If the ALPHABETICAL_INDEX tag is set to YES, an alphabetical index of all
# compounds will be generated. Enable this if the project contains a lot of
# classes, structs, unions or interfaces.
# The default value is: YES.
ALPHABETICAL_INDEX = YES
# The COLS_IN_ALPHA_INDEX tag can be used to specify the number of columns in
# which the alphabetical index list will be split.
# Minimum value: 1, maximum value: 20, default value: 5.
# This tag requires that the tag ALPHABETICAL_INDEX is set to YES.
COLS_IN_ALPHA_INDEX = 5
# In case all classes in a project start with a common prefix, all classes will
# be put under the same header in the alphabetical index. The IGNORE_PREFIX tag
# can be used to specify a prefix (or a list of prefixes) that should be ignored
# while generating the index headers.
# This tag requires that the tag ALPHABETICAL_INDEX is set to YES.
IGNORE_PREFIX =
#---------------------------------------------------------------------------
# Configuration options related to the HTML output
#---------------------------------------------------------------------------
# If the GENERATE_HTML tag is set to YES, doxygen will generate HTML output
# The default value is: YES.
GENERATE_HTML = NO
# The HTML_OUTPUT tag is used to specify where the HTML docs will be put. If a
# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
# it.
# The default directory is: html.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_OUTPUT = html
# The HTML_FILE_EXTENSION tag can be used to specify the file extension for each
# generated HTML page (for example: .htm, .php, .asp).
# The default value is: .html.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_FILE_EXTENSION = .html
# The HTML_HEADER tag can be used to specify a user-defined HTML header file for
# each generated HTML page. If the tag is left blank doxygen will generate a
# standard header.
#
# To get valid HTML the header file that includes any scripts and style sheets
# that doxygen needs, which is dependent on the configuration options used (e.g.
# the setting GENERATE_TREEVIEW). It is highly recommended to start with a
# default header using
# doxygen -w html new_header.html new_footer.html new_stylesheet.css
# YourConfigFile
# and then modify the file new_header.html. See also section "Doxygen usage"
# for information on how to generate the default header that doxygen normally
# uses.
# Note: The header is subject to change so you typically have to regenerate the
# default header when upgrading to a newer version of doxygen. For a description
# of the possible markers and block names see the documentation.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_HEADER =
# The HTML_FOOTER tag can be used to specify a user-defined HTML footer for each
# generated HTML page. If the tag is left blank doxygen will generate a standard
# footer. See HTML_HEADER for more information on how to generate a default
# footer and what special commands can be used inside the footer. See also
# section "Doxygen usage" for information on how to generate the default footer
# that doxygen normally uses.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_FOOTER =
# The HTML_STYLESHEET tag can be used to specify a user-defined cascading style
# sheet that is used by each HTML page. It can be used to fine-tune the look of
# the HTML output. If left blank doxygen will generate a default style sheet.
# See also section "Doxygen usage" for information on how to generate the style
# sheet that doxygen normally uses.
# Note: It is recommended to use HTML_EXTRA_STYLESHEET instead of this tag, as
# it is more robust and this tag (HTML_STYLESHEET) will in the future become
# obsolete.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_STYLESHEET =
# The HTML_EXTRA_STYLESHEET tag can be used to specify additional user-defined
# cascading style sheets that are included after the standard style sheets
# created by doxygen. Using this option one can overrule certain style aspects.
# This is preferred over using HTML_STYLESHEET since it does not replace the
# standard style sheet and is therefore more robust against future updates.
# Doxygen will copy the style sheet files to the output directory.
# Note: The order of the extra style sheet files is of importance (e.g. the last
# style sheet in the list overrules the setting of the previous ones in the
# list). For an example see the documentation.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_EXTRA_STYLESHEET =
# The HTML_EXTRA_FILES tag can be used to specify one or more extra images or
# other source files which should be copied to the HTML output directory. Note
# that these files will be copied to the base HTML output directory. Use the
# $relpath^ marker in the HTML_HEADER and/or HTML_FOOTER files to load these
# files. In the HTML_STYLESHEET file, use the file name only. Also note that the
# files will be copied as-is; there are no commands or markers available.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_EXTRA_FILES =
# The HTML_COLORSTYLE_HUE tag controls the color of the HTML output. Doxygen
# will adjust the colors in the style sheet and background images according to
# this color. Hue is specified as an angle on a colorwheel, see
# http://en.wikipedia.org/wiki/Hue for more information. For instance the value
# 0 represents red, 60 is yellow, 120 is green, 180 is cyan, 240 is blue, 300
# purple, and 360 is red again.
# Minimum value: 0, maximum value: 359, default value: 220.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_COLORSTYLE_HUE = 220
# The HTML_COLORSTYLE_SAT tag controls the purity (or saturation) of the colors
# in the HTML output. For a value of 0 the output will use grayscales only. A
# value of 255 will produce the most vivid colors.
# Minimum value: 0, maximum value: 255, default value: 100.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_COLORSTYLE_SAT = 100
# The HTML_COLORSTYLE_GAMMA tag controls the gamma correction applied to the
# luminance component of the colors in the HTML output. Values below 100
# gradually make the output lighter, whereas values above 100 make the output
# darker. The value divided by 100 is the actual gamma applied, so 80 represents
# a gamma of 0.8, The value 220 represents a gamma of 2.2, and 100 does not
# change the gamma.
# Minimum value: 40, maximum value: 240, default value: 80.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_COLORSTYLE_GAMMA = 80
# If the HTML_TIMESTAMP tag is set to YES then the footer of each generated HTML
# page will contain the date and time when the page was generated. Setting this
# to YES can help to show when doxygen was last run and thus if the
# documentation is up to date.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_TIMESTAMP = NO
# If the HTML_DYNAMIC_SECTIONS tag is set to YES then the generated HTML
# documentation will contain sections that can be hidden and shown after the
# page has loaded.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_DYNAMIC_SECTIONS = NO
# With HTML_INDEX_NUM_ENTRIES one can control the preferred number of entries
# shown in the various tree structured indices initially; the user can expand
# and collapse entries dynamically later on. Doxygen will expand the tree to
# such a level that at most the specified number of entries are visible (unless
# a fully collapsed tree already exceeds this amount). So setting the number of
# entries 1 will produce a full collapsed tree by default. 0 is a special value
# representing an infinite number of entries and will result in a full expanded
# tree by default.
# Minimum value: 0, maximum value: 9999, default value: 100.
# This tag requires that the tag GENERATE_HTML is set to YES.
HTML_INDEX_NUM_ENTRIES = 100
# If the GENERATE_DOCSET tag is set to YES, additional index files will be
# generated that can be used as input for Apple's Xcode 3 integrated development
# environment (see: http://developer.apple.com/tools/xcode/), introduced with
# OSX 10.5 (Leopard). To create a documentation set, doxygen will generate a
# Makefile in the HTML output directory. Running make will produce the docset in
# that directory and running make install will install the docset in
# ~/Library/Developer/Shared/Documentation/DocSets so that Xcode will find it at
# startup. See http://developer.apple.com/tools/creatingdocsetswithdoxygen.html
# for more information.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
GENERATE_DOCSET = NO
# This tag determines the name of the docset feed. A documentation feed provides
# an umbrella under which multiple documentation sets from a single provider
# (such as a company or product suite) can be grouped.
# The default value is: Doxygen generated docs.
# This tag requires that the tag GENERATE_DOCSET is set to YES.
DOCSET_FEEDNAME = "Doxygen generated docs"
# This tag specifies a string that should uniquely identify the documentation
# set bundle. This should be a reverse domain-name style string, e.g.
# com.mycompany.MyDocSet. Doxygen will append .docset to the name.
# The default value is: org.doxygen.Project.
# This tag requires that the tag GENERATE_DOCSET is set to YES.
DOCSET_BUNDLE_ID = org.doxygen.Project
# The DOCSET_PUBLISHER_ID tag specifies a string that should uniquely identify
# the documentation publisher. This should be a reverse domain-name style
# string, e.g. com.mycompany.MyDocSet.documentation.
# The default value is: org.doxygen.Publisher.
# This tag requires that the tag GENERATE_DOCSET is set to YES.
DOCSET_PUBLISHER_ID = org.doxygen.Publisher
# The DOCSET_PUBLISHER_NAME tag identifies the documentation publisher.
# The default value is: Publisher.
# This tag requires that the tag GENERATE_DOCSET is set to YES.
DOCSET_PUBLISHER_NAME = Publisher
# If the GENERATE_HTMLHELP tag is set to YES then doxygen generates three
# additional HTML index files: index.hhp, index.hhc, and index.hhk. The
# index.hhp is a project file that can be read by Microsoft's HTML Help Workshop
# (see: http://www.microsoft.com/en-us/download/details.aspx?id=21138) on
# Windows.
#
# The HTML Help Workshop contains a compiler that can convert all HTML output
# generated by doxygen into a single compiled HTML file (.chm). Compiled HTML
# files are now used as the Windows 98 help format, and will replace the old
# Windows help format (.hlp) on all Windows platforms in the future. Compressed
# HTML files also contain an index, a table of contents, and you can search for
# words in the documentation. The HTML workshop also contains a viewer for
# compressed HTML files.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
GENERATE_HTMLHELP = NO
# The CHM_FILE tag can be used to specify the file name of the resulting .chm
# file. You can add a path in front of the file if the result should not be
# written to the html output directory.
# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
CHM_FILE =
# The HHC_LOCATION tag can be used to specify the location (absolute path
# including file name) of the HTML help compiler (hhc.exe). If non-empty,
# doxygen will try to run the HTML help compiler on the generated index.hhp.
# The file has to be specified with full path.
# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
HHC_LOCATION =
# The GENERATE_CHI flag controls if a separate .chi index file is generated
# (YES) or that it should be included in the master .chm file (NO).
# The default value is: NO.
# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
GENERATE_CHI = NO
# The CHM_INDEX_ENCODING is used to encode HtmlHelp index (hhk), content (hhc)
# and project file content.
# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
CHM_INDEX_ENCODING =
# The BINARY_TOC flag controls whether a binary table of contents is generated
# (YES) or a normal table of contents (NO) in the .chm file. Furthermore it
# enables the Previous and Next buttons.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
BINARY_TOC = NO
# The TOC_EXPAND flag can be set to YES to add extra items for group members to
# the table of contents of the HTML help documentation and to the tree view.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTMLHELP is set to YES.
TOC_EXPAND = NO
# If the GENERATE_QHP tag is set to YES and both QHP_NAMESPACE and
# QHP_VIRTUAL_FOLDER are set, an additional index file will be generated that
# can be used as input for Qt's qhelpgenerator to generate a Qt Compressed Help
# (.qch) of the generated HTML documentation.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
GENERATE_QHP = NO
# If the QHG_LOCATION tag is specified, the QCH_FILE tag can be used to specify
# the file name of the resulting .qch file. The path specified is relative to
# the HTML output folder.
# This tag requires that the tag GENERATE_QHP is set to YES.
QCH_FILE =
# The QHP_NAMESPACE tag specifies the namespace to use when generating Qt Help
# Project output. For more information please see Qt Help Project / Namespace
# (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#namespace).
# The default value is: org.doxygen.Project.
# This tag requires that the tag GENERATE_QHP is set to YES.
QHP_NAMESPACE = org.doxygen.Project
# The QHP_VIRTUAL_FOLDER tag specifies the namespace to use when generating Qt
# Help Project output. For more information please see Qt Help Project / Virtual
# Folders (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#virtual-
# folders).
# The default value is: doc.
# This tag requires that the tag GENERATE_QHP is set to YES.
QHP_VIRTUAL_FOLDER = doc
# If the QHP_CUST_FILTER_NAME tag is set, it specifies the name of a custom
# filter to add. For more information please see Qt Help Project / Custom
# Filters (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#custom-
# filters).
# This tag requires that the tag GENERATE_QHP is set to YES.
QHP_CUST_FILTER_NAME =
# The QHP_CUST_FILTER_ATTRS tag specifies the list of the attributes of the
# custom filter to add. For more information please see Qt Help Project / Custom
# Filters (see: http://qt-project.org/doc/qt-4.8/qthelpproject.html#custom-
# filters).
# This tag requires that the tag GENERATE_QHP is set to YES.
QHP_CUST_FILTER_ATTRS =
# The QHP_SECT_FILTER_ATTRS tag specifies the list of the attributes this
# project's filter section matches. Qt Help Project / Filter Attributes (see:
# http://qt-project.org/doc/qt-4.8/qthelpproject.html#filter-attributes).
# This tag requires that the tag GENERATE_QHP is set to YES.
QHP_SECT_FILTER_ATTRS =
# The QHG_LOCATION tag can be used to specify the location of Qt's
# qhelpgenerator. If non-empty doxygen will try to run qhelpgenerator on the
# generated .qhp file.
# This tag requires that the tag GENERATE_QHP is set to YES.
QHG_LOCATION =
# If the GENERATE_ECLIPSEHELP tag is set to YES, additional index files will be
# generated, together with the HTML files, they form an Eclipse help plugin. To
# install this plugin and make it available under the help contents menu in
# Eclipse, the contents of the directory containing the HTML and XML files needs
# to be copied into the plugins directory of eclipse. The name of the directory
# within the plugins directory should be the same as the ECLIPSE_DOC_ID value.
# After copying Eclipse needs to be restarted before the help appears.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
GENERATE_ECLIPSEHELP = NO
# A unique identifier for the Eclipse help plugin. When installing the plugin
# the directory name containing the HTML and XML files should also have this
# name. Each documentation set should have its own identifier.
# The default value is: org.doxygen.Project.
# This tag requires that the tag GENERATE_ECLIPSEHELP is set to YES.
ECLIPSE_DOC_ID = org.doxygen.Project
# If you want full control over the layout of the generated HTML pages it might
# be necessary to disable the index and replace it with your own. The
# DISABLE_INDEX tag can be used to turn on/off the condensed index (tabs) at top
# of each HTML page. A value of NO enables the index and the value YES disables
# it. Since the tabs in the index contain the same information as the navigation
# tree, you can set this option to YES if you also set GENERATE_TREEVIEW to YES.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
DISABLE_INDEX = NO
# The GENERATE_TREEVIEW tag is used to specify whether a tree-like index
# structure should be generated to display hierarchical information. If the tag
# value is set to YES, a side panel will be generated containing a tree-like
# index structure (just like the one that is generated for HTML Help). For this
# to work a browser that supports JavaScript, DHTML, CSS and frames is required
# (i.e. any modern browser). Windows users are probably better off using the
# HTML help feature. Via custom style sheets (see HTML_EXTRA_STYLESHEET) one can
# further fine-tune the look of the index. As an example, the default style
# sheet generated by doxygen has an example that shows how to put an image at
# the root of the tree instead of the PROJECT_NAME. Since the tree basically has
# the same information as the tab index, you could consider setting
# DISABLE_INDEX to YES when enabling this option.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
GENERATE_TREEVIEW = NO
# The ENUM_VALUES_PER_LINE tag can be used to set the number of enum values that
# doxygen will group on one line in the generated HTML documentation.
#
# Note that a value of 0 will completely suppress the enum values from appearing
# in the overview section.
# Minimum value: 0, maximum value: 20, default value: 4.
# This tag requires that the tag GENERATE_HTML is set to YES.
ENUM_VALUES_PER_LINE = 4
# If the treeview is enabled (see GENERATE_TREEVIEW) then this tag can be used
# to set the initial width (in pixels) of the frame in which the tree is shown.
# Minimum value: 0, maximum value: 1500, default value: 250.
# This tag requires that the tag GENERATE_HTML is set to YES.
TREEVIEW_WIDTH = 250
# If the EXT_LINKS_IN_WINDOW option is set to YES, doxygen will open links to
# external symbols imported via tag files in a separate window.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
EXT_LINKS_IN_WINDOW = NO
# Use this tag to change the font size of LaTeX formulas included as images in
# the HTML documentation. When you change the font size after a successful
# doxygen run you need to manually remove any form_*.png images from the HTML
# output directory to force them to be regenerated.
# Minimum value: 8, maximum value: 50, default value: 10.
# This tag requires that the tag GENERATE_HTML is set to YES.
FORMULA_FONTSIZE = 10
# Use the FORMULA_TRANPARENT tag to determine whether or not the images
# generated for formulas are transparent PNGs. Transparent PNGs are not
# supported properly for IE 6.0, but are supported on all modern browsers.
#
# Note that when changing this option you need to delete any form_*.png files in
# the HTML output directory before the changes have effect.
# The default value is: YES.
# This tag requires that the tag GENERATE_HTML is set to YES.
FORMULA_TRANSPARENT = YES
# Enable the USE_MATHJAX option to render LaTeX formulas using MathJax (see
# http://www.mathjax.org) which uses client side Javascript for the rendering
# instead of using pre-rendered bitmaps. Use this if you do not have LaTeX
# installed or if you want to formulas look prettier in the HTML output. When
# enabled you may also need to install MathJax separately and configure the path
# to it using the MATHJAX_RELPATH option.
# The default value is: NO.
# This tag requires that the tag GENERATE_HTML is set to YES.
USE_MATHJAX = NO
# When MathJax is enabled you can set the default output format to be used for
# the MathJax output. See the MathJax site (see:
# http://docs.mathjax.org/en/latest/output.html) for more details.
# Possible values are: HTML-CSS (which is slower, but has the best
# compatibility), NativeMML (i.e. MathML) and SVG.
# The default value is: HTML-CSS.
# This tag requires that the tag USE_MATHJAX is set to YES.
MATHJAX_FORMAT = HTML-CSS
# When MathJax is enabled you need to specify the location relative to the HTML
# output directory using the MATHJAX_RELPATH option. The destination directory
# should contain the MathJax.js script. For instance, if the mathjax directory
# is located at the same level as the HTML output directory, then
# MATHJAX_RELPATH should be ../mathjax. The default value points to the MathJax
# Content Delivery Network so you can quickly see the result without installing
# MathJax. However, it is strongly recommended to install a local copy of
# MathJax from http://www.mathjax.org before deployment.
# The default value is: http://cdn.mathjax.org/mathjax/latest.
# This tag requires that the tag USE_MATHJAX is set to YES.
MATHJAX_RELPATH = http://cdn.mathjax.org/mathjax/latest
# The MATHJAX_EXTENSIONS tag can be used to specify one or more MathJax
# extension names that should be enabled during MathJax rendering. For example
# MATHJAX_EXTENSIONS = TeX/AMSmath TeX/AMSsymbols
# This tag requires that the tag USE_MATHJAX is set to YES.
MATHJAX_EXTENSIONS =
# The MATHJAX_CODEFILE tag can be used to specify a file with javascript pieces
# of code that will be used on startup of the MathJax code. See the MathJax site
# (see: http://docs.mathjax.org/en/latest/output.html) for more details. For an
# example see the documentation.
# This tag requires that the tag USE_MATHJAX is set to YES.
MATHJAX_CODEFILE =
# When the SEARCHENGINE tag is enabled doxygen will generate a search box for
# the HTML output. The underlying search engine uses javascript and DHTML and
# should work on any modern browser. Note that when using HTML help
# (GENERATE_HTMLHELP), Qt help (GENERATE_QHP), or docsets (GENERATE_DOCSET)
# there is already a search function so this one should typically be disabled.
# For large projects the javascript based search engine can be slow, then
# enabling SERVER_BASED_SEARCH may provide a better solution. It is possible to
# search using the keyboard; to jump to the search box use <access key> + S
# (what the <access key> is depends on the OS and browser, but it is typically
# <CTRL>, <ALT>/<option>, or both). Inside the search box use the <cursor down
# key> to jump into the search results window, the results can be navigated
# using the <cursor keys>. Press <Enter> to select an item or <escape> to cancel
# the search. The filter options can be selected when the cursor is inside the
# search box by pressing <Shift>+<cursor down>. Also here use the <cursor keys>
# to select a filter and <Enter> or <escape> to activate or cancel the filter
# option.
# The default value is: YES.
# This tag requires that the tag GENERATE_HTML is set to YES.
SEARCHENGINE = YES
# When the SERVER_BASED_SEARCH tag is enabled the search engine will be
# implemented using a web server instead of a web client using Javascript. There
# are two flavors of web server based searching depending on the EXTERNAL_SEARCH
# setting. When disabled, doxygen will generate a PHP script for searching and
# an index file used by the script. When EXTERNAL_SEARCH is enabled the indexing
# and searching needs to be provided by external tools. See the section
# "External Indexing and Searching" for details.
# The default value is: NO.
# This tag requires that the tag SEARCHENGINE is set to YES.
SERVER_BASED_SEARCH = NO
# When EXTERNAL_SEARCH tag is enabled doxygen will no longer generate the PHP
# script for searching. Instead the search results are written to an XML file
# which needs to be processed by an external indexer. Doxygen will invoke an
# external search engine pointed to by the SEARCHENGINE_URL option to obtain the
# search results.
#
# Doxygen ships with an example indexer (doxyindexer) and search engine
# (doxysearch.cgi) which are based on the open source search engine library
# Xapian (see: http://xapian.org/).
#
# See the section "External Indexing and Searching" for details.
# The default value is: NO.
# This tag requires that the tag SEARCHENGINE is set to YES.
EXTERNAL_SEARCH = NO
# The SEARCHENGINE_URL should point to a search engine hosted by a web server
# which will return the search results when EXTERNAL_SEARCH is enabled.
#
# Doxygen ships with an example indexer (doxyindexer) and search engine
# (doxysearch.cgi) which are based on the open source search engine library
# Xapian (see: http://xapian.org/). See the section "External Indexing and
# Searching" for details.
# This tag requires that the tag SEARCHENGINE is set to YES.
SEARCHENGINE_URL =
# When SERVER_BASED_SEARCH and EXTERNAL_SEARCH are both enabled the unindexed
# search data is written to a file for indexing by an external tool. With the
# SEARCHDATA_FILE tag the name of this file can be specified.
# The default file is: searchdata.xml.
# This tag requires that the tag SEARCHENGINE is set to YES.
SEARCHDATA_FILE = searchdata.xml
# When SERVER_BASED_SEARCH and EXTERNAL_SEARCH are both enabled the
# EXTERNAL_SEARCH_ID tag can be used as an identifier for the project. This is
# useful in combination with EXTRA_SEARCH_MAPPINGS to search through multiple
# projects and redirect the results back to the right project.
# This tag requires that the tag SEARCHENGINE is set to YES.
EXTERNAL_SEARCH_ID =
# The EXTRA_SEARCH_MAPPINGS tag can be used to enable searching through doxygen
# projects other than the one defined by this configuration file, but that are
# all added to the same external search index. Each project needs to have a
# unique id set via EXTERNAL_SEARCH_ID. The search mapping then maps the id of
# to a relative location where the documentation can be found. The format is:
# EXTRA_SEARCH_MAPPINGS = tagname1=loc1 tagname2=loc2 ...
# This tag requires that the tag SEARCHENGINE is set to YES.
EXTRA_SEARCH_MAPPINGS =
#---------------------------------------------------------------------------
# Configuration options related to the LaTeX output
#---------------------------------------------------------------------------
# If the GENERATE_LATEX tag is set to YES, doxygen will generate LaTeX output.
# The default value is: YES.
GENERATE_LATEX = NO
# The LATEX_OUTPUT tag is used to specify where the LaTeX docs will be put. If a
# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
# it.
# The default directory is: latex.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_OUTPUT = latex
# The LATEX_CMD_NAME tag can be used to specify the LaTeX command name to be
# invoked.
#
# Note that when enabling USE_PDFLATEX this option is only used for generating
# bitmaps for formulas in the HTML output, but not in the Makefile that is
# written to the output directory.
# The default file is: latex.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_CMD_NAME = latex
# The MAKEINDEX_CMD_NAME tag can be used to specify the command name to generate
# index for LaTeX.
# The default file is: makeindex.
# This tag requires that the tag GENERATE_LATEX is set to YES.
MAKEINDEX_CMD_NAME = makeindex
# If the COMPACT_LATEX tag is set to YES, doxygen generates more compact LaTeX
# documents. This may be useful for small projects and may help to save some
# trees in general.
# The default value is: NO.
# This tag requires that the tag GENERATE_LATEX is set to YES.
COMPACT_LATEX = NO
# The PAPER_TYPE tag can be used to set the paper type that is used by the
# printer.
# Possible values are: a4 (210 x 297 mm), letter (8.5 x 11 inches), legal (8.5 x
# 14 inches) and executive (7.25 x 10.5 inches).
# The default value is: a4.
# This tag requires that the tag GENERATE_LATEX is set to YES.
PAPER_TYPE = a4
# The EXTRA_PACKAGES tag can be used to specify one or more LaTeX package names
# that should be included in the LaTeX output. The package can be specified just
# by its name or with the correct syntax as to be used with the LaTeX
# \usepackage command. To get the times font for instance you can specify :
# EXTRA_PACKAGES=times or EXTRA_PACKAGES={times}
# To use the option intlimits with the amsmath package you can specify:
# EXTRA_PACKAGES=[intlimits]{amsmath}
# If left blank no extra packages will be included.
# This tag requires that the tag GENERATE_LATEX is set to YES.
EXTRA_PACKAGES =
# The LATEX_HEADER tag can be used to specify a personal LaTeX header for the
# generated LaTeX document. The header should contain everything until the first
# chapter. If it is left blank doxygen will generate a standard header. See
# section "Doxygen usage" for information on how to let doxygen write the
# default header to a separate file.
#
# Note: Only use a user-defined header if you know what you are doing! The
# following commands have a special meaning inside the header: $title,
# $datetime, $date, $doxygenversion, $projectname, $projectnumber,
# $projectbrief, $projectlogo. Doxygen will replace $title with the empty
# string, for the replacement values of the other commands the user is referred
# to HTML_HEADER.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_HEADER =
# The LATEX_FOOTER tag can be used to specify a personal LaTeX footer for the
# generated LaTeX document. The footer should contain everything after the last
# chapter. If it is left blank doxygen will generate a standard footer. See
# LATEX_HEADER for more information on how to generate a default footer and what
# special commands can be used inside the footer.
#
# Note: Only use a user-defined footer if you know what you are doing!
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_FOOTER =
# The LATEX_EXTRA_STYLESHEET tag can be used to specify additional user-defined
# LaTeX style sheets that are included after the standard style sheets created
# by doxygen. Using this option one can overrule certain style aspects. Doxygen
# will copy the style sheet files to the output directory.
# Note: The order of the extra style sheet files is of importance (e.g. the last
# style sheet in the list overrules the setting of the previous ones in the
# list).
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_EXTRA_STYLESHEET =
# The LATEX_EXTRA_FILES tag can be used to specify one or more extra images or
# other source files which should be copied to the LATEX_OUTPUT output
# directory. Note that the files will be copied as-is; there are no commands or
# markers available.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_EXTRA_FILES =
# If the PDF_HYPERLINKS tag is set to YES, the LaTeX that is generated is
# prepared for conversion to PDF (using ps2pdf or pdflatex). The PDF file will
# contain links (just like the HTML output) instead of page references. This
# makes the output suitable for online browsing using a PDF viewer.
# The default value is: YES.
# This tag requires that the tag GENERATE_LATEX is set to YES.
PDF_HYPERLINKS = YES
# If the USE_PDFLATEX tag is set to YES, doxygen will use pdflatex to generate
# the PDF file directly from the LaTeX files. Set this option to YES, to get a
# higher quality PDF documentation.
# The default value is: YES.
# This tag requires that the tag GENERATE_LATEX is set to YES.
USE_PDFLATEX = YES
# If the LATEX_BATCHMODE tag is set to YES, doxygen will add the \batchmode
# command to the generated LaTeX files. This will instruct LaTeX to keep running
# if errors occur, instead of asking the user for help. This option is also used
# when generating formulas in HTML.
# The default value is: NO.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_BATCHMODE = NO
# If the LATEX_HIDE_INDICES tag is set to YES then doxygen will not include the
# index chapters (such as File Index, Compound Index, etc.) in the output.
# The default value is: NO.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_HIDE_INDICES = NO
# If the LATEX_SOURCE_CODE tag is set to YES then doxygen will include source
# code with syntax highlighting in the LaTeX output.
#
# Note that which sources are shown also depends on other settings such as
# SOURCE_BROWSER.
# The default value is: NO.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_SOURCE_CODE = NO
# The LATEX_BIB_STYLE tag can be used to specify the style to use for the
# bibliography, e.g. plainnat, or ieeetr. See
# http://en.wikipedia.org/wiki/BibTeX and \cite for more info.
# The default value is: plain.
# This tag requires that the tag GENERATE_LATEX is set to YES.
LATEX_BIB_STYLE = plain
#---------------------------------------------------------------------------
# Configuration options related to the RTF output
#---------------------------------------------------------------------------
# If the GENERATE_RTF tag is set to YES, doxygen will generate RTF output. The
# RTF output is optimized for Word 97 and may not look too pretty with other RTF
# readers/editors.
# The default value is: NO.
GENERATE_RTF = NO
# The RTF_OUTPUT tag is used to specify where the RTF docs will be put. If a
# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
# it.
# The default directory is: rtf.
# This tag requires that the tag GENERATE_RTF is set to YES.
RTF_OUTPUT = rtf
# If the COMPACT_RTF tag is set to YES, doxygen generates more compact RTF
# documents. This may be useful for small projects and may help to save some
# trees in general.
# The default value is: NO.
# This tag requires that the tag GENERATE_RTF is set to YES.
COMPACT_RTF = NO
# If the RTF_HYPERLINKS tag is set to YES, the RTF that is generated will
# contain hyperlink fields. The RTF file will contain links (just like the HTML
# output) instead of page references. This makes the output suitable for online
# browsing using Word or some other Word compatible readers that support those
# fields.
#
# Note: WordPad (write) and others do not support links.
# The default value is: NO.
# This tag requires that the tag GENERATE_RTF is set to YES.
RTF_HYPERLINKS = NO
# Load stylesheet definitions from file. Syntax is similar to doxygen's config
# file, i.e. a series of assignments. You only have to provide replacements,
# missing definitions are set to their default value.
#
# See also section "Doxygen usage" for information on how to generate the
# default style sheet that doxygen normally uses.
# This tag requires that the tag GENERATE_RTF is set to YES.
RTF_STYLESHEET_FILE =
# Set optional variables used in the generation of an RTF document. Syntax is
# similar to doxygen's config file. A template extensions file can be generated
# using doxygen -e rtf extensionFile.
# This tag requires that the tag GENERATE_RTF is set to YES.
RTF_EXTENSIONS_FILE =
# If the RTF_SOURCE_CODE tag is set to YES then doxygen will include source code
# with syntax highlighting in the RTF output.
#
# Note that which sources are shown also depends on other settings such as
# SOURCE_BROWSER.
# The default value is: NO.
# This tag requires that the tag GENERATE_RTF is set to YES.
RTF_SOURCE_CODE = NO
#---------------------------------------------------------------------------
# Configuration options related to the man page output
#---------------------------------------------------------------------------
# If the GENERATE_MAN tag is set to YES, doxygen will generate man pages for
# classes and files.
# The default value is: NO.
GENERATE_MAN = NO
# The MAN_OUTPUT tag is used to specify where the man pages will be put. If a
# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
# it. A directory man3 will be created inside the directory specified by
# MAN_OUTPUT.
# The default directory is: man.
# This tag requires that the tag GENERATE_MAN is set to YES.
MAN_OUTPUT = man
# The MAN_EXTENSION tag determines the extension that is added to the generated
# man pages. In case the manual section does not start with a number, the number
# 3 is prepended. The dot (.) at the beginning of the MAN_EXTENSION tag is
# optional.
# The default value is: .3.
# This tag requires that the tag GENERATE_MAN is set to YES.
MAN_EXTENSION = .3
# The MAN_SUBDIR tag determines the name of the directory created within
# MAN_OUTPUT in which the man pages are placed. If defaults to man followed by
# MAN_EXTENSION with the initial . removed.
# This tag requires that the tag GENERATE_MAN is set to YES.
MAN_SUBDIR =
# If the MAN_LINKS tag is set to YES and doxygen generates man output, then it
# will generate one additional man file for each entity documented in the real
# man page(s). These additional files only source the real man page, but without
# them the man command would be unable to find the correct page.
# The default value is: NO.
# This tag requires that the tag GENERATE_MAN is set to YES.
MAN_LINKS = NO
#---------------------------------------------------------------------------
# Configuration options related to the XML output
#---------------------------------------------------------------------------
# If the GENERATE_XML tag is set to YES, doxygen will generate an XML file that
# captures the structure of the code including all documentation.
# The default value is: NO.
GENERATE_XML = YES
# The XML_OUTPUT tag is used to specify where the XML pages will be put. If a
# relative path is entered the value of OUTPUT_DIRECTORY will be put in front of
# it.
# The default directory is: xml.
# This tag requires that the tag GENERATE_XML is set to YES.
XML_OUTPUT = xml
# If the XML_PROGRAMLISTING tag is set to YES, doxygen will dump the program
# listings (including syntax highlighting and cross-referencing information) to
# the XML output. Note that enabling this will significantly increase the size
# of the XML output.
# The default value is: YES.
# This tag requires that the tag GENERATE_XML is set to YES.
XML_PROGRAMLISTING = YES
#---------------------------------------------------------------------------
# Configuration options related to the DOCBOOK output
#---------------------------------------------------------------------------
# If the GENERATE_DOCBOOK tag is set to YES, doxygen will generate Docbook files
# that can be used to generate PDF.
# The default value is: NO.
GENERATE_DOCBOOK = NO
# The DOCBOOK_OUTPUT tag is used to specify where the Docbook pages will be put.
# If a relative path is entered the value of OUTPUT_DIRECTORY will be put in
# front of it.
# The default directory is: docbook.
# This tag requires that the tag GENERATE_DOCBOOK is set to YES.
DOCBOOK_OUTPUT = docbook
# If the DOCBOOK_PROGRAMLISTING tag is set to YES, doxygen will include the
# program listings (including syntax highlighting and cross-referencing
# information) to the DOCBOOK output. Note that enabling this will significantly
# increase the size of the DOCBOOK output.
# The default value is: NO.
# This tag requires that the tag GENERATE_DOCBOOK is set to YES.
DOCBOOK_PROGRAMLISTING = NO
#---------------------------------------------------------------------------
# Configuration options for the AutoGen Definitions output
#---------------------------------------------------------------------------
# If the GENERATE_AUTOGEN_DEF tag is set to YES, doxygen will generate an
# AutoGen Definitions (see http://autogen.sf.net) file that captures the
# structure of the code including all documentation. Note that this feature is
# still experimental and incomplete at the moment.
# The default value is: NO.
GENERATE_AUTOGEN_DEF = NO
#---------------------------------------------------------------------------
# Configuration options related to the Perl module output
#---------------------------------------------------------------------------
# If the GENERATE_PERLMOD tag is set to YES, doxygen will generate a Perl module
# file that captures the structure of the code including all documentation.
#
# Note that this feature is still experimental and incomplete at the moment.
# The default value is: NO.
GENERATE_PERLMOD = NO
# If the PERLMOD_LATEX tag is set to YES, doxygen will generate the necessary
# Makefile rules, Perl scripts and LaTeX code to be able to generate PDF and DVI
# output from the Perl module output.
# The default value is: NO.
# This tag requires that the tag GENERATE_PERLMOD is set to YES.
PERLMOD_LATEX = NO
# If the PERLMOD_PRETTY tag is set to YES, the Perl module output will be nicely
# formatted so it can be parsed by a human reader. This is useful if you want to
# understand what is going on. On the other hand, if this tag is set to NO, the
# size of the Perl module output will be much smaller and Perl will parse it
# just the same.
# The default value is: YES.
# This tag requires that the tag GENERATE_PERLMOD is set to YES.
PERLMOD_PRETTY = YES
# The names of the make variables in the generated doxyrules.make file are
# prefixed with the string contained in PERLMOD_MAKEVAR_PREFIX. This is useful
# so different doxyrules.make files included by the same Makefile don't
# overwrite each other's variables.
# This tag requires that the tag GENERATE_PERLMOD is set to YES.
PERLMOD_MAKEVAR_PREFIX =
#---------------------------------------------------------------------------
# Configuration options related to the preprocessor
#---------------------------------------------------------------------------
# If the ENABLE_PREPROCESSING tag is set to YES, doxygen will evaluate all
# C-preprocessor directives found in the sources and include files.
# The default value is: YES.
ENABLE_PREPROCESSING = YES
# If the MACRO_EXPANSION tag is set to YES, doxygen will expand all macro names
# in the source code. If set to NO, only conditional compilation will be
# performed. Macro expansion can be done in a controlled way by setting
# EXPAND_ONLY_PREDEF to YES.
# The default value is: NO.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
MACRO_EXPANSION = NO
# If the EXPAND_ONLY_PREDEF and MACRO_EXPANSION tags are both set to YES then
# the macro expansion is limited to the macros specified with the PREDEFINED and
# EXPAND_AS_DEFINED tags.
# The default value is: NO.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
EXPAND_ONLY_PREDEF = NO
# If the SEARCH_INCLUDES tag is set to YES, the include files in the
# INCLUDE_PATH will be searched if a #include is found.
# The default value is: YES.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
SEARCH_INCLUDES = YES
# The INCLUDE_PATH tag can be used to specify one or more directories that
# contain include files that are not input files but should be processed by the
# preprocessor.
# This tag requires that the tag SEARCH_INCLUDES is set to YES.
INCLUDE_PATH =
# You can use the INCLUDE_FILE_PATTERNS tag to specify one or more wildcard
# patterns (like *.h and *.hpp) to filter out the header-files in the
# directories. If left blank, the patterns specified with FILE_PATTERNS will be
# used.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
INCLUDE_FILE_PATTERNS =
# The PREDEFINED tag can be used to specify one or more macro names that are
# defined before the preprocessor is started (similar to the -D option of e.g.
# gcc). The argument of the tag is a list of macros of the form: name or
# name=definition (no spaces). If the definition and the "=" are omitted, "=1"
# is assumed. To prevent a macro definition from being undefined via #undef or
# recursively expanded use the := operator instead of the = operator.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
PREDEFINED =
# If the MACRO_EXPANSION and EXPAND_ONLY_PREDEF tags are set to YES then this
# tag can be used to specify a list of macro names that should be expanded. The
# macro definition that is found in the sources will be used. Use the PREDEFINED
# tag if you want to use a different macro definition that overrules the
# definition found in the source code.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
EXPAND_AS_DEFINED =
# If the SKIP_FUNCTION_MACROS tag is set to YES then doxygen's preprocessor will
# remove all references to function-like macros that are alone on a line, have
# an all uppercase name, and do not end with a semicolon. Such function macros
# are typically used for boiler-plate code, and will confuse the parser if not
# removed.
# The default value is: YES.
# This tag requires that the tag ENABLE_PREPROCESSING is set to YES.
SKIP_FUNCTION_MACROS = YES
#---------------------------------------------------------------------------
# Configuration options related to external references
#---------------------------------------------------------------------------
# The TAGFILES tag can be used to specify one or more tag files. For each tag
# file the location of the external documentation should be added. The format of
# a tag file without this location is as follows:
# TAGFILES = file1 file2 ...
# Adding location for the tag files is done as follows:
# TAGFILES = file1=loc1 "file2 = loc2" ...
# where loc1 and loc2 can be relative or absolute paths or URLs. See the
# section "Linking to external documentation" for more information about the use
# of tag files.
# Note: Each tag file must have a unique name (where the name does NOT include
# the path). If a tag file is not located in the directory in which doxygen is
# run, you must also specify the path to the tagfile here.
TAGFILES =
# When a file name is specified after GENERATE_TAGFILE, doxygen will create a
# tag file that is based on the input files it reads. See section "Linking to
# external documentation" for more information about the usage of tag files.
GENERATE_TAGFILE =
# If the ALLEXTERNALS tag is set to YES, all external class will be listed in
# the class index. If set to NO, only the inherited external classes will be
# listed.
# The default value is: NO.
ALLEXTERNALS = NO
# If the EXTERNAL_GROUPS tag is set to YES, all external groups will be listed
# in the modules index. If set to NO, only the current project's groups will be
# listed.
# The default value is: YES.
EXTERNAL_GROUPS = YES
# If the EXTERNAL_PAGES tag is set to YES, all external pages will be listed in
# the related pages index. If set to NO, only the current project's pages will
# be listed.
# The default value is: YES.
EXTERNAL_PAGES = YES
# The PERL_PATH should be the absolute path and name of the perl script
# interpreter (i.e. the result of 'which perl').
# The default file (with absolute path) is: /usr/bin/perl.
PERL_PATH = /usr/bin/perl
#---------------------------------------------------------------------------
# Configuration options related to the dot tool
#---------------------------------------------------------------------------
# If the CLASS_DIAGRAMS tag is set to YES, doxygen will generate a class diagram
# (in HTML and LaTeX) for classes with base or super classes. Setting the tag to
# NO turns the diagrams off. Note that this option also works with HAVE_DOT
# disabled, but it is recommended to install and use dot, since it yields more
# powerful graphs.
# The default value is: YES.
CLASS_DIAGRAMS = YES
# You can define message sequence charts within doxygen comments using the \msc
# command. Doxygen will then run the mscgen tool (see:
# http://www.mcternan.me.uk/mscgen/)) to produce the chart and insert it in the
# documentation. The MSCGEN_PATH tag allows you to specify the directory where
# the mscgen tool resides. If left empty the tool is assumed to be found in the
# default search path.
MSCGEN_PATH =
# You can include diagrams made with dia in doxygen documentation. Doxygen will
# then run dia to produce the diagram and insert it in the documentation. The
# DIA_PATH tag allows you to specify the directory where the dia binary resides.
# If left empty dia is assumed to be found in the default search path.
DIA_PATH =
# If set to YES the inheritance and collaboration graphs will hide inheritance
# and usage relations if the target is undocumented or is not a class.
# The default value is: YES.
HIDE_UNDOC_RELATIONS = YES
# If you set the HAVE_DOT tag to YES then doxygen will assume the dot tool is
# available from the path. This tool is part of Graphviz (see:
# http://www.graphviz.org/), a graph visualization toolkit from AT&T and Lucent
# Bell Labs. The other options in this section have no effect if this option is
# set to NO
# The default value is: NO.
HAVE_DOT = NO
# The DOT_NUM_THREADS specifies the number of dot invocations doxygen is allowed
# to run in parallel. When set to 0 doxygen will base this on the number of
# processors available in the system. You can set it explicitly to a value
# larger than 0 to get control over the balance between CPU load and processing
# speed.
# Minimum value: 0, maximum value: 32, default value: 0.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_NUM_THREADS = 0
# When you want a differently looking font in the dot files that doxygen
# generates you can specify the font name using DOT_FONTNAME. You need to make
# sure dot is able to find the font, which can be done by putting it in a
# standard location or by setting the DOTFONTPATH environment variable or by
# setting DOT_FONTPATH to the directory containing the font.
# The default value is: Helvetica.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_FONTNAME = Helvetica
# The DOT_FONTSIZE tag can be used to set the size (in points) of the font of
# dot graphs.
# Minimum value: 4, maximum value: 24, default value: 10.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_FONTSIZE = 10
# By default doxygen will tell dot to use the default font as specified with
# DOT_FONTNAME. If you specify a different font using DOT_FONTNAME you can set
# the path where dot can find it using this tag.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_FONTPATH =
# If the CLASS_GRAPH tag is set to YES then doxygen will generate a graph for
# each documented class showing the direct and indirect inheritance relations.
# Setting this tag to YES will force the CLASS_DIAGRAMS tag to NO.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
CLASS_GRAPH = YES
# If the COLLABORATION_GRAPH tag is set to YES then doxygen will generate a
# graph for each documented class showing the direct and indirect implementation
# dependencies (inheritance, containment, and class references variables) of the
# class with other documented classes.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
COLLABORATION_GRAPH = YES
# If the GROUP_GRAPHS tag is set to YES then doxygen will generate a graph for
# groups, showing the direct groups dependencies.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
GROUP_GRAPHS = YES
# If the UML_LOOK tag is set to YES, doxygen will generate inheritance and
# collaboration diagrams in a style similar to the OMG's Unified Modeling
# Language.
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
UML_LOOK = NO
# If the UML_LOOK tag is enabled, the fields and methods are shown inside the
# class node. If there are many fields or methods and many nodes the graph may
# become too big to be useful. The UML_LIMIT_NUM_FIELDS threshold limits the
# number of items for each type to make the size more manageable. Set this to 0
# for no limit. Note that the threshold may be exceeded by 50% before the limit
# is enforced. So when you set the threshold to 10, up to 15 fields may appear,
# but if the number exceeds 15, the total amount of fields shown is limited to
# 10.
# Minimum value: 0, maximum value: 100, default value: 10.
# This tag requires that the tag HAVE_DOT is set to YES.
UML_LIMIT_NUM_FIELDS = 10
# If the TEMPLATE_RELATIONS tag is set to YES then the inheritance and
# collaboration graphs will show the relations between templates and their
# instances.
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
TEMPLATE_RELATIONS = NO
# If the INCLUDE_GRAPH, ENABLE_PREPROCESSING and SEARCH_INCLUDES tags are set to
# YES then doxygen will generate a graph for each documented file showing the
# direct and indirect include dependencies of the file with other documented
# files.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
INCLUDE_GRAPH = YES
# If the INCLUDED_BY_GRAPH, ENABLE_PREPROCESSING and SEARCH_INCLUDES tags are
# set to YES then doxygen will generate a graph for each documented file showing
# the direct and indirect include dependencies of the file with other documented
# files.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
INCLUDED_BY_GRAPH = YES
# If the CALL_GRAPH tag is set to YES then doxygen will generate a call
# dependency graph for every global function or class method.
#
# Note that enabling this option will significantly increase the time of a run.
# So in most cases it will be better to enable call graphs for selected
# functions only using the \callgraph command. Disabling a call graph can be
# accomplished by means of the command \hidecallgraph.
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
CALL_GRAPH = NO
# If the CALLER_GRAPH tag is set to YES then doxygen will generate a caller
# dependency graph for every global function or class method.
#
# Note that enabling this option will significantly increase the time of a run.
# So in most cases it will be better to enable caller graphs for selected
# functions only using the \callergraph command. Disabling a caller graph can be
# accomplished by means of the command \hidecallergraph.
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
CALLER_GRAPH = NO
# If the GRAPHICAL_HIERARCHY tag is set to YES then doxygen will graphical
# hierarchy of all classes instead of a textual one.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
GRAPHICAL_HIERARCHY = YES
# If the DIRECTORY_GRAPH tag is set to YES then doxygen will show the
# dependencies a directory has on other directories in a graphical way. The
# dependency relations are determined by the #include relations between the
# files in the directories.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
DIRECTORY_GRAPH = YES
# The DOT_IMAGE_FORMAT tag can be used to set the image format of the images
# generated by dot. For an explanation of the image formats see the section
# output formats in the documentation of the dot tool (Graphviz (see:
# http://www.graphviz.org/)).
# Note: If you choose svg you need to set HTML_FILE_EXTENSION to xhtml in order
# to make the SVG files visible in IE 9+ (other browsers do not have this
# requirement).
# Possible values are: png, jpg, gif, svg, png:gd, png:gd:gd, png:cairo,
# png:cairo:gd, png:cairo:cairo, png:cairo:gdiplus, png:gdiplus and
# png:gdiplus:gdiplus.
# The default value is: png.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_IMAGE_FORMAT = png
# If DOT_IMAGE_FORMAT is set to svg, then this option can be set to YES to
# enable generation of interactive SVG images that allow zooming and panning.
#
# Note that this requires a modern browser other than Internet Explorer. Tested
# and working are Firefox, Chrome, Safari, and Opera.
# Note: For IE 9+ you need to set HTML_FILE_EXTENSION to xhtml in order to make
# the SVG files visible. Older versions of IE do not have SVG support.
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
INTERACTIVE_SVG = NO
# The DOT_PATH tag can be used to specify the path where the dot tool can be
# found. If left blank, it is assumed the dot tool can be found in the path.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_PATH =
# The DOTFILE_DIRS tag can be used to specify one or more directories that
# contain dot files that are included in the documentation (see the \dotfile
# command).
# This tag requires that the tag HAVE_DOT is set to YES.
DOTFILE_DIRS =
# The MSCFILE_DIRS tag can be used to specify one or more directories that
# contain msc files that are included in the documentation (see the \mscfile
# command).
MSCFILE_DIRS =
# The DIAFILE_DIRS tag can be used to specify one or more directories that
# contain dia files that are included in the documentation (see the \diafile
# command).
DIAFILE_DIRS =
# When using plantuml, the PLANTUML_JAR_PATH tag should be used to specify the
# path where java can find the plantuml.jar file. If left blank, it is assumed
# PlantUML is not used or called during a preprocessing step. Doxygen will
# generate a warning when it encounters a \startuml command in this case and
# will not generate output for the diagram.
PLANTUML_JAR_PATH =
# When using plantuml, the specified paths are searched for files specified by
# the !include statement in a plantuml block.
PLANTUML_INCLUDE_PATH =
# The DOT_GRAPH_MAX_NODES tag can be used to set the maximum number of nodes
# that will be shown in the graph. If the number of nodes in a graph becomes
# larger than this value, doxygen will truncate the graph, which is visualized
# by representing a node as a red box. Note that doxygen if the number of direct
# children of the root node in a graph is already larger than
# DOT_GRAPH_MAX_NODES then the graph will not be shown at all. Also note that
# the size of a graph can be further restricted by MAX_DOT_GRAPH_DEPTH.
# Minimum value: 0, maximum value: 10000, default value: 50.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_GRAPH_MAX_NODES = 50
# The MAX_DOT_GRAPH_DEPTH tag can be used to set the maximum depth of the graphs
# generated by dot. A depth value of 3 means that only nodes reachable from the
# root by following a path via at most 3 edges will be shown. Nodes that lay
# further from the root node will be omitted. Note that setting this option to 1
# or 2 may greatly reduce the computation time needed for large code bases. Also
# note that the size of a graph can be further restricted by
# DOT_GRAPH_MAX_NODES. Using a depth of 0 means no depth restriction.
# Minimum value: 0, maximum value: 1000, default value: 0.
# This tag requires that the tag HAVE_DOT is set to YES.
MAX_DOT_GRAPH_DEPTH = 0
# Set the DOT_TRANSPARENT tag to YES to generate images with a transparent
# background. This is disabled by default, because dot on Windows does not seem
# to support this out of the box.
#
# Warning: Depending on the platform used, enabling this option may lead to
# badly anti-aliased labels on the edges of a graph (i.e. they become hard to
# read).
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_TRANSPARENT = NO
# Set the DOT_MULTI_TARGETS tag to YES to allow dot to generate multiple output
# files in one run (i.e. multiple -o and -T options on the command line). This
# makes dot run faster, but since only newer versions of dot (>1.8.10) support
# this, this feature is disabled by default.
# The default value is: NO.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_MULTI_TARGETS = NO
# If the GENERATE_LEGEND tag is set to YES doxygen will generate a legend page
# explaining the meaning of the various boxes and arrows in the dot generated
# graphs.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
GENERATE_LEGEND = YES
# If the DOT_CLEANUP tag is set to YES, doxygen will remove the intermediate dot
# files that are used to generate the various graphs.
# The default value is: YES.
# This tag requires that the tag HAVE_DOT is set to YES.
DOT_CLEANUP = YES
doc/about/index.rst 0 → 100644
浏览文件 @ 021b3a44
ABOUT
=======
PaddlPaddle is an easy-to-use, efficient, flexible and scalable deep learning platform,
which is originally developed by Baidu scientists and engineers for the purpose of applying deep learning to many products at Baidu.
PaddlePaddle is now open source but far from complete, which is intended to be built upon, improved, scaled, and extended.
We hope to build an active open source community both by providing feedback and by actively contributing to the source code.
Credits
--------
We owe many thanks to `all contributors and developers <https://github.com/PaddlePaddle/Paddle/blob/develop/authors>`_ of PaddlePaddle!
doc/algorithm/rnn/bi_lstm.jpg 已删除 120000 → 0
浏览文件 @ 4f1bf30d
../../demo/sentiment_analysis/bi_lstm.jpg
\ No newline at end of file
doc/algorithm/rnn/encoder-decoder-attention-model.png 已删除 120000 → 0
浏览文件 @ 4f1bf30d
../../demo/text_generation/encoder-decoder-attention-model.png
\ No newline at end of file
doc/ui/data_provider/index.rst → doc/api/data_provider/index.rst
浏览文件 @ 021b3a44
DataProvider Introduction Introduction
========================= ==============
DataProvider is a module that loads training or testing data into cpu or gpu DataProvider is a module that loads training or testing data into cpu or gpu
memory for the following triaining or testing process. memory for the following triaining or testing process.
......
doc/ui/data_provider/pydataprovider2.rst → doc/api/data_provider/pydataprovider2.rst
浏览文件 @ 021b3a44
How to use PyDataProvider2 PyDataProvider2
========================== =================
We highly recommand users to use PyDataProvider2 to provide training or testing 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 data to PaddlePaddle. The user only needs to focus on how to read a single
......
doc/api/index.rst 0 → 100644
浏览文件 @ 021b3a44
API
====
DataProvider API
----------------
.. toctree::
:maxdepth: 1
data_provider/index.rst
data_provider/pydataprovider2.rst
Model Config API
----------------
.. toctree::
:maxdepth: 1
trainer_config_helpers/index.rst
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
\ No newline at end of file
doc/ui/predict/swig_py_paddle_en.rst → doc/api/predict/swig_py_paddle_en.rst
浏览文件 @ 021b3a44
Python Prediction API Python Prediction
===================== ==================
PaddlePaddle offers a set of clean prediction interfaces for python with the help of PaddlePaddle offers a set of clean prediction interfaces for python with the help of
SWIG. The main steps of predict values in python are: SWIG. The main steps of predict values in python are:
......
doc/ui/api/trainer_config_helpers/attrs.rst → doc/api/trainer_config_helpers/attrs.rst
浏览文件 @ 021b3a44
Parameter and Extra Layer Attribute Parameter Attributes
=================================== =======================
.. automodule:: paddle.trainer_config_helpers.attrs .. automodule:: paddle.trainer_config_helpers.attrs
:members: :members:
doc/build/docker_install.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
Docker installation guide
==========================
PaddlePaddle provide the `Docker <https://www.docker.com/>`_ image. `Docker`_ is a lightweight container utilities. The performance of PaddlePaddle in `Docker`_ container is basically as same as run it in a normal linux. The `Docker`_ is a very convenient way to deliver the binary release for linux programs.
.. note::
The `Docker`_ image is the recommended way to run PaddlePaddle
PaddlePaddle Docker images
--------------------------
There are 12 `images <https://hub.docker.com/r/paddledev/paddle/tags/>`_ for PaddlePaddle, and the name is :code:`paddle-dev/paddle`, tags are\:
+-----------------+------------------+------------------------+-----------------------+
| | normal | devel | demo |
+=================+==================+========================+=======================+
| CPU | cpu-latest | cpu-devel-latest | cpu-demo-latest |
+-----------------+------------------+------------------------+-----------------------+
| GPU | gpu-latest | gpu-devel-latest | gpu-demo-latest |
+-----------------+------------------+------------------------+-----------------------+
| CPU WITHOUT AVX | cpu-noavx-latest | cpu-devel-noavx-latest | cpu-demo-noavx-latest |
+-----------------+------------------+------------------------+-----------------------+
| GPU WITHOUT AVX | gpu-noavx-latest | gpu-devel-noavx-latest | gpu-demo-noavx-latest |
+-----------------+------------------+------------------------+-----------------------+
And the three columns are:
* normal\: The docker image only contains binary of PaddlePaddle.
* devel\: The docker image contains PaddlePaddle binary, source code and essential build environment.
* demo\: The docker image contains the dependencies to run PaddlePaddle demo.
And the four rows are:
* CPU\: CPU Version. Support CPU which has :code:`AVX` instructions.
* GPU\: GPU Version. Support GPU, and cpu has :code:`AVX` instructions.
* CPU WITHOUT AVX\: CPU Version, which support most CPU even doesn't have :code:`AVX` instructions.
* GPU WITHOUT AVX\: GPU Version, which support most CPU even doesn't have :code:`AVX` instructions.
User can choose any version depends on machine. The following script can help you to detect your CPU support :code:`AVX` or not.
.. code-block:: bash
if cat /proc/cpuinfo | grep -q avx ; then echo "Support AVX"; else echo "Not support AVX"; fi
If the output is :code:`Support AVX`, then you can choose the AVX version of PaddlePaddle, otherwise, you need select :code:`noavx` version of PaddlePaddle. For example, the CPU develop version of PaddlePaddle is :code:`paddle-dev/paddle:cpu-devel-latest`.
The PaddlePaddle images don't contain any entry command. You need to write your entry command to use this image. See :code:`Remote Access` part or just use following command to run a :code:`bash`
.. code-block:: bash
docker run -it paddledev/paddle:cpu-latest /bin/bash
Download and Run Docker images
------------------------------
You have to install Docker in your machine which has linux kernel version 3.10+ first. You can refer to the official guide https://docs.docker.com/engine/installation/ for further information.
You can use :code:`docker pull ` to download images first, or just launch a container with :code:`docker run` \:
.. code-block:: bash
docker run -it paddledev/paddle:cpu-latest
If you want to launch container with GPU support, you need to set some environment variables at the same time:
.. code-block:: bash
export CUDA_SO="$(\ls /usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')"
export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
docker run ${CUDA_SO} ${DEVICES} -it paddledev/paddle:gpu-latest
Some notes for docker
---------------------
Performance
+++++++++++
Since Docker is based on the lightweight virtual containers, the CPU computing performance maintains well. And GPU driver and equipments are all mapped to the container, so the GPU computing performance would not be seriously affected.
If you use high performance nic, such as RDMA(RoCE 40GbE or IB 56GbE), Ethernet(10GbE), it is recommended to use config "-net = host".
Remote access
+++++++++++++
If you want to enable ssh access background, you need to build an image by yourself. Please refer to official guide https://docs.docker.com/engine/reference/builder/ for further information.
Following is a simple Dockerfile with ssh:
.. literalinclude:: ../../doc_cn/build_and_install/install/paddle_ssh.Dockerfile
Then you can build an image with Dockerfile and launch a container:
.. code-block:: bash
# cd into Dockerfile directory
docker build . -t paddle_ssh
# run container, and map host machine port 8022 to container port 22
docker run -d -p 8022:22 --name paddle_ssh_machine paddle_ssh
Now, you can ssh on port 8022 to access the container, username is root, password is also root:
.. code-block:: bash
ssh -p 8022 root@YOUR_HOST_MACHINE
You can stop and delete the container as following:
.. code-block:: bash
# stop
docker stop paddle_ssh_machine
# delete
docker rm paddle_ssh_machine
doc/cluster/index.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
Cluster Train
====================
.. toctree::
:glob:
opensource/cluster_train.md
internal/index.md
doc/conf.py.in
浏览文件 @ 021b3a44
...@@ -25,20 +25,6 @@ sys.path.insert(0, '@PROJ_ROOT@/python') ...@@ -25,20 +25,6 @@ sys.path.insert(0, '@PROJ_ROOT@/python')
templates_path = ["@PROJ_ROOT@/doc/templates"] templates_path = ["@PROJ_ROOT@/doc/templates"]
# -- Doxygen Settings
breathe_projects = {
'paddle': '@PADDLE_DOXYGEN_OUTPUT@/xml'
}
breathe_default_project = 'paddle'
breathe_domain_by_extension = {
'h': 'cpp', # mapping XXX.h XXX.cpp together
}
breathe_default_members = {
'protected-members','undoc-members'
}
# -- General configuration ------------------------------------------------ # -- General configuration ------------------------------------------------
# General information about the project. # General information about the project.
...@@ -62,7 +48,6 @@ extensions = [ ...@@ -62,7 +48,6 @@ extensions = [
'sphinx.ext.autosummary', 'sphinx.ext.autosummary',
'sphinx.ext.mathjax', 'sphinx.ext.mathjax',
'sphinx.ext.napoleon', 'sphinx.ext.napoleon',
'breathe'
] ]
......
doc/dev/index.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
Development Guide
=================
.. toctree::
:maxdepth: 1
layer.md
new_layer/new_layer.rst
../source/index.md
doc/dev/layer.md 已删除 100644 → 0
浏览文件 @ 4f1bf30d
# Layer Documents
* [Layer Source Code Document](../source/gserver/layers/index.rst)
* [Layer Python API Document](../ui/api/trainer_config_helpers/index.rst)
doc/introduction/index.md → doc/getstarted/basic_usage/basic_usage.rst
浏览文件 @ 021b3a44
# Introduction Basic Usage
=============
PaddlePaddle is a deep learning platform open-sourced by Baidu. With PaddlePaddle, you can easily train a classic neural network within a couple lines of configuration, or you can build sophisticated models that provide state-of-the-art performance on difficult learning tasks like sentiment analysis, machine translation, image caption and so on. PaddlePaddle is a deep learning platform open-sourced by Baidu. With PaddlePaddle, you can easily train a classic neural network within a couple lines of configuration, or you can build sophisticated models that provide state-of-the-art performance on difficult learning tasks like sentiment analysis, machine translation, image caption and so on.
## 1. A Classic Problem 1. A Classic Problem
---------------------
Now, to give you a hint of what using PaddlePaddle looks like, let's start with a fundamental learning problem - <a href="https://en.wikipedia.org/wiki/Simple_linear_regression">**simple linear regression**</a> : you have observed a set of two-dimensional data points of `X` and `Y`, where `X` is an explanatory variable and `Y` is corresponding dependent variable, and you want to recover the underlying correlation between `X` and `Y`. Linear regression can be used in many practical scenarios. For example, `X` can be a variable about house size, and `Y` a variable about house price. You can build a model that captures relationship between them by observing real estate markets. Now, to give you a hint of what using PaddlePaddle looks like, let's start with a fundamental learning problem - `simple linear regression <https://en.wikipedia.org/wiki/Simple_linear_regression>`_: you have observed a set of two-dimensional data points of ``X`` and ``Y``, where ``X`` is an explanatory variable and ``Y`` is corresponding dependent variable, and you want to recover the underlying correlation between ``X`` and ``Y``. Linear regression can be used in many practical scenarios. For example, ``X`` can be a variable about house size, and ``Y`` a variable about house price. You can build a model that captures relationship between them by observing real estate markets.
## 2. Prepare the Data 2. Prepare the Data
--------------------
Suppose the true relationship can be characterized as `Y = 2X + 0.3`, let's see how to recover this pattern only from observed data. Here is a piece of python code that feeds synthetic data to PaddlePaddle. The code is pretty self-explanatory, the only extra thing you need to add for PaddlePaddle is a definition of input data types. Suppose the true relationship can be characterized as ``Y = 2X + 0.3``, let's see how to recover this pattern only from observed data. Here is a piece of python code that feeds synthetic data to PaddlePaddle. The code is pretty self-explanatory, the only extra thing you need to add for PaddlePaddle is a definition of input data types.
```python .. code-block:: python
# dataprovider.py
from paddle.trainer.PyDataProvider2 import *
import random
# define data types of input: 2 real numbers # dataprovider.py
@provider(input_types=[dense_vector(1), dense_vector(1)],use_seq=False) from paddle.trainer.PyDataProvider2 import *
def process(settings, input_file): import random
# define data types of input: 2 real numbers
@provider(input_types=[dense_vector(1), dense_vector(1)],use_seq=False)
def process(settings, input_file):
for i in xrange(2000): for i in xrange(2000):
x = random.random() x = random.random()
yield [x], [2*x+0.3] yield [x], [2*x+0.3]
```
## 3. Train a NeuralNetwork in PaddlePaddle 3. Train a NeuralNetwork
-------------------------
To recover this relationship between ``X`` and ``Y``, we use a neural network with one layer of linear activation units and a square error cost layer. Don't worry if you are not familiar with these terminologies, it's just saying that we are starting from a random line ``Y' = wX + b`` , then we gradually adapt ``w`` and ``b`` to minimize the difference between ``Y'`` and ``Y``. Here is what it looks like in PaddlePaddle:
To recover this relationship between `X` and `Y`, we use a neural network with one layer of linear activation units and a square error cost layer. Don't worry if you are not familiar with these terminologies, it's just saying that we are starting from a random line `Y' = wX + b` , then we gradually adapt `w` and `b` to minimize the difference between `Y'` and `Y`. Here is what it looks like in PaddlePaddle: .. code-block:: python
```python # trainer_config.py
# trainer_config.py from paddle.trainer_config_helpers import *
from paddle.trainer_config_helpers import *
# 1. read data. Suppose you saved above python code as dataprovider.py # 1. read data. Suppose you saved above python code as dataprovider.py
data_file = 'empty.list' data_file = 'empty.list'
with open(data_file, 'w') as f: f.writelines(' ') with open(data_file, 'w') as f: f.writelines(' ')
define_py_data_sources2(train_list=data_file, test_list=None, define_py_data_sources2(train_list=data_file, test_list=None,
module='dataprovider', obj='process',args={}) module='dataprovider', obj='process',args={})
# 2. learning algorithm # 2. learning algorithm
settings(batch_size=12, learning_rate=1e-3, learning_method=MomentumOptimizer()) settings(batch_size=12, learning_rate=1e-3, learning_method=MomentumOptimizer())
# 3. Network configuration # 3. Network configuration
x = data_layer(name='x', size=1) x = data_layer(name='x', size=1)
y = data_layer(name='y', size=1) y = data_layer(name='y', size=1)
y_predict = fc_layer(input=x, param_attr=ParamAttr(name='w'), size=1, act=LinearActivation(), bias_attr=ParamAttr(name='b')) y_predict = fc_layer(input=x, param_attr=ParamAttr(name='w'), size=1, act=LinearActivation(), bias_attr=ParamAttr(name='b'))
cost = regression_cost(input=y_predict, label=y) cost = regression_cost(input=y_predict, label=y)
outputs(cost) outputs(cost)
```
Some of the most fundamental usages of PaddlePaddle are demonstrated: Some of the most fundamental usages of PaddlePaddle are demonstrated:
...@@ -55,46 +59,51 @@ Some of the most fundamental usages of PaddlePaddle are demonstrated: ...@@ -55,46 +59,51 @@ Some of the most fundamental usages of PaddlePaddle are demonstrated:
- The second part describes learning algorithm. It defines in what ways adjustments are made to model parameters. PaddlePaddle provides a rich set of optimizers, but a simple momentum based optimizer will suffice here, and it processes 12 data points each time. - The second part describes learning algorithm. It defines in what ways adjustments are made to model parameters. PaddlePaddle provides a rich set of optimizers, but a simple momentum based optimizer will suffice here, and it processes 12 data points each time.
- Finally, the network configuration. It usually is as simple as "stacking" layers. Three kinds of layers are used in this configuration: - Finally, the network configuration. It usually is as simple as "stacking" layers. Three kinds of layers are used in this configuration:
- **Data Layer**: a network always starts with one or more data layers. They provide input data to the rest of the network. In this problem, two data layers are used respectively for `X` and `Y`. - **Data Layer**: a network always starts with one or more data layers. They provide input data to the rest of the network. In this problem, two data layers are used respectively for ``X`` and ``Y``.
- **FC Layer**: FC layer is short for Fully Connected Layer, which connects all the input units to current layer and does the actual computation specified as activation function. Computation layers like this are the fundamental building blocks of a deeper model. - **FC Layer**: FC layer is short for Fully Connected Layer, which connects all the input units to current layer and does the actual computation specified as activation function. Computation layers like this are the fundamental building blocks of a deeper model.
- **Cost Layer**: in training phase, cost layers are usually the last layers of the network. They measure the performance of current model, and provide guidence to adjust parameters. - **Cost Layer**: in training phase, cost layers are usually the last layers of the network. They measure the performance of current model, and provide guidence to adjust parameters.
Now that everything is ready, you can train the network with a simple command line call: Now that everything is ready, you can train the network with a simple command line call:
```
.. code-block:: bash
paddle train --config=trainer_config.py --save_dir=./output --num_passes=30 paddle train --config=trainer_config.py --save_dir=./output --num_passes=30
```
This means that PaddlePaddle will train this network on the synthectic dataset for 30 passes, and save all the models under path `./output`. You will see from the messages printed out during training phase that the model cost is decreasing as time goes by, which indicates we are getting a closer guess.
This means that PaddlePaddle will train this network on the synthectic dataset for 30 passes, and save all the models under path ``./output``. You will see from the messages printed out during training phase that the model cost is decreasing as time goes by, which indicates we are getting a closer guess.
4. Evaluate the Model
-----------------------
## 4. Evaluate the Model Usually, a different dataset that left out during training phase should be used to evalute the models. However, we are lucky enough to know the real answer: ``w=2, b=0.3``, thus a better option is to check out model parameters directly.
Usually, a different dataset that left out during training phase should be used to evalute the models. However, we are lucky enough to know the real answer: `w=2, b=0.3`, thus a better option is to check out model parameters directly. In PaddlePaddle, training is just to get a collection of model parameters, which are ``w`` and ``b`` in this case. Each parameter is saved in an individual file in the popular ``numpy`` array format. Here is the code that reads parameters from last pass.
In PaddlePaddle, training is just to get a collection of model parameters, which are `w` and `b` in this case. Each parameter is saved in an individual file in the popular `numpy` array format. Here is the code that reads parameters from last pass. .. code-block:: python
```python import numpy as np
import numpy as np import os
import os
def load(file_name): def load(file_name):
with open(file_name, 'rb') as f: with open(file_name, 'rb') as f:
f.read(16) # skip header for float type. f.read(16) # skip header for float type.
return np.fromfile(f, dtype=np.float32) return np.fromfile(f, dtype=np.float32)
print 'w=%.6f, b=%.6f' % (load('output/pass-00029/w'), load('output/pass-00029/b')) print 'w=%.6f, b=%.6f' % (load('output/pass-00029/w'), load('output/pass-00029/b'))
# w=1.999743, b=0.300137 # w=1.999743, b=0.300137
```
<center> ![](./parameters.png) </center> .. image:: parameters.png
:align: center
Although starts from a random guess, you can see that value of `w` changes quickly towards 2 and `b` changes quickly towards 0.3. In the end, the predicted line is almost identical with real answer. Although starts from a random guess, you can see that value of ``w`` changes quickly towards 2 and ``b`` changes quickly towards 0.3. In the end, the predicted line is almost identical with real answer.
There, you have recovered the underlying pattern between `X` and `Y` only from observed data. There, you have recovered the underlying pattern between ``X`` and ``Y`` only from observed data.
## 5. Where to Go from Here 5. Where to Go from Here
-------------------------
- <a href="../build/index.html"> Build and Installation </a> - `Install and Build <../build_and_install/index.html>`_
- <a href="../demo/quick_start/index_en.html">Quick Start</a> - `Tutorials <../demo/quick_start/index_en.html>`_
- <a href="../demo/index.html">Example and Demo</a> - `Example and Demo <../demo/index.html>`_
doc/getstarted/build_and_install/docker_install.rst 0 → 100644
浏览文件 @ 021b3a44
PaddlePaddle in Docker Containers
=================================
Docker container is currently the only officially-supported way to
running PaddlePaddle. This is reasonable as Docker now runs on all
major operating systems including Linux, Mac OS X, and Windows.
Please be aware that you will need to change `Dockers settings
<https://github.com/PaddlePaddle/Paddle/issues/627>`_ to make full use
of your hardware resource on Mac OS X and Windows.
CPU-only and GPU Images
-----------------------
For each version of PaddlePaddle, we release 2 Docker images, a
CPU-only one and a CUDA GPU one. We do so by configuring
`dockerhub.com <https://hub.docker.com/r/paddledev/paddle/>`_
automatically runs the following commands:
.. code-block:: base
docker build -t paddle:cpu-noavx -f paddle/scripts/docker/Dockerfile .
docker build -t paddle:gpu-noavx -f paddle/scripts/docker/Dockerfile.gpu .
To run the CPU-only image as an interactive container:
.. code-block:: bash
docker run -it --rm paddledev/paddle:cpu-latest /bin/bash
or, we can run it as a daemon container
.. code-block:: bash
docker run -d -p 2202:22 paddledev/paddle:cpu-latest
and SSH to this container using password :code:`root`:
.. code-block:: bash
ssh -p 2202 root@localhost
An advantage of using SSH is that we can connect to PaddlePaddle from
more than one terminals. For example, one terminal running vi and
another one running Python interpreter. Another advantage is that we
can run the PaddlePaddle container on a remote server and SSH to it
from a laptop.
Above methods work with the GPU image too -- just please don't forget
to install CUDA driver and let Docker knows about it:
.. code-block:: bash
export CUDA_SO="$(\ls /usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')"
export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
docker run ${CUDA_SO} ${DEVICES} -it paddledev/paddle:gpu-latest
Non-AVX Images
--------------
Please be aware that the CPU-only and the GPU images both use the AVX
instruction set, but old computers produced before 2008 do not support
AVX. The following command checks if your Linux computer supports
AVX:
.. code-block:: bash
if cat /proc/cpuinfo | grep -i avx; then echo Yes; else echo No; fi
If it doesn't, we will need to build non-AVX images manually from
source code:
.. code-block:: bash
cd ~
git clone github.com/PaddlePaddle/Paddle
cd Paddle
docker build --build-arg WITH_AVX=OFF -t paddle:cpu-noavx -f paddle/scripts/docker/Dockerfile .
docker build --build-arg WITH_AVX=OFF -t paddle:gpu-noavx -f paddle/scripts/docker/Dockerfile.gpu .
doc/build/index.rst → doc/getstarted/build_and_install/index.rst
浏览文件 @ 021b3a44
...@@ -8,8 +8,6 @@ Install PaddlePaddle ...@@ -8,8 +8,6 @@ Install PaddlePaddle
:maxdepth: 1 :maxdepth: 1
:glob: :glob:
install_*
internal/install_from_jumbo.md
docker_install.rst docker_install.rst
ubuntu_install.rst ubuntu_install.rst
...@@ -25,4 +23,3 @@ Build from Source ...@@ -25,4 +23,3 @@ Build from Source
:glob: :glob:
build_from_source.md build_from_source.md
\ No newline at end of file
contribute_to_paddle.md
doc/getstarted/index.rst 0 → 100644
浏览文件 @ 021b3a44
GET STARTED
============
.. toctree::
:maxdepth: 2
build_and_install/index.rst
basic_usage/basic_usage.rst
doc/cluster/opensource/cluster_train.md → doc/howto/cluster/cluster_train.md
浏览文件 @ 021b3a44
# Distributed Training # How to Run Distributed Training
In this article, we explain how to run distributed Paddle training jobs on clusters. We will create the distributed version of the single-process training example, [recommendation](https://github.com/baidu/Paddle/tree/develop/demo/recommendation). In this article, we explain how to run distributed Paddle training jobs on clusters. We will create the distributed version of the single-process training example, [recommendation](https://github.com/baidu/Paddle/tree/develop/demo/recommendation).
...@@ -9,7 +9,7 @@ In this article, we explain how to run distributed Paddle training jobs on clust ...@@ -9,7 +9,7 @@ In this article, we explain how to run distributed Paddle training jobs on clust
1. Aforementioned scripts use a Python library [fabric](http://www.fabfile.org/) to run SSH commands. We can use `pip` to install fabric: 1. Aforementioned scripts use a Python library [fabric](http://www.fabfile.org/) to run SSH commands. We can use `pip` to install fabric:
```bash ```bash
pip install fabric pip install fabric
``` ```
1. We need to install PaddlePaddle on all nodes in the cluster. To enable GPUs, we need to install CUDA in `/usr/local/cuda`; otherwise Paddle would report errors at runtime. 1. We need to install PaddlePaddle on all nodes in the cluster. To enable GPUs, we need to install CUDA in `/usr/local/cuda`; otherwise Paddle would report errors at runtime.
......
doc/howto/cmd_parameter/index.md 0 → 100644
浏览文件 @ 021b3a44
# How to Set Command-line Parameters
* [Use Case](use_case.md)
* [Arguments](arguments.md)
* [Detailed Descriptions](detail_introduction.md)
doc/build/contribute_to_paddle.md → doc/howto/contribute_to_paddle.md
浏览文件 @ 021b3a44
# Contribute Code # How to Contribute Code
We sincerely appreciate your contributions. You can use fork and pull request We sincerely appreciate your contributions. You can use fork and pull request
workflow to merge your code. workflow to merge your code.
......
doc/algorithm/index.rst → doc/howto/deep_model/index.rst
浏览文件 @ 021b3a44
Algorithm Tutorial How to Configure Deep Models
================== ============================
.. toctree:: .. toctree::
:maxdepth: 1 :maxdepth: 1
......
doc/algorithm/rnn/rnn.rst → doc/howto/deep_model/rnn/rnn.rst
浏览文件 @ 021b3a44
...@@ -42,7 +42,7 @@ Simple Gated Recurrent Neural Network ...@@ -42,7 +42,7 @@ Simple Gated Recurrent Neural Network
Recurrent neural network process a sequence at each time step sequentially. An example of the architecture of LSTM is listed below. Recurrent neural network process a sequence at each time step sequentially. An example of the architecture of LSTM is listed below.
.. image:: ./bi_lstm.jpg .. image:: ../../../tutorials/sentiment_analysis/bi_lstm.jpg
:align: center :align: center
Generally speaking, a recurrent network perform the following operations from :math:`t=1` to :math:`t=T`, or reversely from :math:`t=T` to :math:`t=1`. Generally speaking, a recurrent network perform the following operations from :math:`t=1` to :math:`t=T`, or reversely from :math:`t=T` to :math:`t=1`.
...@@ -101,7 +101,7 @@ Sequence to Sequence Model with Attention ...@@ -101,7 +101,7 @@ Sequence to Sequence Model with Attention
----------------------------------------- -----------------------------------------
We will use the sequence to sequence model with attention as an example to demonstrate how you can configure complex recurrent neural network models. An illustration of the sequence to sequence model with attention is shown in the following figure. We will use the sequence to sequence model with attention as an example to demonstrate how you can configure complex recurrent neural network models. An illustration of the sequence to sequence model with attention is shown in the following figure.
.. image:: ./encoder-decoder-attention-model.png .. image:: ../../../tutorials/text_generation/encoder-decoder-attention-model.png
:align: center :align: center
In this model, the source sequence :math:`S = \{s_1, \dots, s_T\}` is encoded with a bidirectional gated recurrent neural networks. The hidden states of the bidirectional gated recurrent neural network :math:`H_S = \{H_1, \dots, H_T\}` is called *encoder vector* The decoder is a gated recurrent neural network. When decoding each token :math:`y_t`, the gated recurrent neural network generates a set of weights :math:`W_S^t = \{W_1^t, \dots, W_T^t\}`, which are used to compute a weighted sum of the encoder vector. The weighted sum of the encoder vector is utilized to condition the generation of the token :math:`y_t`. In this model, the source sequence :math:`S = \{s_1, \dots, s_T\}` is encoded with a bidirectional gated recurrent neural networks. The hidden states of the bidirectional gated recurrent neural network :math:`H_S = \{H_1, \dots, H_T\}` is called *encoder vector* The decoder is a gated recurrent neural network. When decoding each token :math:`y_t`, the gated recurrent neural network generates a set of weights :math:`W_S^t = \{W_1^t, \dots, W_T^t\}`, which are used to compute a weighted sum of the encoder vector. The weighted sum of the encoder vector is utilized to condition the generation of the token :math:`y_t`.
......
doc/howto/index.rst 0 → 100644
浏览文件 @ 021b3a44
HOW TO
=======
Usage
-------
.. toctree::
:maxdepth: 1
cmd_parameter/index.md
deep_model/index.rst
cluster/cluster_train.md
Development
------------
.. toctree::
:maxdepth: 1
new_layer/index.rst
contribute_to_paddle.md
Optimization
-------------
.. toctree::
:maxdepth: 1
optimization/index.rst
doc/dev/new_layer/new_layer.rst → doc/howto/new_layer/index.rst
浏览文件 @ 021b3a44
================== =======================
Writing New Layers How to Write New Layers
================== =======================
This tutorial will guide you to write customized layers in PaddlePaddle. We will utilize fully connected layer as an example to guide you through the following steps for writing a new layer. This tutorial will guide you to write customized layers in PaddlePaddle. We will utilize fully connected layer as an example to guide you through the following steps for writing a new layer.
......
doc/optimization/index.rst → doc/howto/optimization/index.rst
浏览文件 @ 021b3a44
Performance Tuning How to Tune GPU Performance
================== ===========================
.. toctree:: .. toctree::
:maxdepth: 3 :maxdepth: 3
......
doc/index.rst
浏览文件 @ 021b3a44
...@@ -4,8 +4,9 @@ PaddlePaddle Documentation ...@@ -4,8 +4,9 @@ PaddlePaddle Documentation
.. toctree:: .. toctree::
:maxdepth: 1 :maxdepth: 1
introduction/index.md getstarted/index.rst
user_guide.rst tutorials/index.md
dev/index.rst howto/index.rst
algorithm/index.rst api/index.rst
optimization/index.rst about/index.rst
\ No newline at end of file
doc/introduction/parameters.png 已删除 120000 → 0
浏览文件 @ 4f1bf30d
../../doc_cn/introduction/parameters.png
\ No newline at end of file
doc/demo/index.md → doc/tutorials/index.md
浏览文件 @ 021b3a44
# Examples and demos # TUTORIALS
There are serveral examples and demos here. There are serveral examples and demos here.
## Image ## Image
......
doc/demo/quick_start/index_en.md → doc/tutorials/quick_start/index_en.md
浏览文件 @ 021b3a44
...@@ -59,12 +59,11 @@ To build your text classification system, your code will need to perform five st ...@@ -59,12 +59,11 @@ To build your text classification system, your code will need to perform five st
## Preprocess data into standardized format ## 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. 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/baidu/Paddle) provides scripts for downloading data and preprocessing data as shown below. The data process takes several minutes (about 3 minutes in our machine). `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 ```bash
cd demo/quick_start cd demo/quick_start
./data/get_data.sh ./data/get_data.sh
./preprocess.sh
``` ```
## Transfer Data to Model ## Transfer Data to Model
......
doc/ui/api/trainer_config_helpers/index.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
Model Config Interface
======================
.. toctree::
:maxdepth: 1
optimizers.rst
data_sources.rst
layers.rst
activations.rst
poolings.rst
networks.rst
evaluators.rst
attrs.rst
doc/ui/index.md 已删除 100644 → 0
浏览文件 @ 4f1bf30d
# User Interface
## Data Provider
* [Introduction](data_provider/index.rst)
* [PyDataProvider2](data_provider/pydataprovider2.rst)
## API Reference
* [Model Config Interface](api/trainer_config_helpers/index.md)
## Command Line Argument
* [Use Case](cmd_argument/use_case.md)
* [Argument Outline](cmd_argument/argument_outline.md)
* [Detailed Descriptions](cmd_argument/detail_introduction.md)
## Predict
* [Python Prediction API](predict/swig_py_paddle_en.rst)
doc/user_guide.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
User Guide
==========
.. toctree::
:maxdepth: 1
demo/quick_start/index_en.md
build/index.rst
build/contribute_to_paddle.md
ui/index.md
ui/api/trainer_config_helpers/index.rst
demo/index.md
cluster/index.md
doc_cn/build/docker/build_docker_image.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
构建PaddlePaddle Docker Image
===========================
PaddlePaddle的Docker Image构建源码放置在 :code:`${源码根目录}/paddle/scripts/docker/`目录下。
该Image基于ubuntu 14.04。该目录下有两个文件，Dockerfile和build.sh。其中:
* Dockerfile是docker image的主要描述文件。描述了Docker image的构建步骤、各种参数和维护
人员等等。
* build.sh是docker image的主要构建步骤。
该image的构建在docker 1.12版本测试通过, 低于docker 1.12版本的情况下并没有测试。主要由于旧版本
的docker可能缺乏 :code:`--build-arg` 参数，从而不能在运行编译命令的时候接受参数。
同时，该构建脚本充分考虑了网络不稳定的情况，对于cuda的Toolkit有断点续传和传输速度过小重启下载的
简单优化。
使用脚本构建PaddlePaddle Docker Image
-------------------------------------------
该脚本的使用方法是，进入该源码目录，执行 :code:`docker build .` 命令。可以使用
:code:`--build-arg` 传入的配置参数包括:
* LOWEST\_DL\_SPEED\: 多线程下载过程中，最低线程的下载速度(默认单位是Bytes，可以传入10K,
10M，或者10G这样的单位)。如果小于这个下载速度，那么这个下载线程将会关闭。所有的下载线程关闭时，
下载进程会重启。
* WITH\_GPU\: ON or OFF。是否开启GPU功能。注意，编译PaddlePaddle的GPU版本并不需要一定在具有GPU
的机器上进行。但是，运行PaddlePaddle的GPU版本一定要在具有CUDA的机器上运行。
简单的使用样例为\:
.. code-block:: bash
cd ${源码根目录}/paddle/scripts/docker/
docker build --build-arg LOWEST_DL_SPEED=50K\
--build-arg WITH_GPU=ON \
--tag paddle_gpu:latest .
即可在本地编译出PaddlePaddle的镜像。
doc_cn/build_and_install/index.rst
浏览文件 @ 021b3a44
...@@ -8,9 +8,7 @@ PaddlePaddle提供数个预编译的二进制来进行安装，包括Docker镜 ...@@ -8,9 +8,7 @@ PaddlePaddle提供数个预编译的二进制来进行安装，包括Docker镜
.. toctree:: .. toctree::
:maxdepth: 1 :maxdepth: 1
:glob:
使用Jumbo安装(对内) <../build/internal/install_from_jumbo.rst>
install/docker_install.rst install/docker_install.rst
install/ubuntu_install.rst install/ubuntu_install.rst
...@@ -25,8 +23,5 @@ PaddlePaddle提供数个预编译的二进制来进行安装，包括Docker镜 ...@@ -25,8 +23,5 @@ PaddlePaddle提供数个预编译的二进制来进行安装，包括Docker镜
.. toctree:: .. toctree::
:maxdepth: 1 :maxdepth: 1
:glob:
源码下载(对内) <../build/internal/download_paddle_source_zh_cn.rst>
从源码编译安装(对内) <../build/internal/build_from_source_zh_cn.rst>
cmake/index.rst cmake/index.rst
doc_cn/build_and_install/install/docker_install.rst
浏览文件 @ 021b3a44
安装PaddlePaddle的Docker镜像 安装PaddlePaddle的Docker镜像
============================ ============================
PaddlePaddle提供了Docker的使用镜像。PaddlePaddle推荐使用Docker进行PaddlePaddle的部署和 PaddlePaddle项目提供官方 `Docker <https://www.docker.com/>`_ 镜像。Docker镜像是我们目前唯一官方支持的部署和运行方式。
运行。Docker是一个基于容器的轻量级虚拟环境。具有和宿主机相近的运行效率，并提供
了非常方便的二进制分发手段。
下述内容将分为如下几个类别描述。 下述内容将分为如下几个类别描述。
...@@ -41,7 +39,7 @@ PaddlePaddle提供的Docker镜像版本 ...@@ -41,7 +39,7 @@ PaddlePaddle提供的Docker镜像版本
* CPU WITHOUT AVX: CPU版本，不支持AVX指令集的CPU也可以运行 * CPU WITHOUT AVX: CPU版本，不支持AVX指令集的CPU也可以运行
* GPU WITHOUT AVX: GPU版本，不需要AVX指令集的CPU也可以运行。 * GPU WITHOUT AVX: GPU版本，不需要AVX指令集的CPU也可以运行。
用户可以选择对应版本的docker image。使用如下脚本可以确定本机的CPU知否支持 :code:`AVX` 指令集\: 用户可以选择对应版本的docker image。使用如下脚本可以确定本机的CPU是否支持 :code:`AVX` 指令集\:
.. code-block:: bash .. code-block:: bash
...@@ -67,7 +65,7 @@ mac osx或者是windows机器，请参考 ...@@ -67,7 +65,7 @@ mac osx或者是windows机器，请参考
.. code-block:: bash .. code-block:: bash
$ docker run -it paddledev/paddlepaddle:cpu-latest $ docker run -it paddledev/paddle:cpu-latest
即可启动和进入PaddlePaddle的container。如果运行GPU版本的PaddlePaddle，则需要先将 即可启动和进入PaddlePaddle的container。如果运行GPU版本的PaddlePaddle，则需要先将
cuda相关的Driver和设备映射进container中，脚本类似于 cuda相关的Driver和设备映射进container中，脚本类似于
...@@ -76,7 +74,7 @@ cuda相关的Driver和设备映射进container中，脚本类似于 ...@@ -76,7 +74,7 @@ cuda相关的Driver和设备映射进container中，脚本类似于
$ export CUDA_SO="$(\ls /usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')" $ export CUDA_SO="$(\ls /usr/lib64/libcuda* | xargs -I{} echo '-v {}:{}') $(\ls /usr/lib64/libnvidia* | xargs -I{} echo '-v {}:{}')"
$ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}') $ export DEVICES=$(\ls /dev/nvidia* | xargs -I{} echo '--device {}:{}')
$ docker run ${CUDA_SO} ${DEVICES} -it paddledev/paddlepaddle:latest-gpu $ docker run ${CUDA_SO} ${DEVICES} -it paddledev/paddle:gpu-latest
进入Docker container后，运行 :code:`paddle version` 即可打印出PaddlePaddle的版本和构建 进入Docker container后，运行 :code:`paddle version` 即可打印出PaddlePaddle的版本和构建
信息。安装完成的PaddlePaddle主体包括三个部分， :code:`paddle` 脚本， python的 信息。安装完成的PaddlePaddle主体包括三个部分， :code:`paddle` 脚本， python的
......
doc_cn/build_and_install/install/ubuntu_install.rst
浏览文件 @ 021b3a44
使用deb包在Ubuntu上安装PaddlePaddle Ubuntu部署PaddlePaddle
=================================== ===================================
PaddlePaddle目前支持使用deb包安装。Paddle的 :code:`deb` 安装包在ubuntu 14.04中正确，但理论上支持其他的 debian 发行版。 PaddlePaddle提供了ubuntu 14.04 deb安装包。
安装
------
PaddlePaddle的ubuntu安装包分为四个版本，他们是 cpu、gpu、cpu-noavx、gpu-noavx 四个版本。其中 noavx 用于不支持AVX指令集的cpu。安装包的下载地址是\: https://github.com/baidu/Paddle/releases/ 安装包的下载地址是\: https://github.com/PaddlePaddle/Paddle/releases
它包含四个版本\:
用户需要先将PaddlePaddle安装包下载到本地，然后执行如下 :code:`gdebi` 命令即可完成安装。 * cpu版本: 支持主流x86处理器平台, 使用了avx指令集。
.. code-block:: shell * cpu-noavx版本：支持主流x86处理器平台，没有使用avx指令集。
* gpu版本：支持主流x86处理器平台，支持nvidia cuda平台，使用了avx指令集。
gdebi paddle-*-cpu*.deb * gpu-noavx版本：支持主流x86处理器平台，支持nvidia cuda平台，没有使用avx指令集。
如果 :code:`gdebi` 没有安装,则需要使用 :code:`sudo apt-get install gdebi`, 来安装 :code:`gdebi` 。 下载完相关安装包后，执行:
.. code-block:: shell
或者使用下面一条命令安装. sudo apt-get install gdebi
gdebi paddle-*-cpu.deb
或者:
.. code-block:: shell .. code-block:: shell
dpkg -i paddle-*-cpu*.deb dpkg -i paddle-*-cpu.deb
apt-get install -f apt-get install -f
在 :code:`dpkg -i` 的时候如果报一些依赖未找到的错误是正常的， 在 :code:`dpkg -i` 的时候如果报一些依赖未找到的错误是正常的，
在 :code:`apt-get install -f` 里会继续安装 PaddlePaddle。 在 :code:`apt-get install -f` 里会继续安装 PaddlePaddle。
需要注意的是，如果使用GPU版本的PaddlePaddle，请安装CUDA 7.5 和CUDNN 5到本地环境中， 安装完成后，可以使用命令 :code:`paddle version` 查看安装后的paddle 版本:
并设置好对应的环境变量(LD_LIBRARY_PATH等等)。
安装完成后,可以使用命令 :code:`paddle version` 查看安装后的paddle 版本。可能的输出为
.. literalinclude:: paddle_version.txt .. literalinclude:: paddle_version.txt
...@@ -39,45 +46,16 @@ PaddlePaddle的ubuntu安装包分为四个版本，他们是 cpu、gpu、cpu-noa ...@@ -39,45 +46,16 @@ PaddlePaddle的ubuntu安装包分为四个版本，他们是 cpu、gpu、cpu-noa
libcudart.so/libcudnn.so找不到 libcudart.so/libcudnn.so找不到
++++++++++++++++++++++++++++++ ++++++++++++++++++++++++++++++
安装完成PaddlePaddle后，运行 :code:`paddle train` 报错\: 安装完成后，运行 :code:`paddle train` 报错\:
.. code-block:: shell .. code-block:: shell
0831 12:36:04.151525 1085 hl_dso_loader.cc:70] Check failed: nullptr != *dso_handle For Gpu version of PaddlePaddle, it couldn't find CUDA library: libcudart.so Please make sure you already specify its path.Note: for training data on Cpu using Gpu version of PaddlePaddle,you must specify libcudart.so via LD_LIBRARY_PATH. 0831 12:36:04.151525 1085 hl_dso_loader.cc:70] Check failed: nullptr != *dso_handle For Gpu version of PaddlePaddle, it couldn't find CUDA library: libcudart.so Please make sure you already specify its path.Note: for training data on Cpu using Gpu version of PaddlePaddle,you must specify libcudart.so via LD_LIBRARY_PATH.
PaddlePaddle使用运行时动态连接CUDA的so，如果在 LD_LIBRARY_PATH里面找不到这些动态 原因是未设置cuda运行时环境变量。 如果使用GPU版本的PaddlePaddle，请安装CUDA 7.5 和CUDNN 5到本地环境中，并设置：
库的话，会报寻找不到这些动态库。
解决方法很简单，就是将这些动态库加到环境变量里面。比较可能的命令如下。
.. code-block:: text
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
CUDA Driver找不到
+++++++++++++++++
运行 :code:`paddle train` 报错\: .. code-block:: shell
.. code-block:: text
F0831 12:39:16.699000 1090 hl_cuda_device.cc:530] Check failed: cudaSuccess == cudaStat (0 vs. 35) Cuda Error: CUDA driver version is insufficient for CUDA runtime version
PaddlePaddle运行时如果没有寻找到cuda的driver，变会报这个错误。解决办法是将cuda
driver添加到LD_LIBRARY_PATH中。比较可能的命令如下。
.. code-block:: text
export LD_LIBRARY_PATH=/usr/lib64:$LD_LIBRARY_PATH
config文件找不到
++++++++++++++++
运行 :code:`paddle train` 得到结果\:
.. code-block:: text
F0831 20:53:07.525789 1302 TrainerMain.cpp:94] Check failed: config != nullptr no valid config export LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/cuda/lib:$LD_LIBRARY_PATH
export PATH=/usr/local/cuda/bin:$PATH
PaddlePaddle在运行时找不到对应的config文件，说明命令行参数 :code:`config` 没有设置。
而这个一般说明PaddlePaddle已经安装完毕了。
\ No newline at end of file
doc_cn/demo/quick_start/index.md
浏览文件 @ 021b3a44
...@@ -32,13 +32,11 @@ ...@@ -32,13 +32,11 @@
## 数据格式准备(Data Preparation) ## 数据格式准备(Data Preparation)
在本问题中，我们使用[Amazon电子产品评论数据](http://jmcauley.ucsd.edu/data/amazon/)， 在本问题中，我们使用[Amazon电子产品评论数据](http://jmcauley.ucsd.edu/data/amazon/)，
将评论分为好评(正样本)和差评(负样本)两类。[源码](https://github.com/baidu/Paddle)的`demo/quick_start`里提供了数据下载脚本 将评论分为好评(正样本)和差评(负样本)两类。[源码](https://github.com/PaddlePaddle/Paddle)的`demo/quick_start`里提供了下载已经预处理数据的脚本（如果想从最原始的数据处理，可以使用脚本 `./demo/quick_start/data/proc_from_raw_data/get_data.sh`）。
和预处理脚本。
```bash ```bash
cd demo/quick_start cd demo/quick_start
./data/get_data.sh ./data/get_data.sh
./preprocess.sh
``` ```
## 数据向模型传送(Transfer Data to Model) ## 数据向模型传送(Transfer Data to Model)
...@@ -143,7 +141,7 @@ PyDataProvider2</a>。 ...@@ -143,7 +141,7 @@ PyDataProvider2</a>。
我们将以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 我们将以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<a href = "../../../doc/layer.html">Layer文档</a>。 连接请参考<a href = "../../../doc/layer.html">Layer文档</a>。
所有配置在[源码](https://github.com/baidu/Paddle)`demo/quick_start`目录，首先列举逻辑回归网络。 所有配置在[源码](https://github.com/PaddlePaddle/Paddle)`demo/quick_start`目录，首先列举逻辑回归网络。
### 逻辑回归模型(Logistic Regression) ### 逻辑回归模型(Logistic Regression)
......
doc_cn/faq/index.rst
浏览文件 @ 021b3a44
...@@ -4,22 +4,18 @@ PaddlePaddle常见问题 ...@@ -4,22 +4,18 @@ PaddlePaddle常见问题
.. contents:: .. contents::
1. 如何减少PaddlePaddle的内存占用 1. 如何减少内存占用
--------------------------------- ---------------------------------
神经网络的训练本身是一个非常消耗内存和显存的工作。经常会消耗数十G的内存和数G的显存。 神经网络的训练本身是一个非常消耗内存和显存的工作，经常会消耗数10GB的内存和数GB的显存。
PaddlePaddle的内存占用主要分为如下几个方面\: PaddlePaddle的内存占用主要分为如下几个方面\:
* DataProvider缓冲池内存 (只针对内存) * DataProvider缓冲池内存（只针对内存）
* 神经元激活内存 （针对内存和显存） * 神经元激活内存（针对内存和显存）
* 参数内存 (针对内存和显存) * 参数内存 （针对内存和显存）
* 其他内存杂项 * 其他内存杂项
这其中，其他内存杂项是指PaddlePaddle本身所用的一些内存，包括字符串分配，临时变量等等， 其中，其他内存杂项是指PaddlePaddle本身所用的一些内存，包括字符串分配，临时变量等等，暂不考虑在内。
这些内存就不考虑如何缩减了。
其他的内存的减少方法依次为
减少DataProvider缓冲池内存 减少DataProvider缓冲池内存
++++++++++++++++++++++++++ ++++++++++++++++++++++++++
...@@ -39,28 +35,28 @@ PyDataProvider使用的是异步加载，同时在内存里直接随即选取数 ...@@ -39,28 +35,28 @@ PyDataProvider使用的是异步加载，同时在内存里直接随即选取数
.. literalinclude:: reduce_min_pool_size.py .. literalinclude:: reduce_min_pool_size.py
这样做可以极大的减少内存占用，并且可能会加速训练过程。 详细文档参考 `这里 这样做可以极大的减少内存占用，并且可能会加速训练过程，详细文档参考 `这里
<../ui/data_provider/pydataprovider2.html#provider>`_ 。 <../ui/data_provider/pydataprovider2.html#provider>`_ 。
神经元激活内存 神经元激活内存
++++++++++++++ ++++++++++++++
神经网络在训练的时候，会对每一个激活暂存一些数据，包括激活，參差等等。 神经网络在训练的时候，会对每一个激活暂存一些数据，如神经元激活值等。
在反向传递的时候，这些数据会被用来更新参数。这些数据使用的内存主要和两个参数有关系， 在反向传递的时候，这些数据会被用来更新参数。这些数据使用的内存主要和两个参数有关系，
一是batch size，另一个是每条序列(Sequence)长度。所以，其实也是和每个mini-batch中包含 一是batch size，另一个是每条序列(Sequence)长度。所以，其实也是和每个mini-batch中包含
的时间步信息成正比。 的时间步信息成正比。
所以，做法可以有两种。他们是 所以做法可以有两种：
* 减小batch size。 即在网络配置中 :code:`settings(batch_size=1000)` 设置成一个小一些的值。但是batch size本身是神经网络的超参数，减小batch size可能会对训练结果产生影响。 * 减小batch size。 即在网络配置中 :code:`settings(batch_size=1000)` 设置成一个小一些的值。但是batch size本身是神经网络的超参数，减小batch size可能会对训练结果产生影响。
* 减小序列的长度，或者直接扔掉非常长的序列。比如，一个数据集大部分序列长度是100-200, * 减小序列的长度，或者直接扔掉非常长的序列。比如，一个数据集大部分序列长度是100-200,
但是突然有一个10000长的序列，就很容易导致内存超限。特别是在LSTM等RNN中。 但是突然有一个10000长的序列，就很容易导致内存超限，特别是在LSTM等RNN中。
参数内存 参数内存
++++++++ ++++++++
PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需要使用不同大小的内存。 PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需要使用不同大小的内存。
例如如果使用 :code:`adadelta` 算法，则需要使用参数规模大约5倍的内存。 如果参数保存下来的 例如使用 :code:`adadelta` 算法，则需要使用等于权重参数规模大约5倍的内存。举例，如果参数保存下来的模型目录
文件为 :code:`100M`， 那么该优化算法至少需要 :code:`500M` 的内存。 文件为 :code:`100M`， 那么该优化算法至少需要 :code:`500M` 的内存。
可以考虑使用一些优化算法，例如 :code:`momentum`。 可以考虑使用一些优化算法，例如 :code:`momentum`。
...@@ -68,11 +64,11 @@ PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需 ...@@ -68,11 +64,11 @@ PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需
2. 如何加速PaddlePaddle的训练速度 2. 如何加速PaddlePaddle的训练速度
--------------------------------- ---------------------------------
PaddlePaddle是神经网络训练平台，加速PaddlePaddle训练有如下几个方面\： 加速PaddlePaddle训练可以考虑从以下几个方面\：
* 减少数据载入的耗时 * 减少数据载入的耗时
* 加速训练速度 * 加速训练速度
* 利用更多的计算资源 * 利用分布式训练驾驭更多的计算资源
减少数据载入的耗时 减少数据载入的耗时
++++++++++++++++++ ++++++++++++++++++
...@@ -108,25 +104,20 @@ PaddlePaddle支持Sparse的训练，sparse训练需要训练特征是 :code:`spa ...@@ -108,25 +104,20 @@ PaddlePaddle支持Sparse的训练，sparse训练需要训练特征是 :code:`spa
利用更多的计算资源可以分为一下几个方式来进行\: 利用更多的计算资源可以分为一下几个方式来进行\:
* 单机CPU训练 * 单机CPU训练
* 使用多线程训练。设置命令行参数 :code:`trainer_count`，即可以设置参与训练的线程数量。使用方法为 :code:`paddle train --trainer_count=4` * 使用多线程训练。设置命令行参数 :code:`trainer_count`。
* 单机GPU训练 * 单机GPU训练
* 使用显卡训练。设置命令行参数 :code:`use_gpu`。 使用方法为 :code:`paddle train --use_gpu=true` * 使用显卡训练。设置命令行参数 :code:`use_gpu`。
* 使用多块显卡训练。设置命令行参数 :code:`use_gpu` 和 :code:`trainer_count`。使用 :code:`--use_gpu=True` 开启GPU训练，使用 :code:`trainer_count` 指定显卡数量。使用方法为 :code:`paddle train --use_gpu=true --trainer_count=4` * 使用多块显卡训练。设置命令行参数 :code:`use_gpu` 和 :code:`trainer_count` 。
* 多机训练 * 多机训练
* 使用多机训练的方法也比较简单，需要先在每个节点启动 :code:`paddle pserver`，在使用 :code:`paddle train --pservers=192.168.100.1,192.168.100.2` 来指定每个pserver的ip地址 * 具体的多机训练方法参考 `多机训练文档 <../ui/data_provider/pydataprovider2.html#provider>`_ 。
* 具体的多机训练方法参考 `多机训练 <TBD>`_ 文档。
3. 遇到“非法指令”或者是“illegal instruction” 3. 遇到“非法指令”或者是“illegal instruction”
-------------------------------------------- --------------------------------------------
paddle在进行计算的时候为了提升计算性能，使用了avx指令。部分老的cpu型号无法支持这样的指令。通常来说执行下grep avx /proc/cpuinfo看看是否有输出即可知道是否支持。（另：用此方法部分虚拟机可能检测到支持avx指令但是实际运行会挂掉，请当成是不支持，看下面的解决方案） PaddlePaddle使用avx SIMD指令提高cpu执行效率，因此错误的使用二进制发行版可能会导致这种错误，请选择正确的版本。
解决办法是\:
* 使用 NO_AVX的 `安装包 <../build_and_install/index.html>`_ 或者 `Docker image <../build_and_install/install/docker_install.html>`_
* 或者，使用 :code:`-DWITH_AVX=OFF` 重新编译PaddlePaddle。
4. 如何选择SGD算法的学习率 4. 如何选择SGD算法的学习率
-------------------------- --------------------------
...@@ -158,7 +149,7 @@ paddle在进行计算的时候为了提升计算性能，使用了avx指令。 ...@@ -158,7 +149,7 @@ paddle在进行计算的时候为了提升计算性能，使用了avx指令。
6. 如何共享参数 6. 如何共享参数
--------------- ---------------
PaddlePaddle的参数使用名字 :code:`name` 作为参数的ID，相同名字的参数，会共享参数。设置参数的名字，可以使用 :code:`ParamAttr(name="YOUR_PARAM_NAME")` 来设置。更方便的设置方式，是想要共享的参数使用同样的 :code:`ParamAttr` 对象。 PaddlePaddle的参数使用名字 :code:`name` 作为参数的ID，相同名字的参数，会共享参数。设置参数的名字，可以使用 :code:`ParamAttr(name="YOUR_PARAM_NAME")` 来设置。更方便的设置方式，是使得要共享的参数使用同样的 :code:`ParamAttr` 对象。
简单的全连接网络，参数共享的配置示例为\: 简单的全连接网络，参数共享的配置示例为\:
...@@ -208,9 +199,18 @@ PaddlePaddle的参数使用名字 :code:`name` 作为参数的ID，相同名字 ...@@ -208,9 +199,18 @@ PaddlePaddle的参数使用名字 :code:`name` 作为参数的ID，相同名字
paddle package is already in your PYTHONPATH. But unittest need a clean environment. paddle package is already in your PYTHONPATH. But unittest need a clean environment.
Please uninstall paddle package before start unittest. Try to 'pip uninstall paddle'. Please uninstall paddle package before start unittest. Try to 'pip uninstall paddle'.
解决办法是：卸载paddle包 :code:`pip uninstall paddle`。 解决办法是：
* 卸载PaddlePaddle包 :code:`pip uninstall paddle`, 清理掉老旧的PaddlePaddle安装包，使得单元测试有一个干净的环境。如果PaddlePaddle包已经在python的site-packages里面，单元测试会引用site-packages里面的python包，而不是源码目录里 :code:`/python` 目录下的python包。同时，即便设置 :code:`PYTHONPATH` 到 :code:`/python` 也没用，因为python的搜索路径是优先已经安装的python包。
9. CMake源码编译, 找到的PythonLibs和PythonInterp版本不一致
----------------------------------------------------------
这是目前CMake寻找Python的逻辑存在缺陷，如果系统安装了多个Python版本，CMake找到的Python库和Python解释器版本可能有不一致现象，导致编译PaddlePaddle失败。正确的解决方法是，
用户强制指定特定的Python版本，具体操作如下：
.. code-block:: bash
原因是：单元测试使用了一个旧版本的python包，而没有测试到代码中实际修改的python包。即单元测试需要一个干净的环境： cmake .. -DPYTHON_EXECUTABLE=<exc_path> -DPYTHON_LIBRARY=<lib_path> -DPYTHON_INCLUDE_DIR=<inc_path>
* 如果paddle包已经在python的site-packages里面了，那么单元测试时使用的paddle包，就是site-packages里面的python包，而不是源码目录里 :code:`/python` 目录下的python包。 用户需要指定本机上Python的路径：``<exc_path>``, ``<lib_path>``, ``<inc_path>``
* 即便设置了 :code:`PYTHONPATH` 到 :code:`/python` 也没用，因为python的搜索路径是优先已经安装的python包。
\ No newline at end of file
doc_cn/howto/build_docker_image.rst 0 → 100644
浏览文件 @ 021b3a44
构建PaddlePaddle的Docker Image
==============================
PaddlePaddle的Docker Image构建源码放置在 ``${源码根目录}/paddle/scripts/docker/`` 目录下。该目录有三类文件：
- Dockerfile：Docker Image的描述文件，包括构建步骤、各种参数和维护人员等。
- 一共维护了12个Dockerfile，Dockerfile.m4是它们的模板。
- PaddlePaddle中所有的Image都基于ubuntu 14.04。
- build.sh：Docker Image的构建脚本，使用方式见下一小节。
- generate.sh：通过Dockerfile.m4模板生成不同的Dockerfile。
使用脚本构建Docker Image
------------------------
进入源码目录，执行 ``docker build`` 命令，即可在本地编译出PaddlePaddle的镜像。简单的使用样例为
.. code-block:: bash
cd ${源码根目录}/paddle/scripts/docker/
docker build --build-arg LOWEST_DL_SPEED=50K \
--build-arg WITH_GPU=ON \
--tag paddle_gpu:latest .
其中，``--build-arg`` 传入的配置参数包括:
- LOWEST\_DL\_SPEED\: 在多线程下载过程中，设置下载线程的最低速度。
- 默认单位是Bytes，但可以传入10K、10M、或10G等这样的单位。
- 如果小于这个速度，那么这个线程将会关闭。当所有的线程都关闭了，那么下载进程将会重启。
- WITH\_GPU\: ON or OFF，是否开启GPU功能。注意，
- **编译** PaddlePaddle的GPU版本 **不一定** 要在具有GPU的机器上进行。
- **运行** PaddlePaddle的GPU版本 **一定** 要在具有GPU的机器上运行。
注意：所有Image的构建在Docker 1.12版本测试通过, 低于1.12的版本并没有测试。原因是旧版本可能缺乏 ``--build-arg`` 参数，从而不能在运行编译命令的时候接受参数。
doc_cn/index.rst
浏览文件 @ 021b3a44
...@@ -16,6 +16,7 @@ PaddlePaddle文档 ...@@ -16,6 +16,7 @@ PaddlePaddle文档
-------- --------
* `新写Layer <../doc/dev/new_layer/index.html>`_ * `新写Layer <../doc/dev/new_layer/index.html>`_
* `如何贡献文档 <howto/how_to_write_docs/index.html>`_ * `如何贡献文档 <howto/how_to_write_docs/index.html>`_
* `如何构建Docker Image <howto/build_docker_image.html>`_
算法教程 算法教程
-------- --------
......
doc_cn/introduction/index.md 已删除 100644 → 0
浏览文件 @ 4f1bf30d
# 简介
PaddlePaddle 是起源于百度的开源深度学习平台。它是简单易用的：你可以通过简单的十数行配置搭建经典的神经网络模型；它也是高效强大的：PaddlePaddle可以支撑复杂集群环境下超大模型的训练，令你受益于深度学习的前沿成果。在百度内部，已经有大量产品线使用了基于PaddlePaddle的深度学习技术。
这份简短的介绍将像你展示如何利用PaddlePaddle解决一个经典的学习问题。
## 1. 一个经典的任务
让我们从一个基础问题开始：<a href="https://www.baidu.com/s?wd=单变量线性回归">单变量的线性回归</a>。问题假定观测到了一批二维空间上的点`(x, y) `，并且已知 `x` 和 `y` 之间存在着某种线性关系，我们的目标是通过观测数据还原这个线性关系。作为一个简单基础的模型，线性回归却有着广泛的应用场景。比如可以想象一个资产定价的简化场景，其中 `x` 对应于房屋的大小，`y` 对应于房屋价格。我们可以通过观察市场上房屋的情况获得二者之间的关系，从而为新房屋的定价提供参考。
## 2. 准备数据
假设变量 `X` 和 `Y` 的真实关系为： `Y = 2X + 0.3`，这里展示如何使用观测数据还原这一线性关系。如下Python代码将随机产生2000个观测点，它们将被用作PaddlePaddle的输入。产生PaddlePaddle的输入数据和写一段普通的Python脚本几乎一样，你唯一需要增加的就是定义输入数据的类型。
```python
# -*- coding:utf-8 -*-
# dataprovider.py
from paddle.trainer.PyDataProvider2 import *
import random
# 定义输入数据的类型: 2个浮点数
@provider(input_types=[dense_vector(1), dense_vector(1)],use_seq=False)
def process(settings, input_file):
for i in xrange(2000):
x = random.random()
yield [x], [2*x+0.3]
```
## 3. 训练模型
为了还原 `Y = 2X + 0.3`，我们先从一条随机的直线 `Y' = wX + b` 开始，然后利用观测数据调整 `w` 和 `b` 使得 `Y'` 和 `Y` 的差距不断减小，最终趋于相同。这个过程就是模型的训练过程，而 `w` 和 `b` 就是模型的参数，即我们的训练目标。
在PaddlePaddle里，该模型的网络配置如下。
```python
# -*- coding:utf-8 -*-
# trainer_config.py
from paddle.trainer_config_helpers import *
# 1. 定义数据来源，调用上面的process函数获得观测数据
data_file = 'empty.list'
with open(data_file, 'w') as f: f.writelines(' ')
define_py_data_sources2(train_list=data_file, test_list=None,
module='dataprovider', obj='process',args={})
# 2. 学习算法。控制如何改变模型参数 w 和 b
settings(batch_size=12, learning_rate=1e-3, learning_method=MomentumOptimizer())
# 3. 神经网络配置
x = data_layer(name='x', size=1)
y = data_layer(name='y', size=1)
# 线性计算单元: y_predict = wx + b
y_predict = fc_layer(input=x, param_attr=ParamAttr(name='w'), size=1, act=LinearActivation(), bias_attr=ParamAttr(name='b'))
# 损失计算，度量 y_predict 和真实 y 之间的差距
cost = regression_cost(input=y_predict, label=y)
outputs(cost)
```
这段简短的配置展示了PaddlePaddle的基本用法：
- 首先，第一部分定义了数据输入。一般情况下，PaddlePaddle先从一个文件列表里获得数据文件地址，然后交给用户自定义的函数（例如上面的`process`函数）进行读入和预处理从而得到真实输入。本文中由于输入数据是随机生成的不需要读输入文件，所以放一个空列表（`empty.list`）即可。
- 第二部分主要是选择学习算法，它定义了模型参数如何改变。PaddlePaddle提供了很多优秀的学习算法，但这里使用一个简单的基于momentum的算法就足够了，它每次读取12个数据进行计算和模型更新。
- 最后一部分是神经网络的配置。由于PaddlePaddle已经实现了丰富的网络单元（Layer），所以很多时候你需要做的只是声明正确的网络单元并把它们拼接起来。这里使用了三种网络单元：
- **数据层**：数据层 `data_layer` 是神经网络的入口，它读入数据并将它们传输到下游的其它单元。这里数据层有两个，分别对应于变量 `X` 和 `Y`。
- **全连接层**：全连接层 `fc_layer` 是基础的计算单元，这里利用它建模变量之间的线性关系。计算单元是神经网络的核心，PaddlePaddle支持大量的计算单元和任意深度的网络连接，从而可以挖掘复杂的数据关系。
- **回归损失层**：回归损失层 `regression_cost`是众多损失函数层的一种，它们在训练过程作为网络的出口，用来计算模型的表现，并指导模型参数的改变。
这样定义了网络结构并保存为`trainer_config.py`之后，运行训练命令即可：
```
paddle train --config=trainer_config.py --save_dir=./output --num_passes=30
```
PaddlePaddle将在观测数据集上迭代训练30轮，并将每轮的模型结果存放在 `./output` 路径下。从输出日志可以看到，随着轮数增加损失函数的输出在不断的减小，这意味着模型在不断的改进，直到逼近真实解：` Y = 2X + 0.3 `
## 4. 模型检验
训练完成后，我们希望能够检验模型的好坏。一种常用的做法是用模型对另外一组数据进行预测，然后评价预测的效果。但在这个例子中，由于已经知道了真实答案，我们可以直接观察模型的参数是否符合预期来进行检验。
PaddlePaddle将每个模型参数作为一个numpy数组单独存为一个文件，所以可以利用如下方法读取模型的参数。
```python
import numpy as np
import os
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)
print 'w=%.6f, b=%.6f' % (load('output/pass-00029/w'), load('output/pass-00029/b'))
# w=1.999743, b=0.300137
```
<center> ![](./parameters.png) </center>
从图中可以看到，虽然 `w` 和 `b` 都使用随机值初始化，但在起初的几轮训练中它们都在快速逼近真实值，并且后续仍在不断改进，使得最终得到的模型几乎与真实模型重合。
这样，我们就完成了对单变量线性回归问题的解决：将数据输入PaddlePaddle，训练模型，最后验证结果。
## 5. 推荐后续阅读
- <a href="../build_and_install/index.html">安装/编译</a>：PaddlePaddle的安装与编译文档。
- <a href="../demo/quick_start/index.html">快速入门 </a>：使用商品评论分类任务，系统性的介绍如何一步步改进，最终得到产品级的深度模型。
- <a href="../demo/index.html">示例</a>：各种实用案例，涵盖图像、文本、推荐等多个领域。
doc_cn/introduction/index.rst 0 → 100644
浏览文件 @ 021b3a44
简介
====
PaddlePaddle是源于百度的一个深度学习平台。这份简短的介绍将向你展示如何利用PaddlePaddle来解决一个经典的线性回归问题。
1. 一个经典的任务
-----------------
我们展示如何用PaddlePaddle解决 `单变量的线性回归 <https://www.baidu.com/s?wd=单变量线性回归>`_ 问题。线性回归的输入是一批点 `(x, y)` ，其中 `y = wx + b + ε`， 而 ε 是一个符合高斯分布的随机变量。线性回归的输出是从这批点估计出来的参数 `w` 和 `b` 。
一个例子是房产估值。我们假设房产的价格（y）是其大小（x）的一个线性函数，那么我们可以通过收集市场上房子的大小和价格，用来估计线性函数的参数w 和 b。
2. 准备数据
-----------
假设变量 `x` 和 `y` 的真实关系为： `y = 2x + 0.3 + ε`，这里展示如何使用观测数据来拟合这一线性关系。首先，Python代码将随机产生2000个观测点，作为线性回归的输入。下面脚本符合PaddlePaddle期待的读取数据的Python程序的模式。
.. code-block:: python
# dataprovider.py
from paddle.trainer.PyDataProvider2 import *
import random
# 定义输入数据的类型: 2个浮点数
@provider(input_types=[dense_vector(1), dense_vector(1)],use_seq=False)
def process(settings, input_file):
for i in xrange(2000):
x = random.random()
yield [x], [2*x+0.3]
3. 训练模型
-----------
为了还原 `y = 2x + 0.3`，我们先从一条随机的直线 `y' = wx + b` 开始，然后利用观测数据调整 `w` 和 `b` 使得 `y'` 和 `y` 的差距不断减小，最终趋于接近。这个过程就是模型的训练过程，而 `w` 和 `b` 就是模型的参数，即我们的训练目标。
在PaddlePaddle里，该模型的网络配置如下。
.. code-block:: python
# trainer_config.py
from paddle.trainer_config_helpers import *
# 1. 定义数据来源，调用上面的process函数获得观测数据
data_file = 'empty.list'
with open(data_file, 'w') as f: f.writelines(' ')
define_py_data_sources2(train_list=data_file, test_list=None,
module='dataprovider', obj='process',args={})
# 2. 学习算法。控制如何改变模型参数 w 和 b
settings(batch_size=12, learning_rate=1e-3, learning_method=MomentumOptimizer())
# 3. 神经网络配置
x = data_layer(name='x', size=1)
y = data_layer(name='y', size=1)
# 线性计算网络层: ȳ = wx + b
ȳ = fc_layer(input=x, param_attr=ParamAttr(name='w'), size=1, act=LinearActivation(), bias_attr=ParamAttr(name='b'))
# 计算误差函数，即 ȳ 和真实 y 之间的距离
cost = regression_cost(input= ȳ, label=y)
outputs(cost)
这段简短的配置展示了PaddlePaddle的基本用法：
- 第一部分定义了数据输入。一般情况下，PaddlePaddle先从一个文件列表里获得数据文件地址，然后交给用户自定义的函数（例如上面的 `process`函数）进行读入和预处理从而得到真实输入。本文中由于输入数据是随机生成的不需要读输入文件，所以放一个空列表（`empty.list`）即可。
- 第二部分主要是选择学习算法，它定义了模型参数改变的规则。PaddlePaddle提供了很多优秀的学习算法，这里使用一个基于momentum的随机梯度下降(SGD)算法，该算法每批量(batch)读取12个采样数据进行随机梯度计算来更新更新。
- 最后一部分是神经网络的配置。由于PaddlePaddle已经实现了丰富的网络层，所以很多时候你需要做的只是定义正确的网络层并把它们连接起来。这里使用了三种网络单元：
- **数据层**：数据层 `data_layer` 是神经网络的入口，它读入数据并将它们传输到接下来的网络层。这里数据层有两个，分别对应于变量 `x` 和 `y`。
- **全连接层**：全连接层 `fc_layer` 是基础的计算单元，这里利用它建模变量之间的线性关系。计算单元是神经网络的核心，PaddlePaddle支持大量的计算单元和任意深度的网络连接，从而可以拟合任意的函数来学习复杂的数据关系。
- **回归误差代价层**：回归误差代价层 `regression_cost` 是众多误差代价函数层的一种，它们在训练过程作为网络的出口，用来计算模型的误差，是模型参数优化的目标函数。
定义了网络结构并保存为 `trainer_config.py` 之后，运行以下训练命令：
.. code-block:: bash
paddle train --config=trainer_config.py --save_dir=./output --num_passes=30
PaddlePaddle将在观测数据集上迭代训练30轮，并将每轮的模型结果存放在 `./output` 路径下。从输出日志可以看到，随着轮数增加误差代价函数的输出在不断的减小，这意味着模型在训练数据上不断的改进，直到逼近真实解：` y = 2x + 0.3 `
4. 模型检验
-----------
训练完成后，我们希望能够检验模型的好坏。一种常用的做法是用学习的模型对另外一组测试数据进行预测，评价预测的效果。在这个例子中，由于已经知道了真实答案，我们可以直接观察模型的参数是否符合预期来进行检验。
PaddlePaddle将每个模型参数作为一个numpy数组单独存为一个文件，所以可以利用如下方法读取模型的参数。
.. code-block:: python
import numpy as np
import os
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)
print 'w=%.6f, b=%.6f' % (load('output/pass-00029/w'), load('output/pass-00029/b'))
# w=1.999743, b=0.300137
.. image:: ./parameters.png
:align: center
:scale: 80 %
从图中可以看到，虽然 `w` 和 `b` 都使用随机值初始化，但在起初的几轮训练中它们都在快速逼近真实值，并且后续仍在不断改进，使得最终得到的模型几乎与真实模型一致。
这样，我们用PaddlePaddle解决了单变量线性回归问题， 包括数据输入、模型训练和最后的结果验证。
5. 推荐后续阅读
---------------
- `安装/编译 <../build_and_install/index.html>`_ ：PaddlePaddle的安装与编译文档。
- `快速入门 <../demo/quick_start/index.html>`_ ：使用商品评论分类任务，系统性的介绍如何一步步改进，最终得到产品级的深度模型。
- `示例 <../demo/index.html>`_ ：各种实用案例，涵盖图像、文本、推荐等多个领域。
\ No newline at end of file
doc_cn/ui/cmd/index.rst
浏览文件 @ 021b3a44
PaddlePaddle的命令行参数 命令
======================== ====
安装好PaddlePaddle后，在命令行直接敲击 ``paddle`` 或 ``paddle --help`` 会显示如下一些命令行参数。 安装好PaddlePaddle后，在命令行直接敲击 ``paddle`` 或 ``paddle --help`` 会显示如下一些命令。
* ``train`` Start a paddle_trainer * ``train`` Start a paddle_trainer
启动一个PaddlePaddle训练进程。 ``paddle train`` 可以通过命令行参数 ``-local=true`` 启动一个单机的训练进程；也可以和 ``paddle pserver`` 一起使用启动多机的分布式训练进程。 启动一个PaddlePaddle训练进程。 ``paddle train`` 可以通过命令行参数 ``-local=true`` 启动一个单机的训练进程；也可以和 ``paddle pserver`` 一起使用启动多机的分布式训练进程。
* ``pserver`` Start a paddle_pserver_main * ``pserver`` Start a paddle_pserver_main
在多机分布式训练下启动PaddlePaddle的parameter server进程。 在多机分布式训练下启动PaddlePaddle的parameter server进程。
* ``version`` Print paddle version * ``version`` Print paddle version
用于打印当前PaddlePaddle的版本和编译选项相关信息。 用于打印当前PaddlePaddle的版本和编译选项相关信息。常见的输出格式如下：1）第一行说明了PaddlePaddle的版本信息；2）第二行开始说明了一些主要的编译选项，具体意义可以参考 `编译参数选项文件 <../../build_and_install/cmake/compile_options.html>`_ 。
.. literalinclude:: paddle_version.txt
* ``merge_model`` Start a paddle_merge_model * ``merge_model`` Start a paddle_merge_model
用于将PaddlePaddle的模型参数文件和模型配置文件打包成一个文件，方便做部署分发。 用于将PaddlePaddle的模型参数文件和模型配置文件打包成一个文件，方便做部署分发。
* ``dump_config`` Dump the trainer config as proto string * ``dump_config`` Dump the trainer config as proto string
用于将PaddlePaddle的模型配置文件以proto string的格式打印出来。 用于将PaddlePaddle的模型配置文件以proto string的格式打印出来。
* ``make_diagram`` * ``make_diagram``
使用graphviz对PaddlePaddle的模型配置文件进行绘制。 使用graphviz对PaddlePaddle的模型配置文件进行绘制。
\ No newline at end of file
更详细的介绍请参考各命令行参数文档。
.. toctree::
:glob:
paddle_train.rst
paddle_pserver.rst
paddle_version.rst
merge_model.rst
dump_config.rst
make_diagram.rst
doc_cn/ui/cmd/paddle_version.rst 已删除 100644 → 0
浏览文件 @ 4f1bf30d
paddle version的命令行参数
==========================
paddle version用于打印当前的版本信息和相关编译选项。常见的输出格式如下。第一行说明了PaddlePaddle的版本信息，后面跟着一些主要的编译选项。编译选项的具体意义可以参考
`编译参数选项文件 <../../build_and_install/cmake/compile_options.html>`_
.. literalinclude:: paddle_version.txt
doc_cn/ui/index.rst
浏览文件 @ 021b3a44
...@@ -11,21 +11,23 @@ ...@@ -11,21 +11,23 @@
data_provider/index.rst data_provider/index.rst
命令行参数 命令及命令行参数
========== ================
.. toctree:: .. toctree::
:maxdepth: 1
cmd/index.rst cmd/index.rst
* `参数用例 <../../doc/ui/cmd_argument/use_case.html>`_
* `参数分类 <../../doc/ui/cmd_argument/argument_outline.html>`_ * `参数分类 <../../doc/ui/cmd_argument/argument_outline.html>`_
* `参数描述 <../../doc/ui/cmd_argument/detail_introduction.html>`_ * `参数描述 <../../doc/ui/cmd_argument/detail_introduction.html>`_
* `参数用例 <../../doc/ui/cmd_argument/use_case.html>`_
预测 预测
==== ====
.. toctree:: .. toctree::
:maxdepth: 1
predict/swig_py_paddle.rst predict/swig_py_paddle.rst
doc_cn/ui/predict/swig_py_paddle.rst
浏览文件 @ 021b3a44
PaddlePaddle的Python预测接口 基于Python的预测
================================== ================
PaddlePaddle目前使用Swig对其常用的预测接口进行了封装，使在Python环境下的预测接口更加简单。 预测流程
在Python环境下预测结果，主要分为以下几个步骤。 --------
* 读入解析训练配置 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。
* 构造GradientMachine
* 准备数据
* 预测
典型的预测代码如下，使用mnist手写识别作为样例, 完整代码见 基于Python的模型预测，主要包括以下五个步骤。
:code:`src_root/doc/ui/predict/predict_sample.py` 。
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:: ../../../doc/ui/predict/predict_sample.py .. literalinclude:: ../../../doc/ui/predict/predict_sample.py
:language: python :language: python
:lines: 15-18,90-100,101-104 :lines: 15-18,121-136
主要的软件包为py_paddle.swig_paddle，这个软件包文档相对完善。可以使用python的
:code:`help()` 函数查询文档。主要步骤为: 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 。
* 在程序开始阶段，使用 :code:`swig_paddle.initPaddle()` 传入命令行参数初始化
PaddlePaddle。详细的命令行参数请参考
`命令行参数 <../cmd_argument/detail_introduction.html>`_ 。
* 接下来使用 :code:`parse_config()` 解析训练时的配置文件。这里要注意预测数据通常
不包含label, 而且预测网络通常直接输出最后一层的结果而不是像训练时一样以cost
layer作为输出，所以用于预测的配置文件要做相应的修改。
* 使用 :code:`swig_paddle.GradientMachine.createFromConfigproto()` 根据上一步解
析好的配置创建神经网络。
* 创建一个 :code:`DataProviderConverter` 对象converter。
- swig_paddle接受的原始数据是C++的Matrix，也就是直接写内存的float数组。
这个接口并不用户友好。所以，我们提供了一个工具类DataProviderConverter。
这个工具类接收和PyDataProvider2一样的输入数据，详情请参考
`PyDataProvider2文档 <../../../doc/ui/data_provider/pydataprovider2.html>`_ 。
* 最后使用 :code:`forwardTest()` 直接提取出神经网络Output层的输出结果。典型的输出结果为\:
.. code-block:: text .. code-block:: text
[{'id': None, 'value': array([[ 5.53018653e-09, 1.12194102e-05, 1.96644767e-09, [{'id': None, 'value': array(
[[ 5.53018653e-09, 1.12194102e-05, 1.96644767e-09,
1.43630644e-02, 1.51111044e-13, 9.85625684e-01, 1.43630644e-02, 1.51111044e-13, 9.85625684e-01,
2.08823112e-10, 2.32777140e-08, 2.00186201e-09, 2.08823112e-10, 2.32777140e-08, 2.00186201e-09,
1.15501715e-08], 1.15501715e-08],
...@@ -45,4 +53,4 @@ PaddlePaddle目前使用Swig对其常用的预测接口进行了封装，使在P ...@@ -45,4 +53,4 @@ PaddlePaddle目前使用Swig对其常用的预测接口进行了封装，使在P
2.70634608e-08, 3.48565123e-08, 5.25639710e-09, 2.70634608e-08, 3.48565123e-08, 5.25639710e-09,
4.48684503e-08]], dtype=float32)}] 4.48684503e-08]], dtype=float32)}]
其中，value即为softmax层的输出。由于数据是两条，所以输出的value包含两个向量 。
paddle/api/Arguments.cpp
浏览文件 @ 021b3a44
...@@ -27,11 +27,6 @@ Arguments* Arguments::createArguments(size_t slotNum) { ...@@ -27,11 +27,6 @@ Arguments* Arguments::createArguments(size_t slotNum) {
void Arguments::resize(size_t slotNum) { m->outputs.resize(slotNum); } void Arguments::resize(size_t slotNum) { m->outputs.resize(slotNum); }
Matrix* Arguments::getSlotValue(size_t idx) const throw(RangeError) {
auto& a = m->getArg(idx);
return Matrix::createByPaddleMatrixPtr(&a.value);
}
Arguments::Arguments() : m(new ArgumentsPrivate()) {} Arguments::Arguments() : m(new ArgumentsPrivate()) {}
Arguments::~Arguments() { delete m; } Arguments::~Arguments() { delete m; }
...@@ -43,6 +38,16 @@ Arguments* Arguments::createByPaddleArgumentVector(void* ptr) { ...@@ -43,6 +38,16 @@ Arguments* Arguments::createByPaddleArgumentVector(void* ptr) {
return args; return args;
} }
Matrix* Arguments::getSlotValue(size_t idx) const throw(RangeError) {
auto& a = m->getArg(idx);
return Matrix::createByPaddleMatrixPtr(&a.value);
}
Matrix* Arguments::getSlotGrad(size_t idx) const throw(RangeError) {
auto& a = m->getArg(idx);
return Matrix::createByPaddleMatrixPtr(&a.grad);
}
IVector* Arguments::getSlotIds(size_t idx) const throw(RangeError) { IVector* Arguments::getSlotIds(size_t idx) const throw(RangeError) {
auto& a = m->getArg(idx); auto& a = m->getArg(idx);
return IVector::createByPaddleVectorPtr(&a.ids); return IVector::createByPaddleVectorPtr(&a.ids);
...@@ -58,6 +63,11 @@ void Arguments::setSlotValue(size_t idx, Matrix* mat) throw(RangeError) { ...@@ -58,6 +63,11 @@ void Arguments::setSlotValue(size_t idx, Matrix* mat) throw(RangeError) {
a.value = m->cast<paddle::Matrix>(mat->getSharedPtr()); a.value = m->cast<paddle::Matrix>(mat->getSharedPtr());
} }
void Arguments::setSlotGrad(size_t idx, Matrix* mat) throw(RangeError) {
auto& a = m->getArg(idx);
a.grad = m->cast<paddle::Matrix>(mat->getSharedPtr());
}
void Arguments::setSlotIn(size_t idx, Matrix* mat) throw(RangeError) { void Arguments::setSlotIn(size_t idx, Matrix* mat) throw(RangeError) {
auto& a = m->getArg(idx); auto& a = m->getArg(idx);
a.in = m->cast<paddle::Matrix>(mat->getSharedPtr()); a.in = m->cast<paddle::Matrix>(mat->getSharedPtr());
......
paddle/api/Paddle.swig
浏览文件 @ 021b3a44
...@@ -194,4 +194,3 @@ namespace std { ...@@ -194,4 +194,3 @@ namespace std {
%ignore ParameterTraverseCallbackPrivate; %ignore ParameterTraverseCallbackPrivate;
%include "utils/GlobalConstants.h" %include "utils/GlobalConstants.h"
%include "api/PaddleAPI.h" %include "api/PaddleAPI.h"
paddle/api/PaddleAPI.h
浏览文件 @ 021b3a44
...@@ -156,12 +156,15 @@ public: ...@@ -156,12 +156,15 @@ public:
* @param dim1 dimension of data. * @param dim1 dimension of data.
* @param dim2 dimension of data. * @param dim2 dimension of data.
* @param copy true if copy into a new matrix, false will create * @param copy true if copy into a new matrix, false will create
* matrix inplace. * matrix inplace. copy = false should be used with extreme
* care because Matrix will share the memory with the given
* numpy array. If the numpy array object is no longer valid,
* the memory space will not be usable.
*/ */
static Matrix* createCpuDenseFromNumpy(float* data, static Matrix* createCpuDenseFromNumpy(float* data,
int dim1, int dim1,
int dim2, int dim2,
bool copy = false); bool copy = true);
/// Create Gpu Dense Matrix from numpy matrix, dtype=float32 /// Create Gpu Dense Matrix from numpy matrix, dtype=float32
static Matrix* createGpuDenseFromNumpy(float* data, int dim1, int dim2); static Matrix* createGpuDenseFromNumpy(float* data, int dim1, int dim2);
...@@ -271,11 +274,18 @@ public: ...@@ -271,11 +274,18 @@ public:
*/ */
static Vector* createCpuVectorFromNumpy(float* data, static Vector* createCpuVectorFromNumpy(float* data,
int dim, int dim,
bool copy = false); bool copy = true);
/// Create Gpu Vector from numpy array, which dtype=float32 /// Create Gpu Vector from numpy array, which dtype=float32
static Vector* createGpuVectorFromNumpy(float* data, int dim); static Vector* createGpuVectorFromNumpy(float* data, int dim);
/**
* copy from another vector
* throw(RangeError) if size of src vector is different from size of this
* vector
*/
void copyFrom(Vector* src) throw(RangeError);
/// Cast to numpy array inplace. /// Cast to numpy array inplace.
void toNumpyArrayInplace(float** view_data, int* dim1) throw(UnsupportError); void toNumpyArrayInplace(float** view_data, int* dim1) throw(UnsupportError);
...@@ -339,7 +349,7 @@ public: ...@@ -339,7 +349,7 @@ public:
*/ */
static IVector* createCpuVectorFromNumpy(int* data, static IVector* createCpuVectorFromNumpy(int* data,
int dim, int dim,
bool copy = false); bool copy = true);
/** /**
* Create Gpu IVector from numpy array, which dtype=int32 * Create Gpu IVector from numpy array, which dtype=int32
*/ */
...@@ -418,6 +428,7 @@ public: ...@@ -418,6 +428,7 @@ public:
* the param idx is the slot id * the param idx is the slot id
*/ */
Matrix* getSlotValue(size_t idx) const throw(RangeError); Matrix* getSlotValue(size_t idx) const throw(RangeError);
Matrix* getSlotGrad(size_t idx) const throw(RangeError);
IVector* getSlotIds(size_t idx) const throw(RangeError); IVector* getSlotIds(size_t idx) const throw(RangeError);
Matrix* getSlotIn(size_t idx) const throw(RangeError); Matrix* getSlotIn(size_t idx) const throw(RangeError);
IVector* getSlotSequenceStartPositions(size_t idx) const throw(RangeError); IVector* getSlotSequenceStartPositions(size_t idx) const throw(RangeError);
...@@ -434,6 +445,7 @@ public: ...@@ -434,6 +445,7 @@ public:
* The other param is the input Matrix or vector. * The other param is the input Matrix or vector.
*/ */
void setSlotValue(size_t idx, Matrix* mat) throw(RangeError); void setSlotValue(size_t idx, Matrix* mat) throw(RangeError);
void setSlotGrad(size_t idx, Matrix* mat) throw(RangeError);
void setSlotIn(size_t idx, Matrix* mat) throw(RangeError); void setSlotIn(size_t idx, Matrix* mat) throw(RangeError);
void setSlotIds(size_t idx, IVector* vec) throw(RangeError); void setSlotIds(size_t idx, IVector* vec) throw(RangeError);
void setSlotSequenceStartPositions(size_t idx, void setSlotSequenceStartPositions(size_t idx,
...@@ -535,6 +547,7 @@ public: ...@@ -535,6 +547,7 @@ public:
size_t getID() const; size_t getID() const;
ParameterConfig* getConfig(); ParameterConfig* getConfig();
void setValueUpdated();
private: private:
static Parameter* createFromRawPtr(void* ptr); static Parameter* createFromRawPtr(void* ptr);
......
paddle/api/Parameter.cpp
浏览文件 @ 021b3a44
...@@ -68,3 +68,5 @@ ParameterConfig* Parameter::getConfig() { ...@@ -68,3 +68,5 @@ ParameterConfig* Parameter::getConfig() {
} }
size_t Parameter::getID() const { return m->getPtr()->getID(); } size_t Parameter::getID() const { return m->getPtr()->getID(); }
void Parameter::setValueUpdated() { m->getPtr()->setValueUpdated(); }
paddle/api/Vector.cpp
浏览文件 @ 021b3a44
...@@ -281,6 +281,13 @@ FloatArray Vector::getData() const { ...@@ -281,6 +281,13 @@ FloatArray Vector::getData() const {
} }
} }
void Vector::copyFrom(Vector* src) throw(RangeError) {
if (src->m->vec->getSize() != m->vec->getSize()) {
throw RangeError();
}
m->vec->copyFrom(*src->m->vec);
}
bool Vector::isGpu() const { bool Vector::isGpu() const {
return std::dynamic_pointer_cast<paddle::GpuVector>(m->vec) != nullptr; return std::dynamic_pointer_cast<paddle::GpuVector>(m->vec) != nullptr;
} }
......
paddle/api/test/testMatrix.py
浏览文件 @ 021b3a44
...@@ -68,7 +68,7 @@ class TestMatrix(unittest.TestCase): ...@@ -68,7 +68,7 @@ class TestMatrix(unittest.TestCase):
def test_numpyCpu(self): def test_numpyCpu(self):
numpy_mat = np.matrix([[1, 2], [3, 4], [5, 6]], dtype="float32") numpy_mat = np.matrix([[1, 2], [3, 4], [5, 6]], dtype="float32")
m = swig_paddle.Matrix.createCpuDenseFromNumpy(numpy_mat) m = swig_paddle.Matrix.createCpuDenseFromNumpy(numpy_mat, copy=False)
self.assertEqual((int(m.getHeight()), int(m.getWidth())), self.assertEqual((int(m.getHeight()), int(m.getWidth())),
numpy_mat.shape) numpy_mat.shape)
......
paddle/api/test/testVector.py
浏览文件 @ 021b3a44
...@@ -43,7 +43,7 @@ class TestIVector(unittest.TestCase): ...@@ -43,7 +43,7 @@ class TestIVector(unittest.TestCase):
def test_cpu_numpy(self): def test_cpu_numpy(self):
vec = np.array([1, 3, 4, 65, 78, 1, 4], dtype="int32") vec = np.array([1, 3, 4, 65, 78, 1, 4], dtype="int32")
iv = swig_paddle.IVector.createCpuVectorFromNumpy(vec) iv = swig_paddle.IVector.createCpuVectorFromNumpy(vec, copy=False)
self.assertEqual(vec.shape[0], int(iv.__len__())) self.assertEqual(vec.shape[0], int(iv.__len__()))
vec[4] = 832 vec[4] = 832
for i in xrange(len(iv)): for i in xrange(len(iv)):
...@@ -107,7 +107,7 @@ class TestVector(unittest.TestCase): ...@@ -107,7 +107,7 @@ class TestVector(unittest.TestCase):
def testCpuNumpy(self): def testCpuNumpy(self):
numpy_arr = np.array([1.2, 2.3, 3.4, 4.5], dtype="float32") numpy_arr = np.array([1.2, 2.3, 3.4, 4.5], dtype="float32")
vec = swig_paddle.Vector.createCpuVectorFromNumpy(numpy_arr) vec = swig_paddle.Vector.createCpuVectorFromNumpy(numpy_arr, copy=False)
assert isinstance(vec, swig_paddle.Vector) assert isinstance(vec, swig_paddle.Vector)
numpy_arr[0] = 0.1 numpy_arr[0] = 0.1
for n, v in zip(numpy_arr, vec): for n, v in zip(numpy_arr, vec):
......
paddle/api/test/util.py
浏览文件 @ 021b3a44
...@@ -24,7 +24,9 @@ def doubleEqual(a, b): ...@@ -24,7 +24,9 @@ def doubleEqual(a, b):
def __readFromFile(): def __readFromFile():
for i in xrange(10002): for i in xrange(10002):
yield np.random.rand(784), random.randint(0, 9) label = np.random.randint(0, 9)
sample = np.random.rand(784) + 0.1 * label
yield sample, label
def loadMNISTTrainData(batch_size=100): def loadMNISTTrainData(batch_size=100):
......
paddle/gserver/dataproviders/DataProvider.h
浏览文件 @ 021b3a44
...@@ -271,8 +271,10 @@ public: ...@@ -271,8 +271,10 @@ public:
void finishAsyncLoad() { void finishAsyncLoad() {
stopping_ = true; stopping_ = true;
taskReadySem_.post(); taskReadySem_.post();
if (asyncLoader_) {
asyncLoader_->join(); asyncLoader_->join();
} }
}
void setPending(bool pending) { pending_ = pending; } void setPending(bool pending) { pending_ = pending; }
......
paddle/gserver/layers/BatchNormBaseLayer.cpp
浏览文件 @ 021b3a44
...@@ -60,18 +60,16 @@ bool BatchNormBaseLayer::init(const LayerMap& layerMap, ...@@ -60,18 +60,16 @@ bool BatchNormBaseLayer::init(const LayerMap& layerMap,
void BatchNormBaseLayer::calFeatureMapSize() { void BatchNormBaseLayer::calFeatureMapSize() {
const ImageConfig& conf = config_.inputs(0).image_conf(); const ImageConfig& conf = config_.inputs(0).image_conf();
if (inputLayers_[0]->getOutput().getFrameHeight() == 0 &&
inputLayers_[0]->getOutput().getFrameWidth() == 0) {
imgSize_ = conf.img_size();
imageH_ = imgSize_;
imageW_ = imgSize_;
} else {
imageH_ = inputLayers_[0]->getOutput().getFrameHeight(); imageH_ = inputLayers_[0]->getOutput().getFrameHeight();
imageW_ = inputLayers_[0]->getOutput().getFrameWidth(); imageW_ = inputLayers_[0]->getOutput().getFrameWidth();
} if (imageH_ == 0 && imageW_ == 0) {
imgPixels_ = imageH_ * imageW_; imageH_ = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
imageW_ = conf.img_size();
} else {
getOutput().setFrameHeight(imageH_); getOutput().setFrameHeight(imageH_);
getOutput().setFrameWidth(imageW_); getOutput().setFrameWidth(imageW_);
}
imgPixels_ = imageH_ * imageW_;
} }
} // namespace paddle } // namespace paddle
paddle/gserver/layers/BatchNormBaseLayer.h
浏览文件 @ 021b3a44
...@@ -77,9 +77,8 @@ protected: ...@@ -77,9 +77,8 @@ protected:
MatrixPtr savedMean_; MatrixPtr savedMean_;
MatrixPtr savedInvVar_; MatrixPtr savedInvVar_;
/// Height or width of input image feature, now height is equal to width. /// Height or width of input image feature.
/// imgSize is 1 if the input is fully-connected layer. /// Both of them are 1 if the input is fully-connected layer.
int imgSize_;
int imageH_; int imageH_;
int imageW_; int imageW_;
/// Height * Width. /// Height * Width.
......
paddle/gserver/layers/BilinearInterpLayer.cpp
浏览文件 @ 021b3a44
...@@ -26,15 +26,15 @@ size_t BilinearInterpLayer::getSize() { ...@@ -26,15 +26,15 @@ size_t BilinearInterpLayer::getSize() {
const BilinearInterpConfig& conf = config_.inputs(0).bilinear_interp_conf(); const BilinearInterpConfig& conf = config_.inputs(0).bilinear_interp_conf();
if (inImgH_ == 0) { if (inImgH_ == 0) {
inImgH_ = conf.img_size_y(); inImgH_ = conf.image_conf().img_size_y();
} }
if (inImgW_ == 0) { if (inImgW_ == 0) {
inImgW_ = conf.img_size_x(); inImgW_ = conf.image_conf().img_size();
} }
outImgH_ = conf.out_size_y(); outImgH_ = conf.out_size_y();
outImgW_ = conf.out_size_x(); outImgW_ = conf.out_size_x();
numChannels_ = conf.num_channels(); numChannels_ = conf.image_conf().channels();
CHECK(outImgH_ > 0 && outImgW_ > 0); CHECK(outImgH_ > 0 && outImgW_ > 0);
CHECK(inImgH_ > 0 && inImgW_ > 0); CHECK(inImgH_ > 0 && inImgW_ > 0);
......
paddle/gserver/layers/ConvBaseLayer.cpp
浏览文件 @ 021b3a44
...@@ -38,11 +38,12 @@ bool ConvBaseLayer::init(const LayerMap& layerMap, ...@@ -38,11 +38,12 @@ bool ConvBaseLayer::init(const LayerMap& layerMap,
filterSizeY_.push_back(conf.filter_size_y()); filterSizeY_.push_back(conf.filter_size_y());
filterPixels_.push_back(filterSize_.back() * filterSizeY_.back()); filterPixels_.push_back(filterSize_.back() * filterSizeY_.back());
channels_.push_back(conf.channels()); channels_.push_back(conf.channels());
imgSizeH_.push_back(conf.img_size()); imgSizeH_.push_back(conf.has_img_size_y() ? conf.img_size_y()
: conf.img_size());
imgSizeW_.push_back(conf.img_size()); imgSizeW_.push_back(conf.img_size());
groups_.push_back(conf.groups()); groups_.push_back(conf.groups());
filterChannels_.push_back(conf.filter_channels()); filterChannels_.push_back(conf.filter_channels());
outputH_.push_back(conf.output_x()); outputH_.push_back(conf.has_output_y() ? conf.output_y() : conf.output_x());
outputW_.push_back(conf.output_x()); outputW_.push_back(conf.output_x());
} }
...@@ -91,16 +92,19 @@ size_t ConvBaseLayer::calOutputSize() { ...@@ -91,16 +92,19 @@ size_t ConvBaseLayer::calOutputSize() {
for (size_t i = 0; i < inputLayers_.size(); i++) { for (size_t i = 0; i < inputLayers_.size(); i++) {
inH.push_back(inputLayers_[i]->getOutput().getFrameHeight()); inH.push_back(inputLayers_[i]->getOutput().getFrameHeight());
inW.push_back(inputLayers_[i]->getOutput().getFrameWidth()); inW.push_back(inputLayers_[i]->getOutput().getFrameWidth());
const ConvConfig& conf = config_.inputs(i).conv_conf();
if (isDeconv_) { if (isDeconv_) {
if (inH[i] == 0) inH[i] = config_.inputs(i).conv_conf().output_x(); if (inH[i] == 0)
if (inW[i] == 0) inW[i] = config_.inputs(i).conv_conf().output_x(); inH[i] = conf.has_output_y() ? conf.output_y() : conf.output_x();
if (inW[i] == 0) inW[i] = conf.output_x();
outH.push_back(imageSize( outH.push_back(imageSize(
inH[i], filterSizeY_[i], paddingY_[i], strideY_[i], caffeMode_)); inH[i], filterSizeY_[i], paddingY_[i], strideY_[i], caffeMode_));
outW.push_back(imageSize( outW.push_back(imageSize(
inW[i], filterSize_[i], padding_[i], stride_[i], caffeMode_)); inW[i], filterSize_[i], padding_[i], stride_[i], caffeMode_));
} else { } else {
if (inH[i] == 0) inH[i] = config_.inputs(i).conv_conf().img_size(); if (inH[i] == 0)
if (inW[i] == 0) inW[i] = config_.inputs(i).conv_conf().img_size(); inH[i] = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
if (inW[i] == 0) inW[i] = conf.img_size();
outH.push_back(outputSize( outH.push_back(outputSize(
inH[i], filterSizeY_[i], paddingY_[i], strideY_[i], caffeMode_)); inH[i], filterSizeY_[i], paddingY_[i], strideY_[i], caffeMode_));
outW.push_back(outputSize( outW.push_back(outputSize(
......
paddle/gserver/layers/ConvOperator.cpp
浏览文件 @ 021b3a44
...@@ -93,9 +93,9 @@ private: ...@@ -93,9 +93,9 @@ private:
bool caffeMode_; bool caffeMode_;
int inputOffset_, outputOffset_, weightOffset_; int inputOffset_, outputOffset_, weightOffset_;
int numFilters_; int numFilters_;
int padding_, stride_, filterSize_, channels_, imgSize_; int padding_, stride_, filterSize_, channels_, imgSize_, imgSizeY_;
int paddingY_, strideY_, filterSizeY_; int paddingY_, strideY_, filterSizeY_;
int imgPixels_, filterPixels_, filterChannels_, outputX_, outputs_; int imgPixels_, filterPixels_, filterChannels_, outputX_, outputY_, outputs_;
/// Following member variables are same with CudnnConvLayer. /// Following member variables are same with CudnnConvLayer.
/// There is no explanation here. /// There is no explanation here.
...@@ -144,7 +144,7 @@ void ConvOperator::allocConvWorkSpace(size_t maxWorkSpace) { ...@@ -144,7 +144,7 @@ void ConvOperator::allocConvWorkSpace(size_t maxWorkSpace) {
void ConvOperator::reshape(int batchSize) { void ConvOperator::reshape(int batchSize) {
imageH_ = ins_[0]->getFrameHeight(); imageH_ = ins_[0]->getFrameHeight();
imageW_ = ins_[0]->getFrameWidth(); imageW_ = ins_[0]->getFrameWidth();
if (imageH_ == 0) imageH_ = imgSize_; if (imageH_ == 0) imageH_ = imgSizeY_;
if (imageW_ == 0) imageW_ = imgSize_; if (imageW_ == 0) imageW_ = imgSize_;
outputH_ = outputSize(imageH_, filterSizeY_, paddingY_, strideY_, caffeMode_); outputH_ = outputSize(imageH_, filterSizeY_, paddingY_, strideY_, caffeMode_);
outputW_ = outputSize(imageW_, filterSize_, padding_, stride_, caffeMode_); outputW_ = outputSize(imageW_, filterSize_, padding_, stride_, caffeMode_);
...@@ -182,7 +182,10 @@ void ConvOperator::computeConvSizes() { ...@@ -182,7 +182,10 @@ void ConvOperator::computeConvSizes() {
hl_create_tensor_descriptor(&inputDesc_); hl_create_tensor_descriptor(&inputDesc_);
int outputX = int outputX =
outputSize(imgSize_, filterSize_, padding_, stride_, caffeMode_); outputSize(imgSize_, filterSize_, padding_, stride_, caffeMode_);
int outputY =
outputSize(imgSizeY_, filterSizeY_, paddingY_, strideY_, caffeMode_);
CHECK_EQ(outputX, outputX_); CHECK_EQ(outputX, outputX_);
CHECK_EQ(outputY, outputY_);
hl_create_tensor_descriptor(&outputDesc_); hl_create_tensor_descriptor(&outputDesc_);
hl_create_convolution_descriptor(&convDesc_, hl_create_convolution_descriptor(&convDesc_,
inputDesc_, inputDesc_,
...@@ -236,10 +239,12 @@ void ConvOperator::getConvParams() { ...@@ -236,10 +239,12 @@ void ConvOperator::getConvParams() {
filterPixels_ = filterSize_ * filterSizeY_; filterPixels_ = filterSize_ * filterSizeY_;
channels_ = conf.channels(); channels_ = conf.channels();
imgSize_ = conf.img_size(); imgSize_ = conf.img_size();
imgPixels_ = imgSize_ * imgSize_; imgSizeY_ = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
imgPixels_ = imgSize_ * imgSizeY_;
CHECK_EQ(conf.groups(), 1U); CHECK_EQ(conf.groups(), 1U);
filterChannels_ = conf.filter_channels(); filterChannels_ = conf.filter_channels();
outputX_ = conf.output_x(); outputX_ = conf.output_x();
outputY_ = conf.has_output_y() ? conf.output_y() : conf.output_x();
outputs_ = outputX_ * outputX_; outputs_ = outputX_ * outputX_;
} }
......
paddle/gserver/layers/ConvProjection.cpp
浏览文件 @ 021b3a44
...@@ -46,7 +46,7 @@ void ConvProjection::getConvParams() { ...@@ -46,7 +46,7 @@ void ConvProjection::getConvParams() {
filterH_ = conf.filter_size_y(); filterH_ = conf.filter_size_y();
filterW_ = conf.filter_size(); filterW_ = conf.filter_size();
configImgH_ = conf.img_size(); configImgH_ = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
configImgW_ = conf.img_size(); configImgW_ = conf.img_size();
channels_ = conf.channels(); channels_ = conf.channels();
...@@ -58,8 +58,11 @@ void ConvProjection::getConvParams() { ...@@ -58,8 +58,11 @@ void ConvProjection::getConvParams() {
} }
void ConvProjection::initCudnn() { void ConvProjection::initCudnn() {
hl_create_filter_descriptor( hl_create_filter_descriptor(&filterDesc_,
&filterDesc_, channels_, numFilters_, filterH_, filterW_); channels_ / groups_,
numFilters_ / groups_,
filterH_,
filterW_);
hl_create_tensor_descriptor(&inputDesc_); hl_create_tensor_descriptor(&inputDesc_);
hl_create_tensor_descriptor(&outputDesc_); hl_create_tensor_descriptor(&outputDesc_);
hl_create_convolution_descriptor(&convDesc_, hl_create_convolution_descriptor(&convDesc_,
...@@ -86,7 +89,7 @@ void ConvProjection::initCudnn() { ...@@ -86,7 +89,7 @@ void ConvProjection::initCudnn() {
void ConvProjection::reshapeTensorDesc(int batchSize) { void ConvProjection::reshapeTensorDesc(int batchSize) {
hl_tensor_reshape(inputDesc_, hl_tensor_reshape(inputDesc_,
batchSize, batchSize,
channels_, channels_ / groups_,
imageH_, imageH_,
imageW_, imageW_,
channels_ * imageH_ * imageW_, channels_ * imageH_ * imageW_,
...@@ -115,7 +118,7 @@ void ConvProjection::reshapeTensorDesc(int batchSize) { ...@@ -115,7 +118,7 @@ void ConvProjection::reshapeTensorDesc(int batchSize) {
hl_tensor_reshape(outputDesc_, hl_tensor_reshape(outputDesc_,
batchSize, batchSize,
numFilters_, numFilters_ / groups_,
outputH_, outputH_,
outputW_, outputW_,
nStride, nStride,
......
paddle/gserver/layers/DataLayer.cpp
浏览文件 @ 021b3a44
...@@ -49,8 +49,13 @@ void DataLayer::copyDataToOutput(Argument& output) { ...@@ -49,8 +49,13 @@ void DataLayer::copyDataToOutput(Argument& output) {
output.ids->copyFrom(*data_.ids); output.ids->copyFrom(*data_.ids);
} }
} }
if (config_.height() && config_.width()) {
output.setFrameHeight(config_.height());
output.setFrameWidth(config_.width());
} else {
output.setFrameHeight(data_.getFrameHeight());
output.setFrameHeight(data_.getFrameHeight()); output.setFrameHeight(data_.getFrameHeight());
output.setFrameWidth(data_.getFrameWidth()); }
output.cpuSequenceDims = data_.cpuSequenceDims; output.cpuSequenceDims = data_.cpuSequenceDims;
output.sequenceStartPositions = data_.sequenceStartPositions; output.sequenceStartPositions = data_.sequenceStartPositions;
output.subSequenceStartPositions = data_.subSequenceStartPositions; output.subSequenceStartPositions = data_.subSequenceStartPositions;
......
paddle/gserver/layers/ExpandConvBaseLayer.cpp
浏览文件 @ 021b3a44
...@@ -29,16 +29,18 @@ bool ExpandConvBaseLayer::init(const LayerMap &layerMap, ...@@ -29,16 +29,18 @@ bool ExpandConvBaseLayer::init(const LayerMap &layerMap,
* meaning as in conv, we need to swap channels_ and numFilters here for * meaning as in conv, we need to swap channels_ and numFilters here for
* convTrans, and in other functions too. * convTrans, and in other functions too.
* */ * */
int channel;
int numFilters;
/* Initialize the projection */ /* Initialize the projection */
for (auto &inputConfig : config_.inputs()) { for (auto &inputConfig : config_.inputs()) {
const ConvConfig &conf = inputConfig.conv_conf(); const ConvConfig &conf = inputConfig.conv_conf();
numFilters = isDeconv_ ? conf.channels() : numFilters_; int numFilters = isDeconv_ ? conf.channels() : numFilters_;
subM_.push_back(numFilters / conf.groups()); subM_.push_back(numFilters / conf.groups());
subN_.push_back(conf.output_x() * conf.output_x()); subN_.push_back(conf.output_x() *
channel = isDeconv_ ? numFilters_ : conf.channels(); (conf.has_output_y() ? conf.output_y() : conf.output_x()));
subK_.push_back(channel * conf.filter_size() * conf.filter_size() / int channel = isDeconv_ ? numFilters_ : conf.channels();
subK_.push_back(
channel * conf.filter_size() *
(conf.has_filter_size_y() ? conf.filter_size_y() : conf.filter_size()) /
conf.groups()); conf.groups());
/* Consistent caffe mode for multiple input */ /* Consistent caffe mode for multiple input */
caffeMode_ = conf.caffe_mode(); caffeMode_ = conf.caffe_mode();
...@@ -116,11 +118,11 @@ void ExpandConvBaseLayer::expandOneFrame(MatrixPtr image, ...@@ -116,11 +118,11 @@ void ExpandConvBaseLayer::expandOneFrame(MatrixPtr image,
imgSizeH_[inIdx], imgSizeH_[inIdx],
imgSizeW_[inIdx], imgSizeW_[inIdx],
channel, channel,
filterSizeY_[inIdx],
filterSize_[inIdx], filterSize_[inIdx],
filterSize_[inIdx], strideY_[inIdx],
stride_[inIdx], stride_[inIdx],
stride_[inIdx], paddingY_[inIdx],
padding_[inIdx],
padding_[inIdx], padding_[inIdx],
outputH_[inIdx], outputH_[inIdx],
outputW_[inIdx]); outputW_[inIdx]);
...@@ -145,7 +147,7 @@ void ExpandConvBaseLayer::expandFwdOnce(MatrixPtr image, ...@@ -145,7 +147,7 @@ void ExpandConvBaseLayer::expandFwdOnce(MatrixPtr image,
real *expInData = expandInput_->getData(); real *expInData = expandInput_->getData();
for (int g = 0; g < groups_[inIdx]; ++g) { for (int g = 0; g < groups_[inIdx]; ++g) {
MatrixPtr A = MatrixPtr A =
Matrix::create(wgtData, subK, subM, true, useGpu_); // mark transpose Matrix::create(wgtData, subM, subK, false, useGpu_); // mark transpose
MatrixPtr B = Matrix::create(expInData, subK, subN, false, useGpu_); MatrixPtr B = Matrix::create(expInData, subK, subN, false, useGpu_);
MatrixPtr C = Matrix::create(outData, subM, subN, false, useGpu_); MatrixPtr C = Matrix::create(outData, subM, subN, false, useGpu_);
C->mul(A, B, 1, 1); C->mul(A, B, 1, 1);
...@@ -182,7 +184,7 @@ void ExpandConvBaseLayer::bpropActs(MatrixPtr out, ...@@ -182,7 +184,7 @@ void ExpandConvBaseLayer::bpropActs(MatrixPtr out,
// create temporary matrix // create temporary matrix
MatrixPtr C = Matrix::create(expandInData, subK, subN, false, useGpu_); MatrixPtr C = Matrix::create(expandInData, subK, subN, false, useGpu_);
MatrixPtr B = Matrix::create(localGradData, subM, subN, false, useGpu_); MatrixPtr B = Matrix::create(localGradData, subM, subN, false, useGpu_);
MatrixPtr A = Matrix::create(wgtData, subK, subM, false, useGpu_); MatrixPtr A = Matrix::create(wgtData, subM, subK, true, useGpu_);
C->mul(A, B); // mul C->mul(A, B); // mul
// clear the temporary matrix // clear the temporary matrix
...@@ -208,11 +210,11 @@ void ExpandConvBaseLayer::bpropActs(MatrixPtr out, ...@@ -208,11 +210,11 @@ void ExpandConvBaseLayer::bpropActs(MatrixPtr out,
imgSizeH_[inpIdx], imgSizeH_[inpIdx],
imgSizeW_[inpIdx], imgSizeW_[inpIdx],
channel, channel,
filterSizeY_[inpIdx],
filterSize_[inpIdx], filterSize_[inpIdx],
filterSize_[inpIdx], strideY_[inpIdx],
stride_[inpIdx],
stride_[inpIdx], stride_[inpIdx],
padding_[inpIdx], paddingY_[inpIdx],
padding_[inpIdx], padding_[inpIdx],
outputH_[inpIdx], outputH_[inpIdx],
outputW_[inpIdx], outputW_[inpIdx],
...@@ -247,10 +249,10 @@ void ExpandConvBaseLayer::bpropWeights(MatrixPtr image, ...@@ -247,10 +249,10 @@ void ExpandConvBaseLayer::bpropWeights(MatrixPtr image,
// expand-mul one-group by one // expand-mul one-group by one
for (int g = 0; g < groups_[inpIdx]; g++) { for (int g = 0; g < groups_[inpIdx]; g++) {
MatrixPtr A = Matrix::create(expandInData, subK, subN, false, useGpu_); MatrixPtr A = Matrix::create(expandInData, subK, subN, true, useGpu_);
MatrixPtr B = Matrix::create(gradData, subM, subN, true, useGpu_); MatrixPtr B = Matrix::create(gradData, subM, subN, false, useGpu_);
MatrixPtr C = Matrix::create(wGradData, subK, subM, false, useGpu_); MatrixPtr C = Matrix::create(wGradData, subM, subK, false, useGpu_);
C->mul(A, B, 1, 1); C->mul(B, A, 1, 1);
A->clear(); A->clear();
B->clear(); B->clear();
......
paddle/gserver/layers/MaxOutLayer.cpp
浏览文件 @ 021b3a44
...@@ -25,10 +25,10 @@ size_t MaxOutLayer::getSize() { ...@@ -25,10 +25,10 @@ size_t MaxOutLayer::getSize() {
imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight(); imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight();
imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth(); imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth();
if (imgSizeH_ == 0) { if (imgSizeH_ == 0) {
imgSizeH_ = maxoutConf.img_size_y(); imgSizeH_ = maxoutConf.image_conf().img_size_y();
} }
if (imgSizeW_ == 0) { if (imgSizeW_ == 0) {
imgSizeW_ = maxoutConf.img_size_x(); imgSizeW_ = maxoutConf.image_conf().img_size();
} }
featLen_ = imgSizeH_ * imgSizeW_; featLen_ = imgSizeH_ * imgSizeW_;
...@@ -50,7 +50,7 @@ bool MaxOutLayer::init(const LayerMap& layerMap, ...@@ -50,7 +50,7 @@ bool MaxOutLayer::init(const LayerMap& layerMap,
const MaxOutConfig& conf = config_.inputs(0).maxout_conf(); const MaxOutConfig& conf = config_.inputs(0).maxout_conf();
groups_ = conf.groups(); groups_ = conf.groups();
channels_ = conf.channels(); channels_ = conf.image_conf().channels();
CHECK_EQ(channels_ % groups_, 0UL); CHECK_EQ(channels_ % groups_, 0UL);
outputChannels_ = channels_ / groups_; outputChannels_ = channels_ / groups_;
......
paddle/gserver/layers/NormLayer.cpp
浏览文件 @ 021b3a44
...@@ -48,6 +48,9 @@ bool ResponseNormLayer::init(const LayerMap& layerMap, ...@@ -48,6 +48,9 @@ bool ResponseNormLayer::init(const LayerMap& layerMap,
outputX_ = conf.output_x(); outputX_ = conf.output_x();
imgSize_ = conf.img_size(); imgSize_ = conf.img_size();
denoms_ = NULL; denoms_ = NULL;
outputY_ = conf.has_output_y() ? conf.output_y() : conf.output_x();
imgSizeY_ = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
return true; return true;
} }
......
paddle/gserver/layers/NormLayer.h
浏览文件 @ 021b3a44
...@@ -49,7 +49,7 @@ public: ...@@ -49,7 +49,7 @@ public:
*/ */
class ResponseNormLayer : public NormLayer { class ResponseNormLayer : public NormLayer {
protected: protected:
size_t channels_, size_, outputX_, imgSize_; size_t channels_, size_, outputX_, imgSize_, outputY_, imgSizeY_;
float scale_, pow_; float scale_, pow_;
MatrixPtr denoms_; MatrixPtr denoms_;
......
paddle/gserver/layers/NormProjectionLayer.cpp
浏览文件 @ 021b3a44
...@@ -23,7 +23,7 @@ size_t CMRProjectionNormLayer::getSize() { ...@@ -23,7 +23,7 @@ size_t CMRProjectionNormLayer::getSize() {
imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight(); imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight();
imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth(); imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth();
if (imgSizeH_ == 0) { if (imgSizeH_ == 0) {
imgSizeH_ = imgSize_; imgSizeH_ = imgSizeY_;
} }
if (imgSizeW_ == 0) { if (imgSizeW_ == 0) {
imgSizeW_ = imgSize_; imgSizeW_ = imgSize_;
......
paddle/gserver/layers/SpatialPyramidPoolLayer.cpp
浏览文件 @ 021b3a44
...@@ -56,14 +56,14 @@ ProjectionConfig SpatialPyramidPoolLayer::getConfig(size_t imgSizeW, ...@@ -56,14 +56,14 @@ ProjectionConfig SpatialPyramidPoolLayer::getConfig(size_t imgSizeW,
size_t SpatialPyramidPoolLayer::getSize() { size_t SpatialPyramidPoolLayer::getSize() {
CHECK_EQ(inputLayers_.size(), 1UL); CHECK_EQ(inputLayers_.size(), 1UL);
size_t layerSize = 0; size_t layerSize = 0;
const SppConfig& sppConf = config_.inputs(0).spp_conf(); const ImageConfig& conf = config_.inputs(0).spp_conf().image_conf();
imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight(); imgSizeH_ = inputLayers_[0]->getOutput().getFrameHeight();
imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth(); imgSizeW_ = inputLayers_[0]->getOutput().getFrameWidth();
if (imgSizeH_ == 0) { if (imgSizeH_ == 0) {
imgSizeH_ = sppConf.has_img_size_y() ? sppConf.img_size_y() : imgSizeW_; imgSizeH_ = conf.has_img_size_y() ? conf.img_size_y() : conf.img_size();
} }
if (imgSizeW_ == 0) { if (imgSizeW_ == 0) {
imgSizeW_ = sppConf.img_size(); imgSizeW_ = conf.img_size();
} }
size_t outputH = 1; size_t outputH = 1;
...@@ -82,9 +82,10 @@ bool SpatialPyramidPoolLayer::init(const LayerMap& layerMap, ...@@ -82,9 +82,10 @@ bool SpatialPyramidPoolLayer::init(const LayerMap& layerMap,
pyramidHeight_ = sppConf.pyramid_height(); pyramidHeight_ = sppConf.pyramid_height();
poolType_ = sppConf.pool_type(); poolType_ = sppConf.pool_type();
channels_ = sppConf.channels(); const ImageConfig& imageConf = sppConf.image_conf();
imgSizeW_ = sppConf.img_size(); channels_ = imageConf.channels();
imgSizeH_ = sppConf.has_img_size_y() ? sppConf.img_size_y() : imgSizeW_; imgSizeW_ = imageConf.img_size();
imgSizeH_ = imageConf.has_img_size_y() ? imageConf.img_size_y() : imgSizeW_;
poolProjections_.reserve(pyramidHeight_); poolProjections_.reserve(pyramidHeight_);
projCol_.reserve(pyramidHeight_); projCol_.reserve(pyramidHeight_);
projOutput_.resize(pyramidHeight_); projOutput_.resize(pyramidHeight_);
......
paddle/gserver/tests/CMakeLists.txt
浏览文件 @ 021b3a44
...@@ -34,7 +34,22 @@ add_unittest_without_exec(test_ConvTrans ...@@ -34,7 +34,22 @@ add_unittest_without_exec(test_ConvTrans
add_test(NAME test_ConvTrans add_test(NAME test_ConvTrans
COMMAND test_ConvTrans) COMMAND test_ConvTrans)
################# test_ConvUnify #######################
add_unittest_without_exec(test_ConvUnify
test_ConvUnify.cpp
LayerGradUtil.cpp
TestUtil.cpp)
add_test(NAME test_ConvUnify
COMMAND test_ConvUnify)
################# test_BatchNorm #######################
add_unittest_without_exec(test_BatchNorm
test_BatchNorm.cpp
LayerGradUtil.cpp
TestUtil.cpp)
add_test(NAME test_BatchNorm
COMMAND test_BatchNorm)
################## test_Evaluator ####################### ################## test_Evaluator #######################
add_unittest(test_Evaluator add_unittest(test_Evaluator
test_Evaluator.cpp test_Evaluator.cpp
......
paddle/gserver/tests/img_conv_a.conf
浏览文件 @ 021b3a44
...@@ -34,6 +34,7 @@ conv = img_conv_layer(input=data, filter_size=1, filter_size_y=1, ...@@ -34,6 +34,7 @@ conv = img_conv_layer(input=data, filter_size=1, filter_size_y=1,
num_channels=8, num_channels=8,
num_filters=16, stride=1, num_filters=16, stride=1,
bias_attr=True, bias_attr=True,
act=LinearActivation()) act=LinearActivation(),
groups=2)
outputs(concat, conv) outputs(concat, conv)
paddle/gserver/tests/img_conv_b.conf
浏览文件 @ 021b3a44
...@@ -24,7 +24,7 @@ proj2 = conv_projection(input=data, filter_size=1, filter_size_y=1, ...@@ -24,7 +24,7 @@ proj2 = conv_projection(input=data, filter_size=1, filter_size_y=1,
concat = concat_layer(input=[proj1, proj2], bias_attr=False, act=ReluActivation()) concat = concat_layer(input=[proj1, proj2], bias_attr=False, act=ReluActivation())
proj = conv_projection(input=data, filter_size=1, filter_size_y=1, proj = conv_projection(input=data, filter_size=1, filter_size_y=1,
num_channels=8, num_filters=16, stride=1) num_channels=8, num_filters=16, stride=1, groups=2)
with mixed_layer(bias_attr=True, act=LinearActivation()) as conv: with mixed_layer(bias_attr=True, act=LinearActivation()) as conv:
conv += proj conv += proj
......
paddle/gserver/tests/img_conv_c.conf 0 → 100644
浏览文件 @ 021b3a44
#edit-mode: -*- python -*-
# Copyright (c) 2016 Baidu, 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 *
settings(batch_size=10)
data = data_layer(name ="input", size=8*16*16)
conv1 = img_conv_layer(input=data, filter_size=1, filter_size_y=1,
num_channels=8,
num_filters=16, stride=1,
bias_attr=False,
act=ReluActivation(),
layer_type="exconv")
conv2 = img_conv_layer(input=data, filter_size=1, filter_size_y=1,
num_channels=8,
num_filters=16, stride=1,
bias_attr=False,
act=ReluActivation(),
layer_type="exconv")
concat = concat_layer(input=[conv1, conv2])
conv = img_conv_layer(input=data, filter_size=1, filter_size_y=1,
num_channels=8,
num_filters=16, stride=1,
bias_attr=True,
act=LinearActivation(),
groups=2,
layer_type="exconv")
outputs(concat, conv)
paddle/gserver/tests/img_pool_a.conf
浏览文件 @ 021b3a44
...@@ -28,7 +28,6 @@ maxpool = img_pool_layer(input=conv, ...@@ -28,7 +28,6 @@ maxpool = img_pool_layer(input=conv,
stride_y=2, stride_y=2,
padding=1, padding=1,
padding_y=2, padding_y=2,
img_width=16,
pool_type=MaxPooling(), pool_type=MaxPooling(),
) )
avgpool = img_pool_layer(input=conv, avgpool = img_pool_layer(input=conv,
...@@ -39,7 +38,6 @@ avgpool = img_pool_layer(input=conv, ...@@ -39,7 +38,6 @@ avgpool = img_pool_layer(input=conv,
stride_y=2, stride_y=2,
padding=1, padding=1,
padding_y=2, padding_y=2,
img_width=16,
pool_type=AvgPooling(), pool_type=AvgPooling(),
) )
......
paddle/gserver/tests/test_BatchNorm.cpp 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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 <gtest/gtest.h>
#include <vector>
#include <string>
#include "paddle/gserver/layers/DataLayer.h"
#include "ModelConfig.pb.h"
#include "paddle/trainer/Trainer.h"
#include "paddle/utils/GlobalConstants.h"
#include "paddle/gserver/layers/ExpandConvTransLayer.h"
#include "TestUtil.h"
#include "LayerGradUtil.h"
using namespace paddle; // NOLINT
using namespace std; // NOLINT
P_DECLARE_bool(use_gpu);
P_DECLARE_int32(gpu_id);
P_DECLARE_double(checkgrad_eps);
P_DECLARE_bool(thread_local_rand_use_global_seed);
P_DECLARE_bool(prev_batch_state);
// Test that the batchNormLayer can be followed by a ConvLayer
TEST(Layer, batchNorm) {
FLAGS_use_gpu = false;
TestConfig configBN;
const int CHANNELS = 6272;
const int IMG_SIZE = 1;
configBN.layerConfig.set_type("batch_norm");
configBN.layerConfig.set_name("bn");
configBN.layerConfig.set_size(CHANNELS * IMG_SIZE * IMG_SIZE);
configBN.layerConfig.set_active_type("relu");
configBN.biasSize = CHANNELS;
configBN.inputDefs.push_back({INPUT_DATA, "layer_0",
/* dim= */ IMG_SIZE * IMG_SIZE * CHANNELS,
/* paraSize= */ CHANNELS});
configBN.inputDefs.push_back({INPUT_DATA, "layer_1_running_mean",
1, CHANNELS});
configBN.inputDefs.back().isStatic = true;
configBN.inputDefs.push_back({INPUT_DATA, "layer_2_running_var",
1, CHANNELS});
configBN.inputDefs.back().isStatic = true;
LayerInputConfig* input = configBN.layerConfig.add_inputs();
configBN.layerConfig.add_inputs();
configBN.layerConfig.add_inputs();
ImageConfig* img_conf = input->mutable_image_conf();
img_conf->set_channels(CHANNELS);
img_conf->set_img_size(IMG_SIZE);
// Setting up conv-layer config
TestConfig config;
config.biasSize = 64;
config.layerConfig.set_type("exconv");
config.layerConfig.set_num_filters(64);
config.layerConfig.set_partial_sum(1);
config.layerConfig.set_shared_biases(true);
config.inputDefs.push_back({INPUT_DATA, "bn", 6272, 204800});
input = config.layerConfig.add_inputs();
ConvConfig* conv = input->mutable_conv_conf();
conv->set_filter_size(5);
conv->set_filter_size_y(5);
conv->set_channels(128);
conv->set_padding(1);
conv->set_padding_y(1);
conv->set_stride(2);
conv->set_stride_y(2);
conv->set_groups(1);
conv->set_filter_channels(conv->channels() / conv->groups());
conv->set_img_size(7);
conv->set_output_x(3);
config.layerConfig.set_size(conv->output_x() * conv->output_x() *
config.layerConfig.num_filters());
config.layerConfig.set_name("conv");
// data layer initialize
std::vector<DataLayerPtr> dataLayers;
LayerMap layerMap;
vector<Argument> datas;
initDataLayer(configBN, &dataLayers, &datas, &layerMap, "batch_norm",
100, false, false);
// test layer initialize
std::vector<ParameterPtr> parameters;
LayerPtr bnLayer;
initTestLayer(configBN, &layerMap, &parameters, &bnLayer);
std::vector<ParameterPtr> parameters2;
LayerPtr convLayer;
initTestLayer(config, &layerMap, &parameters2, &convLayer);
bnLayer->forward(PASS_GC);
convLayer->forward(PASS_GC);
CHECK_EQ(convLayer->getOutputValue()->getHeight(), 100);
CHECK_EQ(convLayer->getOutputValue()->getWidth(), 576);
}
int main(int argc, char** argv) {
testing::InitGoogleTest(&argc, argv);
initMain(argc, argv);
FLAGS_thread_local_rand_use_global_seed = true;
srand(1);
return RUN_ALL_TESTS();
}
paddle/gserver/tests/test_ConvUnify.cpp 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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 <gtest/gtest.h>
#include <vector>
#include <string>
#include "paddle/gserver/layers/DataLayer.h"
#include "ModelConfig.pb.h"
#include "paddle/trainer/Trainer.h"
#include "paddle/utils/GlobalConstants.h"
#include "paddle/gserver/layers/ExpandConvTransLayer.h"
#include "paddle/math/MathUtils.h"
#include "TestUtil.h"
#include "LayerGradUtil.h"
using namespace paddle; // NOLINT
using namespace std; // NOLINT
P_DECLARE_bool(use_gpu);
P_DECLARE_int32(gpu_id);
P_DECLARE_double(checkgrad_eps);
P_DECLARE_bool(thread_local_rand_use_global_seed);
P_DECLARE_bool(prev_batch_state);
// Do one forward pass of convTrans layer and check to see if its output
// matches the given result
MatrixPtr doOneConvTest(size_t imgSize, size_t output_x, size_t stride,
size_t padding, size_t filter_size, size_t channel,
size_t numfilters, size_t groups, MatrixPtr& inputData,
real* param, bool useGpu) {
TestConfig config;
config.biasSize = numfilters;
if (useGpu) {
config.layerConfig.set_type("cudnn_conv");
} else {
config.layerConfig.set_type("exconv");
}
config.layerConfig.set_num_filters(numfilters);
config.layerConfig.set_partial_sum(1);
config.layerConfig.set_shared_biases(true);
size_t weightSize = channel* filter_size * filter_size *
config.layerConfig.num_filters() / groups;
config.inputDefs.push_back({INPUT_DATA, "layer_0",
imgSize * imgSize * channel,
weightSize});
LayerInputConfig* input = config.layerConfig.add_inputs();
ConvConfig* conv = input->mutable_conv_conf();
conv->set_filter_size(filter_size);
conv->set_filter_size_y(filter_size);
conv->set_channels(channel);
conv->set_padding(padding);
conv->set_padding_y(padding);
conv->set_stride(stride);
conv->set_stride_y(stride);
conv->set_groups(groups);
conv->set_filter_channels(channel/groups);
conv->set_img_size(imgSize);
conv->set_output_x(output_x);
config.layerConfig.set_size(conv->output_x() * conv->output_x() *
config.layerConfig.num_filters());
config.layerConfig.set_name("conv");
std::vector<DataLayerPtr> dataLayers;
LayerMap layerMap;
vector<Argument> datas;
initDataLayer(config, &dataLayers, &datas, &layerMap, "conv",
1, false, useGpu);
dataLayers[0]->getOutputValue()->zeroMem();
dataLayers[0]->getOutputValue()->copyFrom(*inputData);
// test layer initialize
std::vector<ParameterPtr> parameters;
LayerPtr convLayer;
initTestLayer(config, &layerMap, &parameters, &convLayer);
convLayer->getBiasParameter()->zeroMem();
convLayer->getParameters()[0]->zeroMem();
convLayer->getParameters()[0]->getBuf(PARAMETER_VALUE)->copyFrom(param,
weightSize);
convLayer->forward(PASS_GC);
return convLayer->getOutputValue();
}
TEST(Layer, convParaUnified) {
#ifndef PADDLE_ONLY_CPU
MatrixPtr input, resultCpu, resultGpu;
input = Matrix::create(1, 4 * 4, false, false);
float inputData[] = {1, 2, 3, 4,
5, 6, 7, 8,
9, 10, 11, 12,
13, 14, 15, 16};
float param[] = {1, 2, 3, 4, 5, 6, 7, 8, 9,
9, 8, 7, 6, 5, 4, 3, 2, 1};
input->setData(inputData);
resultCpu = doOneConvTest(/* imgSize */ 4,
/* output_x */ 2,
/* stride */ 1,
/* padding */ 0,
/* filter_size */ 3,
/*channel*/ 1,
/*numfilters*/ 2,
/*groups*/ 1,
input, param, false);
resultGpu = doOneConvTest(/* imgSize */ 4,
/* output_x */ 2,
/* stride */ 1,
/* padding */ 0,
/* filter_size */ 3,
/*channel*/ 1,
/*numfilters*/ 2,
/*groups*/ 1,
input, param, true);
checkMatrixEqual(resultCpu, resultGpu);
input = Matrix::create(1, 3 * 3 * 2, false, false);
float inputData2[] = {1, 2, 3,
4, 5, 6,
7, 8, 9,
10, 11, 12,
13, 14, 15,
16, 17, 18};
float param2[] = {1, 2, 3, 4, 5, 6, 7, 8,
8, 7, 6, 5, 4, 3, 2, 1};
input->setData(inputData2);
resultCpu = doOneConvTest(/* imgSize */ 3,
/* output_x */ 2,
/* stride */ 1,
/* padding */ 0,
/* filter_size */ 2,
/*channel*/ 2,
/*numfilters*/ 2,
/*groups*/ 1,
input, param2, false);
resultGpu = doOneConvTest(/* imgSize */ 3,
/* output_x */ 2,
/* stride */ 1,
/* padding */ 0,
/* filter_size */ 2,
/*channel*/ 2,
/*numfilters*/ 2,
/*groups*/ 1,
input, param2, true);
checkMatrixEqual(resultCpu, resultGpu);
float param3[] = {1, 2, 3, 4,
4, 3, 2, 1};
resultCpu = doOneConvTest(/* imgSize */ 3,
/* output_x */ 2,
/* stride */ 1,
/* padding */ 0,
/* filter_size */ 2,
/*channel*/ 2,
/*numfilters*/ 2,
/*groups*/ 2,
input, param3, false);
resultGpu = doOneConvTest(/* imgSize */ 3,
/* output_x */ 2,
/* stride */ 1,
/* padding */ 0,
/* filter_size */ 2,
/*channel*/ 2,
/*numfilters*/ 2,
/*groups*/ 2,
input, param3, true);
checkMatrixEqual(resultCpu, resultGpu);
#endif
}
int main(int argc, char** argv) {
testing::InitGoogleTest(&argc, argv);
initMain(argc, argv);
FLAGS_thread_local_rand_use_global_seed = true;
srand(1);
return RUN_ALL_TESTS();
}
paddle/gserver/tests/test_LayerGrad.cpp
浏览文件 @ 021b3a44
...@@ -166,9 +166,8 @@ TEST(Projection, scaling) { ...@@ -166,9 +166,8 @@ TEST(Projection, scaling) {
} }
} }
#ifndef PADDLE_ONLY_CPU void testProjectionConv(size_t groups) {
TEST(Projection, conv) { const int NUM_FILTERS = 18;
const int NUM_FILTERS = 16;
const int FILTER_SIZE = 2; const int FILTER_SIZE = 2;
const int FILTER_SIZE_Y = 3; const int FILTER_SIZE_Y = 3;
const int CHANNELS = 3; const int CHANNELS = 3;
...@@ -186,7 +185,7 @@ TEST(Projection, conv) { ...@@ -186,7 +185,7 @@ TEST(Projection, conv) {
conv->set_padding_y(1); conv->set_padding_y(1);
conv->set_stride(2); conv->set_stride(2);
conv->set_stride_y(2); conv->set_stride_y(2);
conv->set_groups(1); conv->set_groups(groups);
conv->set_filter_channels(conv->channels() / conv->groups()); conv->set_filter_channels(conv->channels() / conv->groups());
conv->set_img_size(IMAGE_SIZE); conv->set_img_size(IMAGE_SIZE);
int output_x = outputSize(conv->img_size(), int output_x = outputSize(conv->img_size(),
...@@ -203,16 +202,22 @@ TEST(Projection, conv) { ...@@ -203,16 +202,22 @@ TEST(Projection, conv) {
conf.set_input_size(IMAGE_SIZE * IMAGE_SIZE * CHANNELS); conf.set_input_size(IMAGE_SIZE * IMAGE_SIZE * CHANNELS);
conf.set_output_size(output_x * output_y * NUM_FILTERS); conf.set_output_size(output_x * output_y * NUM_FILTERS);
testProjectionGrad( testProjectionGrad(conf,
conf,
INPUT_DATA, INPUT_DATA,
/* parameterSize */ NUM_FILTERS * CHANNELS * FILTER_SIZE * FILTER_SIZE_Y, /* parameterSize */ NUM_FILTERS * CHANNELS * FILTER_SIZE *
FILTER_SIZE_Y / groups,
/* batchSize */ 100, /* batchSize */ 100,
true, true,
false, false,
NUM_FILTERS, NUM_FILTERS,
true); true);
} }
#ifndef PADDLE_ONLY_CPU
TEST(Projection, conv) {
testProjectionConv(1);
testProjectionConv(3);
}
#endif #endif
TEST(Layer, BilinearInterpLayer) { TEST(Layer, BilinearInterpLayer) {
...@@ -223,9 +228,10 @@ TEST(Layer, BilinearInterpLayer) { ...@@ -223,9 +228,10 @@ TEST(Layer, BilinearInterpLayer) {
LayerInputConfig* input = config.layerConfig.add_inputs(); LayerInputConfig* input = config.layerConfig.add_inputs();
BilinearInterpConfig* bilinear = input->mutable_bilinear_interp_conf(); BilinearInterpConfig* bilinear = input->mutable_bilinear_interp_conf();
bilinear->set_img_size_x(32); ImageConfig* image = bilinear->mutable_image_conf();
bilinear->set_img_size_y(32); image->set_img_size(32);
bilinear->set_num_channels(4); image->set_img_size_y(32);
image->set_channels(4);
for (auto useGpu : {false, true}) { for (auto useGpu : {false, true}) {
for (auto outSize : {32, 64}) { for (auto outSize : {32, 64}) {
...@@ -348,7 +354,7 @@ void testConvLayer(const string& type, bool trans, bool useGpu) { ...@@ -348,7 +354,7 @@ void testConvLayer(const string& type, bool trans, bool useGpu) {
config.layerConfig.set_partial_sum(1); config.layerConfig.set_partial_sum(1);
config.layerConfig.set_shared_biases(true); config.layerConfig.set_shared_biases(true);
config.inputDefs.push_back({INPUT_DATA, "layer_0", 768, 288}); config.inputDefs.push_back({INPUT_DATA, "layer_0", 384, 288});
LayerInputConfig* input = config.layerConfig.add_inputs(); LayerInputConfig* input = config.layerConfig.add_inputs();
ConvConfig* conv = input->mutable_conv_conf(); ConvConfig* conv = input->mutable_conv_conf();
conv->set_filter_size(2); conv->set_filter_size(2);
...@@ -361,12 +367,18 @@ void testConvLayer(const string& type, bool trans, bool useGpu) { ...@@ -361,12 +367,18 @@ void testConvLayer(const string& type, bool trans, bool useGpu) {
conv->set_groups(1); conv->set_groups(1);
conv->set_filter_channels(conv->channels() / conv->groups()); conv->set_filter_channels(conv->channels() / conv->groups());
conv->set_img_size(16); conv->set_img_size(16);
conv->set_img_size_y(8);
conv->set_output_x(outputSize(conv->img_size(), conv->set_output_x(outputSize(conv->img_size(),
conv->filter_size(), conv->filter_size(),
conv->padding(), conv->padding(),
conv->stride(), conv->stride(),
/* caffeMode */ true)); /* caffeMode */ true));
config.layerConfig.set_size(conv->output_x() * conv->output_x() * conv->set_output_y(outputSize(conv->img_size_y(),
conv->filter_size_y(),
conv->padding_y(),
conv->stride_y(),
/* caffeMode */ true));
config.layerConfig.set_size(conv->output_x() * conv->output_y() *
config.layerConfig.num_filters()); config.layerConfig.num_filters());
testLayerGrad(config, "conv", 100, trans, useGpu); testLayerGrad(config, "conv", 100, trans, useGpu);
...@@ -466,10 +478,11 @@ TEST(Layer, maxoutLayer) { ...@@ -466,10 +478,11 @@ TEST(Layer, maxoutLayer) {
config.inputDefs.push_back({INPUT_DATA, "layer_0", 4096, 0}); config.inputDefs.push_back({INPUT_DATA, "layer_0", 4096, 0});
LayerInputConfig* input = config.layerConfig.add_inputs(); LayerInputConfig* input = config.layerConfig.add_inputs();
MaxOutConfig* maxout = input->mutable_maxout_conf(); MaxOutConfig* maxout = input->mutable_maxout_conf();
ImageConfig* image = maxout->mutable_image_conf();
maxout->set_img_size_x(32); image->set_img_size(32);
maxout->set_img_size_y(32); image->set_img_size_y(32);
maxout->set_channels(4); image->set_channels(4);
maxout->set_groups(2); maxout->set_groups(2);
for (auto useGpu : {false, true}) { for (auto useGpu : {false, true}) {
...@@ -981,7 +994,7 @@ void testNormLayer(const string& normType, bool trans, bool useGpu) { ...@@ -981,7 +994,7 @@ void testNormLayer(const string& normType, bool trans, bool useGpu) {
config.layerConfig.set_type("norm"); config.layerConfig.set_type("norm");
config.layerConfig.set_active_type("relu"); config.layerConfig.set_active_type("relu");
config.inputDefs.push_back({INPUT_DATA, "layer_0", 3136, 0}); config.inputDefs.push_back({INPUT_DATA, "layer_0", 1568, 0});
LayerInputConfig* input = config.layerConfig.add_inputs(); LayerInputConfig* input = config.layerConfig.add_inputs();
NormConfig* norm = input->mutable_norm_conf(); NormConfig* norm = input->mutable_norm_conf();
norm->set_norm_type(normType); norm->set_norm_type(normType);
...@@ -991,7 +1004,9 @@ void testNormLayer(const string& normType, bool trans, bool useGpu) { ...@@ -991,7 +1004,9 @@ void testNormLayer(const string& normType, bool trans, bool useGpu) {
norm->set_pow(0.75); norm->set_pow(0.75);
norm->set_blocked(0); norm->set_blocked(0);
norm->set_img_size(14); norm->set_img_size(14);
norm->set_img_size_y(7);
norm->set_output_x(norm->img_size()); norm->set_output_x(norm->img_size());
norm->set_output_y(norm->img_size_y());
if (norm->norm_type() == "cmrnorm" || if (norm->norm_type() == "cmrnorm" ||
norm->norm_type() == "cmrnorm-projection") { norm->norm_type() == "cmrnorm-projection") {
norm->set_scale(norm->scale() / norm->size()); norm->set_scale(norm->scale() / norm->size());
...@@ -999,7 +1014,7 @@ void testNormLayer(const string& normType, bool trans, bool useGpu) { ...@@ -999,7 +1014,7 @@ void testNormLayer(const string& normType, bool trans, bool useGpu) {
norm->set_scale(norm->scale() / (norm->size() * norm->size())); norm->set_scale(norm->scale() / (norm->size() * norm->size()));
} }
config.layerConfig.set_size(norm->output_x() * norm->output_x() * config.layerConfig.set_size(norm->output_x() * norm->output_y() *
norm->channels()); norm->channels());
config.biasSize = 0; config.biasSize = 0;
...@@ -1100,11 +1115,12 @@ void testSppLayer(const string& poolType, ...@@ -1100,11 +1115,12 @@ void testSppLayer(const string& poolType,
SppConfig* sppConfig = input->mutable_spp_conf(); SppConfig* sppConfig = input->mutable_spp_conf();
sppConfig->set_pool_type(poolType); sppConfig->set_pool_type(poolType);
sppConfig->set_pyramid_height(pyramidHeight); sppConfig->set_pyramid_height(pyramidHeight);
sppConfig->set_channels(16); ImageConfig* imageConfig = sppConfig->mutable_image_conf();
sppConfig->set_img_size(10); imageConfig->set_channels(16);
sppConfig->set_img_size_y(20); imageConfig->set_img_size(10);
imageConfig->set_img_size_y(20);
int outputSize = (std::pow(4, sppConfig->pyramid_height()) - 1) / (4 - 1); int outputSize = (std::pow(4, sppConfig->pyramid_height()) - 1) / (4 - 1);
config.layerConfig.set_size(outputSize * sppConfig->channels()); config.layerConfig.set_size(outputSize * imageConfig->channels());
testLayerGrad(config, "spp", 100, trans, useGpu); testLayerGrad(config, "spp", 100, trans, useGpu);
} }
...@@ -1414,13 +1430,15 @@ void testBatchNormLayer(const string& type, bool trans, bool useGpu) { ...@@ -1414,13 +1430,15 @@ void testBatchNormLayer(const string& type, bool trans, bool useGpu) {
TestConfig config; TestConfig config;
const int CHANNELS = 10; const int CHANNELS = 10;
const int IMG_SIZE = 16; const int IMG_SIZE = 16;
const int IMG_SIZE_Y = 8;
size_t size = CHANNELS * IMG_SIZE * IMG_SIZE_Y;
config.layerConfig.set_type(type); config.layerConfig.set_type(type);
config.layerConfig.set_size(CHANNELS * IMG_SIZE * IMG_SIZE); config.layerConfig.set_size(size);
config.layerConfig.set_active_type("sigmoid"); config.layerConfig.set_active_type("sigmoid");
config.biasSize = CHANNELS; config.biasSize = CHANNELS;
config.inputDefs.push_back({INPUT_DATA, config.inputDefs.push_back({INPUT_DATA,
"layer_0", "layer_0",
/* dim= */ IMG_SIZE * IMG_SIZE * CHANNELS, /* dim= */ size,
/* paraSize= */ CHANNELS}); /* paraSize= */ CHANNELS});
config.inputDefs.push_back({INPUT_DATA, "layer_1_running_mean", 1, CHANNELS}); config.inputDefs.push_back({INPUT_DATA, "layer_1_running_mean", 1, CHANNELS});
...@@ -1435,6 +1453,7 @@ void testBatchNormLayer(const string& type, bool trans, bool useGpu) { ...@@ -1435,6 +1453,7 @@ void testBatchNormLayer(const string& type, bool trans, bool useGpu) {
ImageConfig* img_conf = input->mutable_image_conf(); ImageConfig* img_conf = input->mutable_image_conf();
img_conf->set_channels(CHANNELS); img_conf->set_channels(CHANNELS);
img_conf->set_img_size(IMG_SIZE); img_conf->set_img_size(IMG_SIZE);
img_conf->set_img_size_y(IMG_SIZE_Y);
testLayerGrad(config, testLayerGrad(config,
"batch_norm", "batch_norm",
...@@ -1461,6 +1480,7 @@ TEST(Operator, conv) { ...@@ -1461,6 +1480,7 @@ TEST(Operator, conv) {
const int FILTER_SIZE_Y = 3; const int FILTER_SIZE_Y = 3;
const int CHANNELS = 3; const int CHANNELS = 3;
const int IMAGE_SIZE = 16; const int IMAGE_SIZE = 16;
const int IMAGE_SIZE_Y = 8;
OperatorConfig& operatorConf = *config.layerConfig.add_operator_confs(); OperatorConfig& operatorConf = *config.layerConfig.add_operator_confs();
operatorConf.set_type("conv"); operatorConf.set_type("conv");
ConvConfig* conv = operatorConf.mutable_conv_conf(); ConvConfig* conv = operatorConf.mutable_conv_conf();
...@@ -1475,19 +1495,22 @@ TEST(Operator, conv) { ...@@ -1475,19 +1495,22 @@ TEST(Operator, conv) {
conv->set_groups(1); conv->set_groups(1);
conv->set_filter_channels(conv->channels() / conv->groups()); conv->set_filter_channels(conv->channels() / conv->groups());
conv->set_img_size(IMAGE_SIZE); conv->set_img_size(IMAGE_SIZE);
int output_x = outputSize(conv->img_size(), conv->set_img_size_y(IMAGE_SIZE_Y);
conv->set_output_x(outputSize(conv->img_size(),
conv->filter_size(), conv->filter_size(),
conv->padding(), conv->padding(),
conv->stride(), conv->stride(),
/* caffeMode */ true); /* caffeMode */ true));
conv->set_output_x(output_x); conv->set_output_y(outputSize(conv->img_size_y(),
config.layerConfig.set_size(output_x * output_x * conv->filter_size_y(),
config.layerConfig.num_filters()); conv->padding_y(),
config.layerConfig.set_size(conv->output_x() * conv->output_x() * conv->stride_y(),
/* caffeMode */ true));
config.layerConfig.set_size(conv->output_x() * conv->output_y() *
NUM_FILTERS); NUM_FILTERS);
config.inputDefs.push_back( config.inputDefs.push_back(
{INPUT_DATA, "layer_0", IMAGE_SIZE * IMAGE_SIZE * CHANNELS, 0}); {INPUT_DATA, "layer_0", IMAGE_SIZE * IMAGE_SIZE_Y * CHANNELS, 0});
config.inputDefs.push_back( config.inputDefs.push_back(
{INPUT_DATA, {INPUT_DATA,
"layer_1", "layer_1",
......
paddle/gserver/tests/test_NetworkCompare.cpp
浏览文件 @ 021b3a44
...@@ -255,6 +255,16 @@ TEST(Compare, img_conv) { ...@@ -255,6 +255,16 @@ TEST(Compare, img_conv) {
compareNetwork(config_file_a, config_file_b); compareNetwork(config_file_a, config_file_b);
FLAGS_use_gpu = useGpu; FLAGS_use_gpu = useGpu;
} }
// Test cudnn_conv and exconv give the same result
TEST(Compare, img_conv2) {
std::string config_file_a = "./gserver/tests/img_conv_a.conf";
std::string config_file_b = "./gserver/tests/img_conv_c.conf";
bool useGpu = FLAGS_use_gpu;
FLAGS_use_gpu = true;
compareNetwork(config_file_a, config_file_b);
FLAGS_use_gpu = useGpu;
}
#endif #endif
P_DEFINE_string(config_file_a, "", "config of one network to compare"); P_DEFINE_string(config_file_a, "", "config of one network to compare");
......
paddle/math/BaseMatrix.cu
浏览文件 @ 021b3a44
...@@ -1240,6 +1240,12 @@ void BaseMatrixT<T>::assignAtOffset(BaseMatrixT& b, int64_t columnOffset) { ...@@ -1240,6 +1240,12 @@ void BaseMatrixT<T>::assignAtOffset(BaseMatrixT& b, int64_t columnOffset) {
} }
} }
DEFINE_MATRIX_BINARY_OP(DeepSwap, T tmp = a; a = b; b = tmp);
template<class T>
void BaseMatrixT<T>::deepSwap(BaseMatrixT& b) {
applyBinary(binary::DeepSwap<T>(), b);
}
template<> template<>
void BaseMatrixT<real>::rowDotMul(size_t destCol, void BaseMatrixT<real>::rowDotMul(size_t destCol,
BaseMatrixT& b, BaseMatrixT& b,
...@@ -1578,11 +1584,6 @@ void BaseMatrixT<real>::minRows(BaseMatrixT& b) { ...@@ -1578,11 +1584,6 @@ void BaseMatrixT<real>::minRows(BaseMatrixT& b) {
applyRow(aggregate::min(), b); applyRow(aggregate::min(), b);
} }
template<>
void BaseMatrixT<real>::sumCols(BaseMatrixT& b) {
applyCol(aggregate::sum(), b);
}
template<> template<>
void BaseMatrixT<real>::maxCols(BaseMatrixT& b) { void BaseMatrixT<real>::maxCols(BaseMatrixT& b) {
applyCol(aggregate::max(), b); applyCol(aggregate::max(), b);
......
paddle/math/BaseMatrix.h
浏览文件 @ 021b3a44
...@@ -455,6 +455,17 @@ public: ...@@ -455,6 +455,17 @@ public:
*/ */
void assign(T p); void assign(T p);
/**
* @code
* swap(this, b)
* example: swap two Matrices
* MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
* MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
* cpuA->deepSwap(*cpuB);
* @endcode
*/
void deepSwap(BaseMatrixT& b);
/** /**
* @code * @code
* this = this + p * this = this + p
...@@ -1007,8 +1018,6 @@ public: ...@@ -1007,8 +1018,6 @@ public:
/// calculate the minimum value of each row of the matrix b. /// calculate the minimum value of each row of the matrix b.
void minRows(BaseMatrixT& b); void minRows(BaseMatrixT& b);
/// calculate the sum of each column of the matrix b.
void sumCols(BaseMatrixT& b);
/// calculate the maximum value of each column of the matrix b. /// calculate the maximum value of each column of the matrix b.
void maxCols(BaseMatrixT& b); void maxCols(BaseMatrixT& b);
/// calculate the minimum value of each column of the matrix b. /// calculate the minimum value of each column of the matrix b.
......
paddle/math/tests/CMakeLists.txt
浏览文件 @ 021b3a44
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
add_simple_unittest(test_ExecViaCpu) add_simple_unittest(test_ExecViaCpu)
add_simple_unittest(test_SIMDFunctions) add_simple_unittest(test_SIMDFunctions)
add_simple_unittest(test_matrix) add_simple_unittest(test_SparseMatrix)
# TODO(yuyang18): Refactor TestUtil.cpp. Remove this cross module reference. # TODO(yuyang18): Refactor TestUtil.cpp. Remove this cross module reference.
add_unittest(test_matrixCompare add_unittest(test_matrixCompare
...@@ -15,3 +15,5 @@ add_simple_unittest(test_CpuGpuVector) ...@@ -15,3 +15,5 @@ add_simple_unittest(test_CpuGpuVector)
add_simple_unittest(test_Allocator) add_simple_unittest(test_Allocator)
add_simple_unittest(test_FPException) add_simple_unittest(test_FPException)
add_simple_unittest(test_GpuProfiler) add_simple_unittest(test_GpuProfiler)
add_simple_unittest(test_BaseMatrix)
add_simple_unittest(test_Matrix)
paddle/math/tests/TensorCheck.h 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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
/**
* This file provides a TensorCheck template function, which can be used to
* compare CpuMatrix and GpuMatrix, CpuVector and GpuVector, and so on.
*/
#include <cmath>
#include "paddle/math/Matrix.h"
namespace autotest {
using paddle::Matrix;
using paddle::CpuMatrix;
using paddle::GpuMatrix;
using paddle::VectorT;
using paddle::CpuVectorT;
using paddle::GpuVectorT;
class AssertEqual {
public:
AssertEqual(real err = 0) : err_(err) {}
inline bool operator()(real a, real b) {
if (err_ == 0) {
if (a != b) {
return false;
}
} else {
if (std::fabs(a - b) > err_) {
if ((std::fabs(a - b) / std::fabs(a)) > (err_ / 10.0f)) {
return false;
}
}
}
return true;
}
private:
real err_;
};
template <typename Tensor>
class CopyToCpu;
template <>
class CopyToCpu<CpuMatrix> {
public:
explicit CopyToCpu(const CpuMatrix& arg) : arg_(arg) {}
const CpuMatrix& copiedArg() const { return arg_; }
private:
const CpuMatrix& arg_;
};
template <>
class CopyToCpu<GpuMatrix> {
public:
explicit CopyToCpu(const GpuMatrix& arg)
: arg_(arg.getHeight(), arg.getWidth()) {
arg_.copyFrom(arg);
}
CpuMatrix& copiedArg() { return arg_; }
private:
CpuMatrix arg_;
};
template <>
class CopyToCpu<Matrix> {
public:
explicit CopyToCpu(const Matrix& arg)
: arg_(arg.getHeight(), arg.getWidth()) {
arg_.copyFrom(arg);
}
CpuMatrix& copiedArg() { return arg_; }
private:
CpuMatrix arg_;
};
template <typename T>
class CopyToCpu<CpuVectorT<T>> {
public:
explicit CopyToCpu(const CpuVectorT<T>& arg) : arg_(arg) {}
const CpuVectorT<T>& copiedArg() const { return arg_; }
private:
const CpuVectorT<T>& arg_;
};
template <typename T>
class CopyToCpu<GpuVectorT<T>> {
public:
explicit CopyToCpu(const GpuVectorT<T>& arg) : arg_(arg.getSize()) {
arg_.copyFrom(arg);
}
CpuVectorT<T>& copiedArg() { return arg_; }
private:
CpuVectorT<T> arg_;
};
template <typename T>
class CopyToCpu<VectorT<T>> {
public:
explicit CopyToCpu(const VectorT<T>& arg) : arg_(arg.getSize()) {
arg_.copyFrom(arg);
}
CpuVectorT<T>& copiedArg() { return arg_; }
private:
CpuVectorT<T> arg_;
};
template <typename AssertEq>
void TensorCheck(AssertEq compare,
const CpuMatrix& matrix1,
const CpuMatrix& matrix2) {
CHECK(matrix1.getHeight() == matrix2.getHeight());
CHECK(matrix1.getWidth() == matrix2.getWidth());
int height = matrix1.getHeight();
int width = matrix1.getWidth();
const real* data1 = matrix1.getData();
const real* data2 = matrix2.getData();
int count = 0;
for (int i = 0; i < height; i++) {
for (int j = 0; j < width; j++) {
real a = data1[i * width + j];
real b = data2[i * width + j];
if (!compare(a, b)) {
count++;
}
}
}
EXPECT_EQ(count, 0) << "There are " << count << " different element.";
}
template <typename AssertEq, class T>
void TensorCheck(AssertEq compare,
const CpuVectorT<T>& vector1,
const CpuVectorT<T>& vector2) {
CHECK(vector1.getSize() == vector2.getSize());
const T* data1 = vector1.getData();
const T* data2 = vector2.getData();
size_t size = vector1.getSize();
int count = 0;
for (size_t i = 0; i < size; i++) {
real a = data1[i];
real b = data2[i];
if (!compare(a, b)) {
count++;
}
}
EXPECT_EQ(count, 0) << "There are " << count << " different element.";
}
template <typename AssertEq, typename Tensor1, typename Tensor2>
void TensorCheck(AssertEq compare,
const Tensor1& tensor1,
const Tensor2& tensor2) {
TensorCheck(compare,
CopyToCpu<Tensor1>(tensor1).copiedArg(),
CopyToCpu<Tensor2>(tensor2).copiedArg());
}
template <typename AssertEq>
void TensorCheck(AssertEq compare, real args1, real args2) {
EXPECT_EQ(compare(args1, args2), true) << "[Test error] args1 = " << args1
<< ", args2 = " << args2;
}
template <typename AssertEq>
void TensorCheck(AssertEq compare, size_t args1, size_t args2) {
EXPECT_EQ(args1, args2) << "[Test error] args1 = " << args1
<< ", args2 = " << args2;
}
template <typename Tensor1, typename Tensor2>
void TensorCheckEqual(const Tensor1& tensor1, const Tensor2& tensor2) {
AssertEqual compare(0);
TensorCheck(compare,
CopyToCpu<Tensor1>(tensor1).copiedArg(),
CopyToCpu<Tensor2>(tensor2).copiedArg());
}
template <typename Tensor1, typename Tensor2>
void TensorCheckErr(const Tensor1& tensor1, const Tensor2& tensor2) {
#ifndef PADDLE_TYPE_DOUBLE
AssertEqual compare(1e-3);
#else
AssertEqual compare(1e-10);
#endif
TensorCheck(compare,
CopyToCpu<Tensor1>(tensor1).copiedArg(),
CopyToCpu<Tensor2>(tensor2).copiedArg());
}
} // namespace autotest
paddle/math/tests/TestUtils.h 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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
/**
* This file provides a AutoCompare calss to simplify the comparison
* of CPU and GPU member functions.
*
* This takes two steps
* 1. Construct an AutoCompare object.
* When constructing an AutoCompare object, you can set the err argument
* to specify the maximum error for CPU and GPU functions.
*
* 2. Use the template functions cmpWithArg or cmpWithoutArg.
* A. [cmpWithArg] Requires the caller construct the cpu arguments.
*
* AutoCompare test;
* Init Argument arg1,arg2...
* test.cmpWithArg(function, arg1, arg2....)
*
* B. [cmpWithoutArg] The caller do not need construct arguments.
* If matrix used in these functions arguments is the same size.
* Such as the element wise function and the aggregate function
* defined in the BaseMatrix.cpp.
*
* AutoCompare test;
* test.cmpWithoutArg<I...>(function, height, width)
*/
#include <gtest/gtest.h>
#include "paddle/math/Matrix.h"
#include "paddle/math/SparseMatrix.h"
#include "TensorCheck.h"
namespace autotest {
using paddle::BaseMatrix;
using paddle::CpuMatrix;
using paddle::GpuMatrix;
using paddle::CpuIVector;
using paddle::GpuIVector;
using paddle::CpuSparseMatrix;
using paddle::GpuSparseMatrix;
template <typename T1, typename T2>
class ReplaceType {
public:
typedef T1 type;
};
template <>
class ReplaceType<BaseMatrix, CpuMatrix> {
public:
typedef CpuMatrix type;
};
template <>
class ReplaceType<BaseMatrix, GpuMatrix> {
public:
typedef GpuMatrix type;
};
template <>
class ReplaceType<Matrix, CpuMatrix> {
public:
typedef CpuMatrix type;
};
template <>
class ReplaceType<Matrix, GpuMatrix> {
public:
typedef GpuMatrix type;
};
// construct a argument
template <typename T>
T construct(int height, int width);
template <>
float construct(int height, int width) {
return 0.5;
}
template <>
double construct(int height, int width) {
return 0.5;
}
template <>
size_t construct(int height, int width) {
size_t offset = std::rand() % (height < width ? height : width);
return offset;
}
template <>
CpuMatrix construct(int height, int width) {
CpuMatrix a(height, width);
return a;
}
template <>
GpuMatrix construct(int height, int width) {
GpuMatrix a(height, width);
return a;
}
// init a argument
template <typename T>
void init(T& v) {
return;
}
template <>
void init(CpuMatrix& v) {
v.randomizeUniform();
}
template <>
void init(GpuMatrix& v) {
v.randomizeUniform();
}
// init a tuple which contains a set of arguments.
template <std::size_t I = 0, typename... Args>
inline typename std::enable_if<I == sizeof...(Args), void>::type initTuple(
std::tuple<Args...>& t) {}
template <std::size_t I = 0, typename... Args>
inline typename std::enable_if <
I<sizeof...(Args), void>::type initTuple(std::tuple<Args...>& t) {
init(std::get<I>(t));
initTuple<I + 1>(t);
}
// copy a argument, copy src to dest
template <typename T1, typename T2>
void copy(T1& dest, T2& src) {
dest = src;
}
template <>
void copy(GpuMatrix& dest, CpuMatrix& src) {
dest.copyFrom(src);
}
// copy a tuple, copy src to dest
template <std::size_t I = 0, typename... Args1, typename... Args2>
inline typename std::enable_if<I == sizeof...(Args1), void>::type copyTuple(
std::tuple<Args1...>& dest, std::tuple<Args2...>& src) {}
template <std::size_t I = 0, typename... Args1, typename... Args2>
inline typename std::enable_if <
I<sizeof...(Args1), void>::type copyTuple(std::tuple<Args1...>& dest,
std::tuple<Args2...>& src) {
copy(std::get<I>(dest), std::get<I>(src));
copyTuple<I + 1>(dest, src);
}
// call member function
template <typename C,
typename FC,
typename R,
typename... FArgs,
typename... Args>
R call(C& obj, R (FC::*f)(FArgs...), Args&&... args) {
return (obj.*f)(args...);
}
template <typename T>
class ReturnType {
public:
typedef T type;
};
template <>
class ReturnType<CpuMatrix> {
public:
typedef GpuMatrix type;
};
template <>
class ReturnType<CpuIVector> {
public:
typedef GpuIVector type;
};
template <>
class ReturnType<CpuSparseMatrix> {
public:
typedef GpuSparseMatrix type;
};
template <typename T>
typename ReturnType<T>::type autoArgs(T& v) {
return v;
}
template <>
GpuMatrix autoArgs(CpuMatrix& v) {
GpuMatrix a(v.getHeight(), v.getWidth());
a.copyFrom(v);
return a;
}
template <>
GpuIVector autoArgs(CpuIVector& v) {
GpuIVector a(v.getSize());
a.copyFrom(v);
return a;
}
template <>
GpuSparseMatrix autoArgs(CpuSparseMatrix& v) {
GpuSparseMatrix a(v.getHeight(),
v.getWidth(),
v.getElementCnt(),
v.getValueType(),
v.getFormat());
a.copyFrom(v, HPPL_STREAM_DEFAULT);
hl_stream_synchronize(HPPL_STREAM_DEFAULT);
return a;
}
class AutoCompare {
public:
/**
* err is the allowed calculation error.
* The smaller the value of err,
* the stricter the comparison is between CPU and GPU calculations.
*/
AutoCompare(size_t height, size_t width, real err = 1e-3)
: cpu(height, width), gpu(height, width), compare(err) {
init(cpu);
copy(gpu, cpu);
}
template <typename C, typename R, typename... FArgs, typename... Args>
void cmpWithArg(R (C::*f)(FArgs...), Args&&... args) {
static_assert(sizeof...(FArgs) == sizeof...(Args),
"size of parameter packs are not equal");
call(cpu, f, args...);
call(gpu, f, autoArgs(args)...);
TensorCheck(compare, cpu, gpu);
}
template <std::size_t... I, typename C, typename R, typename... Args>
void cmpWithoutArg(R (C::*f)(Args...), size_t height, size_t width) {
static_assert(sizeof...(I) == sizeof...(Args),
"size of parameter packs are not equal");
(void)height;
(void)width;
auto tuple1 = std::make_tuple(
construct<typename ReplaceType<
typename std::decay<
typename std::tuple_element<I,
std::tuple<Args...>>::type>::type,
CpuMatrix>::type>(height, width)...);
auto tuple2 = std::make_tuple(
construct<typename ReplaceType<
typename std::decay<
typename std::tuple_element<I,
std::tuple<Args...>>::type>::type,
GpuMatrix>::type>(height, width)...);
initTuple(tuple1);
copyTuple(tuple2, tuple1);
call(cpu, f, std::get<I>(tuple1)...);
call(gpu, f, std::get<I>(tuple2)...);
TensorCheck(compare, cpu, gpu);
}
protected:
CpuMatrix cpu;
GpuMatrix gpu;
AssertEqual compare;
};
} // namespace autotest
paddle/math/tests/test_BaseMatrix.cpp 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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. */
#ifndef PADDLE_ONLY_CPU
/**
* This test file use autotest::AutoCompare and cmpWithoutArg to compares the
* implementation of CPU and GPU member function in
* BaseMatrix.cpp and Matrix.cpp.
*/
#include <gtest/gtest.h>
#include "paddle/math/BaseMatrix.h"
#include "TestUtils.h"
using paddle::BaseMatrix;
using paddle::Matrix;
using autotest::AutoCompare;
// Test all void (BaseMatrix::*)() function
TEST(BaseMatrix, void) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
auto compare = [height, width](void (BaseMatrix::*f)()) {
AutoCompare test(height, width, 1e-5);
test.cmpWithoutArg(f, height, width);
};
compare(&BaseMatrix::neg);
compare(&BaseMatrix::exp);
compare(&BaseMatrix::log);
compare(&BaseMatrix::sqrt);
compare(&BaseMatrix::square);
compare(&BaseMatrix::reciprocal);
compare(&BaseMatrix::abs);
compare(&BaseMatrix::sign);
compare(&BaseMatrix::zero);
compare(&BaseMatrix::one);
}
}
}
// Test all void (BaseMatrix::*)(real) function
TEST(BaseMatrix, real) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
auto compare = [height, width](void (BaseMatrix::*f)(real)) {
AutoCompare test(height, width, 1e-5);
test.cmpWithoutArg<0>(f, height, width);
};
compare(&BaseMatrix::pow);
compare(&BaseMatrix::subScalar);
compare(&BaseMatrix::mulScalar);
compare(&BaseMatrix::divScalar);
compare(&BaseMatrix::assign);
compare(&BaseMatrix::add);
compare(&BaseMatrix::biggerThanScalar);
compare(&BaseMatrix::downClip);
}
}
}
// Test all void (BaseMatrix::*)(BaseMatrix&) function
TEST(BaseMatrix, BaseMatrix) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
auto compare = [height, width](void (BaseMatrix::*f)(BaseMatrix&)) {
AutoCompare test(height, width, 1e-5);
test.cmpWithoutArg<0>(f, height, width);
};
compare(&BaseMatrix::assign);
compare(&BaseMatrix::add);
compare(&BaseMatrix::relu);
compare(&BaseMatrix::reluDerivative);
compare(&BaseMatrix::softrelu);
compare(&BaseMatrix::softreluDerivative);
compare(&BaseMatrix::brelu);
compare(&BaseMatrix::breluDerivative);
compare(&BaseMatrix::square);
compare(&BaseMatrix::squareDerivative);
compare(&BaseMatrix::tanh);
compare(&BaseMatrix::tanhDerivative);
compare(&BaseMatrix::reciprocal);
compare(&BaseMatrix::reciprocalDerivative);
compare(&BaseMatrix::abs);
compare(&BaseMatrix::absDerivative);
compare(&BaseMatrix::sigmoid);
compare(&BaseMatrix::sigmoidDerivative);
compare(&BaseMatrix::expDerivative);
compare(&BaseMatrix::sign);
compare(&BaseMatrix::exp);
compare(&BaseMatrix::log);
compare(&BaseMatrix::sqrt);
compare(&BaseMatrix::dotMul);
compare(&BaseMatrix::dotMulSquare);
compare(&BaseMatrix::dotSquareMul);
compare(&BaseMatrix::addColVector);
compare(&BaseMatrix::addRowVector);
compare(&BaseMatrix::mulRowVector);
compare(&BaseMatrix::divRowVector);
compare(&BaseMatrix::addP2P);
compare(&BaseMatrix::invSqrt);
}
}
}
// Test all void (BaseMatrix::*)(real, real) function
TEST(BaseMatrix, real_real) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
auto compare = [height, width](void (BaseMatrix::*f)(real, real)) {
AutoCompare test(height, width, 1e-5);
test.cmpWithoutArg<0, 1>(f, height, width);
};
compare(&BaseMatrix::add);
compare(&BaseMatrix::clip);
}
}
}
// Test all void (BaseMatrix::*)(BaseMatrix&, real) function
TEST(BaseMatrix, BaseMatrix_real) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
auto compare = [height, width](void (BaseMatrix::*f)(BaseMatrix&, real)) {
AutoCompare test(height, width, 1e-5);
test.cmpWithoutArg<0, 1>(f, height, width);
};
compare(&BaseMatrix::addBias);
compare(&BaseMatrix::add);
compare(&BaseMatrix::sub);
compare(&BaseMatrix::pow);
compare(&BaseMatrix::addScalar);
compare(&BaseMatrix::subScalar);
compare(&BaseMatrix::mulScalar);
compare(&BaseMatrix::divScalar);
compare(&BaseMatrix::scalarDiv);
compare(&BaseMatrix::addSquare);
compare(&BaseMatrix::isEqualTo);
}
}
}
// Test all void (BaseMatrix::*)(BaseMatrix&, BaseMatrix&) function
TEST(BaseMatrix, BaseMatrix_BaseMatrix) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
auto compare = [height,
width](void (BaseMatrix::*f)(BaseMatrix&, BaseMatrix&)) {
AutoCompare test(height, width, 1e-5);
test.cmpWithoutArg<0, 1>(f, height, width);
};
compare(&BaseMatrix::softCrossEntropy);
compare(&BaseMatrix::softCrossEntropyBp);
compare(&BaseMatrix::binaryLabelCrossEntropy);
compare(&BaseMatrix::binaryLabelCrossEntropyBp);
compare(&BaseMatrix::sub);
compare(&BaseMatrix::add2);
compare(&BaseMatrix::dotMul);
compare(&BaseMatrix::dotDiv);
compare(&BaseMatrix::logisticRegressionLoss);
compare(&BaseMatrix::logisticRegressionLossBp);
compare(&BaseMatrix::biggerThan);
compare(&BaseMatrix::max);
compare(&BaseMatrix::dotMulSquare);
compare(&BaseMatrix::dotSquareSquare);
}
}
}
void TestEelementWise(size_t height, size_t width) {
AutoCompare rowScale(height, width);
rowScale.cmpWithoutArg<0, 1, 2>(&BaseMatrix::rowScale, height, width);
AutoCompare rowDotMul(height, width);
rowDotMul.cmpWithoutArg<0, 1, 2>(&BaseMatrix::rowDotMul, height, width);
AutoCompare binaryClassificationError(height, width);
binaryClassificationError.cmpWithoutArg<0, 1, 2, 3>(
&BaseMatrix::binaryClassificationError, height, width);
AutoCompare sumOfSquaresBp(height, width);
sumOfSquaresBp.cmpWithoutArg<0, 1>(&Matrix::sumOfSquaresBp, height, width);
}
void TestAggregateToRow(size_t height, size_t width) {
AutoCompare maxCols(1, width);
maxCols.cmpWithoutArg<0>(&BaseMatrix::maxCols, height, width);
AutoCompare minCols(1, width);
minCols.cmpWithoutArg<0>(&BaseMatrix::minCols, height, width);
AutoCompare addDotMulVMM(1, width);
addDotMulVMM.cmpWithoutArg<0, 1>(&BaseMatrix::addDotMulVMM, height, width);
AutoCompare sumCols(1, width);
sumCols.cmpWithoutArg<0, 1, 2>(&BaseMatrix::sumCols, height, width);
AutoCompare collectBias(1, width);
collectBias.cmpWithoutArg<0, 1>(
static_cast<void (Matrix::*)(Matrix&, real)>(&Matrix::collectBias),
height,
width);
}
void TestAggregateToCol(size_t height, size_t width) {
AutoCompare maxRows(height, 1);
maxRows.cmpWithoutArg<0>(&BaseMatrix::maxRows, height, width);
AutoCompare minRows(height, 1);
minRows.cmpWithoutArg<0>(&BaseMatrix::minRows, height, width);
AutoCompare sumRows(height, 1);
sumRows.cmpWithoutArg<0, 1, 2>(&BaseMatrix::sumRows, height, width);
AutoCompare sumOfSquares(height, 1);
sumOfSquares.cmpWithoutArg<0, 1>(&Matrix::sumOfSquares, height, width);
}
TEST(BaseMatrix, Other) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
TestEelementWise(height, width);
TestAggregateToRow(height, width);
TestAggregateToCol(height, width);
}
}
}
int main(int argc, char** argv) {
testing::InitGoogleTest(&argc, argv);
paddle::initMain(argc, argv);
return RUN_ALL_TESTS();
}
#endif
paddle/math/tests/test_Matrix.cpp 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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. */
#ifndef PADDLE_ONLY_CPU
/**
* This test file use autotest::AutoCompare and cmpWithArg to compares the
* implementation of CPU and GPU member function in Matrix.cpp.
*/
#include <gtest/gtest.h>
#include "TestUtils.h"
using paddle::BaseMatrix;
using paddle::Matrix;
using paddle::CpuMatrix;
using paddle::CpuIVector;
using paddle::CpuSparseMatrix;
using autotest::AutoCompare;
void testBilinearFwdBwd(int numSamples,
int imgSizeH,
int imgSizeW,
int channels) {
int inWidth = imgSizeH * imgSizeW * channels;
int outWidth = 2 * imgSizeH * 2 * imgSizeW * channels;
real ratioH = 0.5;
real ratioW = 0.5;
AutoCompare forward(numSamples, outWidth);
CpuMatrix arg1(numSamples, inWidth);
arg1.randomizeUniform();
forward.cmpWithArg(&Matrix::bilinearForward,
arg1,
imgSizeH,
imgSizeW,
2 * imgSizeH,
2 * imgSizeW,
channels,
ratioH,
ratioW);
AutoCompare backward(numSamples, inWidth);
CpuMatrix arg2(numSamples, outWidth);
arg2.randomizeUniform();
backward.cmpWithArg(&Matrix::bilinearBackward,
arg2,
2 * imgSizeH,
2 * imgSizeW,
imgSizeH,
imgSizeW,
channels,
ratioH,
ratioW);
}
TEST(Matrix, BilinearFwdBwd) {
for (auto numSamples : {5, 10}) {
for (auto channels : {8, 16}) {
for (auto imgSizeH : {14, 28}) {
for (auto imgSizeW : {16, 30}) {
VLOG(3) << " numSamples=" << numSamples << " channels=" << channels
<< " imgSizeH=" << imgSizeH << " imgSizeW=" << imgSizeW;
testBilinearFwdBwd(numSamples, imgSizeH, imgSizeW, channels);
}
}
}
}
}
void testMatrixAddBias(int height, int width, real scale) {
AutoCompare test(height, width);
CpuMatrix arg1(1, width);
arg1.randomizeUniform();
test.cmpWithArg(
static_cast<void (Matrix::*)(Matrix&, real)>(&Matrix::addBias),
arg1,
scale);
}
void testMatrixAddDotMulMMV(int height, int width) {
AutoCompare test(height, width);
CpuMatrix arg1(height, width);
CpuMatrix arg2(1, width);
arg1.randomizeUniform();
arg2.randomizeUniform();
test.cmpWithArg(&BaseMatrix::addDotMulMMV, arg1, arg2);
}
TEST(Matrix, unary) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
VLOG(3) << " height=" << height << " width=" << width;
testMatrixAddBias(height, width, 1.0);
testMatrixAddBias(height, width, 3.5);
testMatrixAddDotMulMMV(height, width);
}
}
}
void testMatrixAddAtOffset(int height, int width1, int width2, int offset) {
AutoCompare test(height, width2);
CpuMatrix arg1(height, width1);
arg1.randomizeUniform();
test.cmpWithArg(&Matrix::addAtOffset, arg1, offset);
}
void testMatrixAssignAtOffset(int height, int width1, int width2, int offset) {
AutoCompare test(height, width2);
CpuMatrix arg1(height, width1);
arg1.randomizeUniform();
test.cmpWithArg(&Matrix::assignAtOffset, arg1, offset);
}
TEST(Matrix, AtOffset) {
for (auto height : {1, 11, 73, 128, 200}) {
for (auto width1 : {1, 32, 100, 512, 1000}) {
for (auto width2 : {1, 32, 100, 512, 1000}) {
int columnOffset = 0;
int offset = std::abs(width1 - width2);
if (offset) {
columnOffset = std::rand() % offset;
}
VLOG(3) << " height=" << height << " width1=" << width1
<< " width2=" << width2 << " columnOffset = " << columnOffset;
testMatrixAddAtOffset(height, width1, width2, columnOffset);
testMatrixAssignAtOffset(height, width1, width2, columnOffset);
}
}
}
}
void testMatrixSelectRows(int numSamples, int tableSize, int inputDim) {
AutoCompare test(numSamples, inputDim);
CpuMatrix arg1(tableSize, inputDim);
CpuIVector arg2(numSamples);
arg1.randomizeUniform();
arg2.rand(tableSize);
test.cmpWithArg(&Matrix::selectRows, arg1, arg2);
}
TEST(Matrix, tableProjection) {
for (auto numSamples : {10, 100, 1000, 10000, 80000}) {
for (auto tableSize : {10, 100}) {
for (auto inputDim : {20, 50}) {
VLOG(3) << " numSamples=" << numSamples << " tableSize=" << tableSize
<< " inputDim=" << inputDim;
testMatrixSelectRows(numSamples, tableSize, inputDim);
}
}
}
}
void testMatrixCopyByRowIndex(int outHeight, int inHeight, int width) {
AutoCompare test(outHeight, width);
CpuMatrix arg1(inHeight, width);
CpuIVector arg2(outHeight);
arg1.randomizeUniform();
arg2.rand(inHeight);
test.cmpWithArg(&Matrix::copyByRowIndex, arg1, arg2);
}
TEST(Matrix, copyByRowIndex) {
for (auto outHeight : {31, 500, 1000}) {
for (auto inHeight : {17, 257, 500, 1200}) {
for (auto width : {512, 1024}) {
VLOG(3) << outHeight << " " << inHeight << " " << width;
testMatrixCopyByRowIndex(outHeight, inHeight, width);
}
}
}
}
void testCosSim(int heightX, int heightY, int width, real scale) {
AutoCompare test(heightX, 1);
CpuMatrix arg1(heightX, width);
CpuMatrix arg2(heightY, width);
arg1.randomizeUniform();
arg2.randomizeUniform();
arg2.add(-0.5);
test.cmpWithArg(&Matrix::cosSim, arg1, arg2, scale);
}
TEST(Matrix, cosSim) {
for (auto heightX : {10, 100, 1000}) {
for (auto heightY : {1, heightX}) {
for (auto width : {10, 100, 1000}) {
for (auto scale : {1.0, 2.0}) {
testCosSim(heightX, heightY, width, scale);
}
}
}
}
}
void testParamReluForward(int height, int width, int w_height, int w_width) {
AutoCompare test(height, width);
CpuMatrix arg1(height, width);
CpuMatrix arg2(w_height, w_width);
arg1.randomizeUniform();
arg2.randomizeUniform();
arg1.add(-0.5);
test.cmpWithArg(&Matrix::paramReluForward, arg1, arg2);
}
void testParamReluBackwardW(int height, int width, int w_height, int w_width) {
AutoCompare test(w_height, w_width);
CpuMatrix arg1(height, width);
CpuMatrix arg2(height, width);
arg1.randomizeUniform();
arg2.randomizeUniform();
arg2.add(-0.5);
test.cmpWithArg(&Matrix::paramReluBackwardW, arg1, arg2);
}
TEST(Matrix, paramRelu) {
for (auto height : {10, 100}) {
for (auto width : {10, 100}) {
for (auto w_height : {1, 2}) {
for (auto w_width : {1, 2}) {
testParamReluForward(height, width, w_height, w_width);
testParamReluBackwardW(height, width, w_height, w_width);
}
}
}
}
}
void testAddSharedBias(int numSamples, int dim, int channel) {
AutoCompare test(numSamples, dim);
CpuMatrix arg1(1, channel);
arg1.randomizeUniform();
test.cmpWithArg(&Matrix::addSharedBias, arg1, 1.0);
}
void testCollectSharedBias(int numSamples, int dim, int channel) {
AutoCompare test(1, channel);
CpuMatrix arg1(numSamples, dim);
arg1.randomizeUniform();
test.cmpWithArg(&Matrix::collectSharedBias, arg1, 1.0);
}
TEST(Matrix, sharedBias) {
for (auto numSamples : {1, 100, 520}) {
for (auto dim : {100 * 16, 100 * 32}) {
for (auto channel : {8, 16}) {
VLOG(3) << " numSamples=" << numSamples << " dim=" << dim
<< " channel=" << channel;
testAddSharedBias(numSamples, dim, channel);
testCollectSharedBias(numSamples, dim, channel);
}
}
}
}
void testMultiBinaryLabelCrossEntropy(int numSamples, int dim) {
AutoCompare forward(numSamples, 1);
CpuMatrix arg1(numSamples, dim);
CpuSparseMatrix arg2(
numSamples, dim, numSamples, paddle::NO_VALUE, paddle::SPARSE_CSR);
CpuMatrix output1(numSamples, dim);
output1.randomizeUniform();
output1.softmax(arg1);
for (int i = 0; i < numSamples; i++) {
const unsigned int id = std::rand() % dim;
arg2.setRow(i, 1, &id, nullptr);
}
forward.cmpWithArg(&Matrix::multiBinaryLabelCrossEntropy, arg1, arg2);
AutoCompare backward(numSamples, dim);
backward.cmpWithArg(&Matrix::multiBinaryLabelCrossEntropyBp, arg1, arg2);
}
TEST(Matrix, multiBinaryCrossEntropy) {
for (auto numSamples : {100, 1000, 10000}) {
for (auto dim : {100, 1000, 10000}) {
VLOG(3) << " numSamples=" << numSamples << " dim=" << dim;
testMultiBinaryLabelCrossEntropy(numSamples, dim);
}
}
}
int main(int argc, char** argv) {
testing::InitGoogleTest(&argc, argv);
paddle::initMain(argc, argv);
return RUN_ALL_TESTS();
}
#endif
paddle/math/tests/test_matrixCompare.cpp
浏览文件 @ 021b3a44
...@@ -22,163 +22,12 @@ limitations under the License. */ ...@@ -22,163 +22,12 @@ limitations under the License. */
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include "paddle/gserver/tests/TestUtil.h" #include "paddle/gserver/tests/TestUtil.h"
#include "paddle/utils/Stat.h" #include "paddle/utils/Stat.h"
#include "TensorCheck.h"
using namespace paddle; // NOLINT using namespace paddle; // NOLINT
using namespace std; // NOLINT using namespace std; // NOLINT
using autotest::TensorCheckEqual;
template <class T> using autotest::TensorCheckErr;
void VectorCheckEqual(const VectorT<T>& vector1, const VectorT<T>& vector2) {
CHECK(vector1.getSize() == vector2.getSize());
const T* data1 = vector1.getData();
const T* data2 = vector2.getData();
size_t size = vector1.getSize();
int count = 0;
for (size_t i = 0; i < size; i++) {
if (data1[i] != data2[i]) {
count++;
}
}
EXPECT_EQ(count, 0) << "There are " << count << " different element.";
}
void MatrixCheckEqual(const Matrix& matrix1, const Matrix& matrix2) {
CHECK(matrix1.getHeight() == matrix2.getHeight());
CHECK(matrix1.getWidth() == matrix2.getWidth());
int height = matrix1.getHeight();
int width = matrix1.getWidth();
const real* data1 = matrix1.getData();
const real* data2 = matrix2.getData();
int count = 0;
for (int i = 0; i < height; i++) {
for (int j = 0; j < width; j++) {
if (data1[i * width + j] != data2[i * width + j]) {
count++;
}
}
}
EXPECT_EQ(count, 0) << "There are " << count << " different element.";
}
void MatrixCheckErr(const Matrix& matrix1, const Matrix& matrix2) {
CHECK(matrix1.getHeight() == matrix2.getHeight());
CHECK(matrix1.getWidth() == matrix2.getWidth());
#ifndef PADDLE_TYPE_DOUBLE
real err = 1e-3;
#else
real err = 1e-10;
#endif
int height = matrix1.getHeight();
int width = matrix1.getWidth();
const real* data1 = matrix1.getData();
const real* data2 = matrix2.getData();
int count = 0;
for (int i = 0; i < height; i++) {
for (int j = 0; j < width; j++) {
real a = data1[i * width + j];
real b = data2[i * width + j];
if (fabs(a - b) > err) {
if ((fabsf(a - b) / fabsf(a)) > (err / 10.0f)) {
count++;
}
}
}
}
EXPECT_EQ(count, 0) << "There are " << count << " different element.";
}
void testBilinearFwdBwd(int numSamples,
int imgSizeH,
int imgSizeW,
int channels) {
int inWidth = imgSizeH * imgSizeW * channels;
int outWidth = 2 * imgSizeH * 2 * imgSizeW * channels;
real ratioH = 0.5;
real ratioW = 0.5;
// forward
MatrixPtr input = CpuMatrix::create(numSamples, inWidth, false, false);
MatrixPtr inputGpu = GpuMatrix::create(numSamples, inWidth, false, true);
MatrixPtr target = CpuMatrix::create(numSamples, outWidth, false, false);
MatrixPtr targetGpu = GpuMatrix::create(numSamples, outWidth, false, true);
MatrixPtr targetCheck = CpuMatrix::create(numSamples, outWidth, false, false);
input->randomizeUniform();
inputGpu->copyFrom(*input);
target->bilinearForward(*input,
imgSizeH,
imgSizeW,
2 * imgSizeH,
2 * imgSizeW,
channels,
ratioH,
ratioW);
targetGpu->bilinearForward(*inputGpu,
imgSizeH,
imgSizeW,
2 * imgSizeH,
2 * imgSizeW,
channels,
ratioH,
ratioW);
// check
targetCheck->copyFrom(*targetGpu);
MatrixCheckErr(*target, *targetCheck);
// backward
MatrixPtr inputGrad = CpuMatrix::create(numSamples, inWidth, false, false);
MatrixPtr inputGpuGrad = GpuMatrix::create(numSamples, inWidth, false, true);
MatrixPtr targetGrad = CpuMatrix::create(numSamples, outWidth, false, false);
MatrixPtr targetGpuGrad =
GpuMatrix::create(numSamples, outWidth, false, true);
MatrixPtr targetCheckGrad =
CpuMatrix::create(numSamples, inWidth, false, false);
inputGrad->randomizeUniform();
targetGrad->randomizeUniform();
inputGpuGrad->copyFrom(*inputGrad);
targetGpuGrad->copyFrom(*targetGrad);
inputGrad->bilinearBackward(*targetGrad,
2 * imgSizeH,
2 * imgSizeW,
imgSizeH,
imgSizeW,
channels,
ratioH,
ratioW);
inputGpuGrad->bilinearBackward(*targetGpuGrad,
2 * imgSizeH,
2 * imgSizeW,
imgSizeH,
imgSizeW,
channels,
ratioH,
ratioW);
// check
targetCheckGrad->copyFrom(*inputGpuGrad);
MatrixCheckErr(*inputGrad, *targetCheckGrad);
}
TEST(Matrix, BilinearFwdBwd) {
for (auto numSamples : {5, 10}) {
for (auto channels : {8, 16}) {
for (auto imgSizeH : {14, 28}) {
for (auto imgSizeW : {16, 30}) {
VLOG(3) << " numSamples=" << numSamples << " channels=" << channels
<< " imgSizeH=" << imgSizeH << " imgSizeW=" << imgSizeW;
testBilinearFwdBwd(numSamples, imgSizeH, imgSizeW, channels);
}
}
}
}
}
void testMatrixProjectionForward(int contextStart, void testMatrixProjectionForward(int contextStart,
int contextLength, int contextLength,
...@@ -232,12 +81,7 @@ void testMatrixProjectionForward(int contextStart, ...@@ -232,12 +81,7 @@ void testMatrixProjectionForward(int contextStart,
beginPad, beginPad,
padding); padding);
// check TensorCheckEqual(*cpuOutput, *gpuOutput);
MatrixPtr outputCheck =
std::make_shared<CpuMatrix>(batchSize, inputDim * contextLength);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckEqual(*cpuOutput, *outputCheck);
} }
void testMatrixProjectionBackward(int contextStart, void testMatrixProjectionBackward(int contextStart,
...@@ -294,15 +138,9 @@ void testMatrixProjectionBackward(int contextStart, ...@@ -294,15 +138,9 @@ void testMatrixProjectionBackward(int contextStart,
beginPad); beginPad);
} }
// check TensorCheckErr(*cpuInputGrad, *gpuInputGrad);
MatrixPtr inputGradCheck = std::make_shared<CpuMatrix>(batchSize, inputDim);
inputGradCheck->copyFrom(*gpuInputGrad);
MatrixCheckErr(*cpuInputGrad, *inputGradCheck);
if (padding) { if (padding) {
MatrixPtr weightGradChcek = std::make_shared<CpuMatrix>(pad, inputDim); TensorCheckErr(*cpuWeightGrad, *gpuWeightGrad);
weightGradChcek->copyFrom(*gpuWeightGrad);
MatrixCheckErr(*cpuWeightGrad, *weightGradChcek);
} }
} }
...@@ -361,15 +199,8 @@ void testMatrixMaxSequence(int batchSize, int inputDim) { ...@@ -361,15 +199,8 @@ void testMatrixMaxSequence(int batchSize, int inputDim) {
cpuOutput->maxSequenceForward(*cpuInput, *cpuSequence, *cpuIndex); cpuOutput->maxSequenceForward(*cpuInput, *cpuSequence, *cpuIndex);
gpuOutput->maxSequenceForward(*gpuInput, *gpuSequence, *gpuIndex); gpuOutput->maxSequenceForward(*gpuInput, *gpuSequence, *gpuIndex);
// check TensorCheckEqual(*cpuOutput, *gpuOutput);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(newBatchSize, inputDim); TensorCheckEqual(*cpuIndex, *gpuIndex);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckEqual(*cpuOutput, *outputCheck);
IVectorPtr indexCheck = nullptr;
IVector::resizeOrCreate(indexCheck, newBatchSize * inputDim, false);
indexCheck->copyFrom(*gpuIndex);
VectorCheckEqual(*cpuIndex, *indexCheck);
// backward // backward
MatrixPtr cpuOutputGrad = std::make_shared<CpuMatrix>(newBatchSize, inputDim); MatrixPtr cpuOutputGrad = std::make_shared<CpuMatrix>(newBatchSize, inputDim);
...@@ -385,10 +216,7 @@ void testMatrixMaxSequence(int batchSize, int inputDim) { ...@@ -385,10 +216,7 @@ void testMatrixMaxSequence(int batchSize, int inputDim) {
cpuInputGrad->maxSequenceBackward(*cpuOutputGrad, *cpuSequence, *cpuIndex); cpuInputGrad->maxSequenceBackward(*cpuOutputGrad, *cpuSequence, *cpuIndex);
gpuInputGrad->maxSequenceBackward(*gpuOutputGrad, *gpuSequence, *gpuIndex); gpuInputGrad->maxSequenceBackward(*gpuOutputGrad, *gpuSequence, *gpuIndex);
// check TensorCheckEqual(*cpuInputGrad, *gpuInputGrad);
MatrixPtr inputGradCheck = std::make_shared<CpuMatrix>(batchSize, inputDim);
inputGradCheck->copyFrom(*gpuInputGrad);
MatrixCheckEqual(*cpuInputGrad, *inputGradCheck);
} }
TEST(Matrix, maxSequence) { TEST(Matrix, maxSequence) {
...@@ -431,6 +259,8 @@ void testMatrixZeroAtOffset(int height, int width) { ...@@ -431,6 +259,8 @@ void testMatrixZeroAtOffset(int height, int width) {
int columnOffset = rand() % width; // NOLINT we just use rand() for test. int columnOffset = rand() % width; // NOLINT we just use rand() for test.
int numColumns = rand() % (width - columnOffset); // NOLINT int numColumns = rand() % (width - columnOffset); // NOLINT
if (numColumns == 0) return;
cpuA->zeroAtOffset(columnOffset, numColumns); cpuA->zeroAtOffset(columnOffset, numColumns);
gpuA->zeroAtOffset(columnOffset, numColumns); gpuA->zeroAtOffset(columnOffset, numColumns);
...@@ -442,304 +272,26 @@ void testMatrixZeroAtOffset(int height, int width) { ...@@ -442,304 +272,26 @@ void testMatrixZeroAtOffset(int height, int width) {
} }
} }
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width); TensorCheckEqual(*cpuA, *gpuA);
outputCheck->copyFrom(*gpuA); TensorCheckEqual(*cpuA, *cpuTest);
MatrixCheckEqual(*cpuA, *outputCheck);
MatrixCheckEqual(*cpuA, *cpuTest);
}
void testMatrixBinaryAdd(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
cpuA->add(*cpuB);
gpuA->add(*gpuB);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckEqual(*cpuA, *outputCheck);
}
void testMatrixAssign(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
gpuA->copyFrom(*cpuA);
cpuA->assign(2.5);
gpuA->assign(2.5);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckEqual(*cpuA, *outputCheck);
}
void testMatrixAdd(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
gpuA->copyFrom(*cpuA);
cpuA->add(2.5);
gpuA->add(2.5);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckEqual(*cpuA, *outputCheck);
}
void testMatrixSqrt(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
gpuA->copyFrom(*cpuA);
cpuA->sqrt();
gpuA->sqrt();
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixTanhDerivative(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
cpuA->tanhDerivative(*cpuB);
gpuA->tanhDerivative(*gpuB);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixTanh(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
cpuA->tanh(*cpuB);
gpuA->tanh(*gpuB);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixTernarySub(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
cpuA->sub(*cpuB, *cpuC);
gpuA->sub(*gpuB, *gpuC);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckEqual(*cpuA, *outputCheck);
}
void testMatrixSumOfSquaresBp(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
cpuA->sumOfSquaresBp(*cpuB, *cpuC);
gpuA->sumOfSquaresBp(*gpuB, *gpuC);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixBinaryRowScale(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, 1);
MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr gpuA1 = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB1 = std::make_shared<GpuMatrix>(height, 1);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
cpuA1->copyFrom(*cpuA);
cpuB1->copyFrom(*cpuB);
gpuA1->copyFrom(*cpuA);
gpuB1->copyFrom(*cpuB);
cpuA->addColVector(*cpuB);
gpuA->addColVector(*gpuB);
cpuA1->addColumnVector(*cpuB1);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckEqual(*cpuA, *outputCheck);
MatrixCheckEqual(*cpuA, *cpuA1);
}
void testMatrixAddBias(int height, int width, real scale) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(1, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(1, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
cpuA->addBias(*cpuB, scale);
gpuA->addBias(*gpuB, scale);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixTernaryRowScale(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
cpuA1->copyFrom(*cpuA);
cpuB1->copyFrom(*cpuB);
cpuC1->copyFrom(*cpuC);
int columnOffset = rand() % width; // NOLINT
cpuA->rowScale(columnOffset, *cpuB, *cpuC);
gpuA->rowScale(columnOffset, *gpuB, *gpuC);
cpuA1->rowScale2(columnOffset, *cpuB1, *cpuC1);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckEqual(*cpuA, *outputCheck);
MatrixCheckEqual(*cpuA, *cpuA1);
}
void testMatrixTernaryRowDotMul(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
cpuA1->copyFrom(*cpuA);
cpuB1->copyFrom(*cpuB);
cpuC1->copyFrom(*cpuC);
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
int columnOffset = rand() % width; // NOLINT
cpuA->rowDotMul(columnOffset, *cpuB, *cpuC);
gpuA->rowDotMul(columnOffset, *gpuB, *gpuC);
cpuA1->rowDotMul2(columnOffset, *cpuB1, *cpuC1);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *cpuA1);
MatrixCheckErr(*cpuA, *outputCheck);
} }
void testMatrixAddDotMulMMV(int height, int width) { void testMatrixDeepSwap(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width); MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width); MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(1, width); MatrixPtr cpuCopyA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width); MatrixPtr cpuCopyB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(1, width);
MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(1, width);
cpuA->randomizeUniform(); cpuA->randomizeUniform();
cpuB->randomizeUniform(); cpuB->randomizeUniform();
cpuC->randomizeUniform(); cpuCopyA->copyFrom(*cpuA);
gpuA->copyFrom(*cpuA); cpuCopyB->copyFrom(*cpuB);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
cpuA1->copyFrom(*cpuA);
cpuB1->copyFrom(*cpuB);
cpuC1->copyFrom(*cpuC);
cpuA->addDotMulMMV(*cpuB, *cpuC); // swap matrix cpuA and cpuB
gpuA->addDotMulMMV(*gpuB, *gpuC); cpuA->deepSwap(*cpuB);
cpuA1->addDotMulMMV2(*cpuB1, *cpuC1);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width); TensorCheckEqual(*cpuA, *cpuCopyB);
outputCheck->copyFrom(*gpuA); TensorCheckEqual(*cpuB, *cpuCopyA);
MatrixCheckErr(*cpuA, *outputCheck);
MatrixCheckEqual(*cpuA, *cpuA1);
} }
void testMatrixTranspose(int height, int width) { void testMatrixTranspose(int height, int width) {
...@@ -753,9 +305,7 @@ void testMatrixTranspose(int height, int width) { ...@@ -753,9 +305,7 @@ void testMatrixTranspose(int height, int width) {
cpu->transpose(cpuT, false); cpu->transpose(cpuT, false);
gpu->transpose(gpuT, false); gpu->transpose(gpuT, false);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(width, height); TensorCheckEqual(*cpuT, *gpuT);
outputCheck->copyFrom(*gpuT);
MatrixCheckEqual(*cpuT, *outputCheck);
} }
void testMatrixInverse(int height) { void testMatrixInverse(int height) {
...@@ -776,529 +326,127 @@ void testMatrixInverse(int height) { ...@@ -776,529 +326,127 @@ void testMatrixInverse(int height) {
cpu->inverse(cpuI, false); cpu->inverse(cpuI, false);
gpu->inverse(gpuI, false); gpu->inverse(gpuI, false);
outputCheck->copyFrom(*gpuI); TensorCheckErr(*cpuI, *gpuI);
MatrixCheckErr(*cpuI, *outputCheck);
outputCheck->mul(cpu, cpuI);
cpu->setDiag(1.0);
MatrixCheckErr(*cpu, *outputCheck);
}
TEST(Matrix, unary) {
for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
VLOG(3) << " height=" << height << " width=" << width;
// applyUnary
testMatrixAssign(height, width);
testMatrixAdd(height, width);
testMatrixSqrt(height, width);
// applyBinary
testMatrixBinaryAdd(height, width);
testMatrixTanh(height, width);
testMatrixTanhDerivative(height, width);
// applyTernary
testMatrixTernarySub(height, width);
testMatrixSumOfSquaresBp(height, width);
// asRowVector
testMatrixAddBias(height, width, 1.0);
testMatrixAddBias(height, width, 3.5);
testMatrixAddDotMulMMV(height, width);
// asColVector
testMatrixTernaryRowScale(height, width);
testMatrixBinaryRowScale(height, width);
// sum
testMatrixGetSum(height, width);
// transpose
testMatrixTranspose(height, width);
}
// inverse
testMatrixInverse(height);
}
}
void testMatrixSoftmax(int height, int width) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width);
cpuInput->randomizeUniform();
gpuInput->copyFrom(*cpuInput);
cpuOutput->zero();
gpuOutput->zero();
cpuInput->softmax(*cpuOutput);
gpuInput->softmax(*gpuOutput);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckErr(*cpuOutput, *outputCheck);
}
void testSequenceSoftmax(int batchSize) {
// forward
int inputDim = 1;
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(batchSize, inputDim);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(batchSize, inputDim);
cpuInput->randomizeUniform();
gpuInput->copyFrom(*cpuInput);
IVectorPtr cpuSequence;
generateSequenceStartPositions(batchSize, cpuSequence);
IVectorPtr gpuSequence = IVector::create(cpuSequence->getSize(), true);
gpuSequence->copyFrom(*cpuSequence);
cpuInput->sequenceSoftmax(*cpuInput, *cpuSequence);
gpuInput->sequenceSoftmax(*gpuInput, *gpuSequence);
// check
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(batchSize, inputDim);
outputCheck->copyFrom(*gpuInput);
MatrixCheckErr(*cpuInput, *outputCheck);
}
void testMatrixSoftmaxThreshold(int height, int width) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width);
cpuInput->randomizeUniform();
cpuInput->getData()[0] = 100.0;
gpuInput->copyFrom(*cpuInput);
cpuOutput->zero();
gpuOutput->zero();
cpuInput->softmax(*cpuOutput);
gpuInput->softmax(*gpuOutput);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuOutput);
// check output zero
int cpuCount = 0;
int gpuCount = 0;
auto zeroNum = [](MatrixPtr out, int& count) {
for (size_t i = 0; i < out->getHeight(); i++) {
for (size_t j = 0; j < out->getWidth(); j++) {
if (out->getElement(i, j) == 0) count++;
}
}
};
zeroNum(cpuOutput, cpuCount);
zeroNum(outputCheck, gpuCount);
EXPECT_EQ(cpuCount, 0) << "Cpu softmax output value 0";
EXPECT_EQ(gpuCount, 0) << "Gpu softmax output value 0";
}
void testMatrixSoftmaxBp(int height, int width) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width);
cpuInput->randomizeUniform();
gpuInput->copyFrom(*cpuInput);
cpuOutput->randomizeUniform();
gpuOutput->copyFrom(*cpuOutput);
gpuOutput->softmaxBackward(*gpuInput);
MatrixPtr sftMaxSum = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr sftMaxDot = std::make_shared<CpuMatrix>(height, width);
sftMaxDot->dotMul(*cpuOutput, *cpuInput);
sftMaxSum->colMerge(*sftMaxDot);
cpuOutput->softmaxDerivative(*cpuInput, *sftMaxSum);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckErr(*cpuOutput, *outputCheck);
}
TEST(Matrix, softmax) {
for (auto height : {1, 11, 73, 128, 200}) {
for (auto width : {1, 32, 100, 512, 1000}) {
VLOG(3) << " height=" << height << " width=" << width;
testMatrixSoftmax(height, width);
testMatrixSoftmaxBp(height, width);
testMatrixSoftmaxThreshold(height, width);
}
testSequenceSoftmax(height);
}
}
void testMatrixAddDotMulVMM(int height, int width, int endCol = 0) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(1, width);
MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC1 = std::make_shared<CpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
cpuA1->copyFrom(*cpuA);
cpuB1->copyFrom(*cpuB);
cpuC1->copyFrom(*cpuC);
if (!endCol) {
cpuA->addDotMulVMM(*cpuB, *cpuC);
gpuA->addDotMulVMM(*gpuB, *gpuC);
cpuA1->addDotMulVMM2(*cpuB1, *cpuC1);
MatrixCheckErr(*cpuA, *cpuA1);
} else {
MatrixPtr subCpuA = cpuA->subColMatrix(0, endCol);
MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
MatrixPtr subCpuC = cpuC->subColMatrix(0, endCol);
MatrixPtr subGpuA = gpuA->subColMatrix(0, endCol);
MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
MatrixPtr subGpuC = gpuC->subColMatrix(0, endCol);
subCpuA->addDotMulVMM(*subCpuB, *subCpuC);
subGpuA->addDotMulVMM(*subGpuB, *subGpuC);
}
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixRowSum(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, 1);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr cpuA1 = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr cpuB1 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA1 = std::make_shared<GpuMatrix>(height, 1);
MatrixPtr gpuB1 = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
cpuA1->copyFrom(*cpuA);
cpuB1->copyFrom(*cpuB);
gpuA1->copyFrom(*cpuA);
gpuB1->copyFrom(*cpuB);
cpuA->colMerge(*cpuB);
gpuA->colMerge(*gpuB);
cpuB1->rowSum(*cpuA1);
gpuB1->rowSum(*gpuA1);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, 1);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
outputCheck->copyFrom(*gpuA1);
MatrixCheckErr(*cpuA1, *outputCheck);
}
void testMatrixRowMax(int height, int width, int endCol = 0) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, 1);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
if (!endCol) {
cpuB->rowMax(*cpuA);
gpuB->rowMax(*gpuA);
} else {
MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
subCpuB->rowMax(*cpuA);
subGpuB->rowMax(*gpuA);
}
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, 1);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixColSum(int height, int width, int endCol = 0) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
if (!endCol) {
cpuA->accumulateColSum(*cpuB);
gpuA->accumulateColSum(*gpuB);
} else {
MatrixPtr subCpuA = cpuA->subColMatrix(0, endCol);
MatrixPtr subGpuA = gpuA->subColMatrix(0, endCol);
MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
subCpuA->accumulateColSum(*subCpuB);
subGpuA->accumulateColSum(*subGpuB);
}
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixColMax(int height, int width, int endCol = 0) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
if (!endCol) {
cpuB->colMax(*cpuA);
gpuB->colMax(*gpuA);
} else {
MatrixPtr subCpuA = cpuA->subColMatrix(0, endCol);
MatrixPtr subGpuA = gpuA->subColMatrix(0, endCol);
MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
subCpuB->colMax(*subCpuA);
subGpuB->colMax(*subGpuA);
}
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixCollectBias(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(1, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(1, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
real scale = 1.0f / (rand() % 10); // NOLINT
cpuA->collectBias(*cpuB, scale);
gpuA->collectBias(*gpuB, scale);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(1, width);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixSumOfSquares(int height, int width, int endCol = 0) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, 1);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, 1);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
if (!endCol) {
cpuA->sumOfSquares(*cpuB, *cpuC);
gpuA->sumOfSquares(*gpuB, *gpuC);
} else {
MatrixPtr subCpuB = cpuB->subColMatrix(0, endCol);
MatrixPtr subCpuC = cpuC->subColMatrix(0, endCol);
MatrixPtr subGpuB = gpuB->subColMatrix(0, endCol);
MatrixPtr subGpuC = gpuC->subColMatrix(0, endCol);
cpuA->sumOfSquares(*subCpuB, *subCpuC);
gpuA->sumOfSquares(*subGpuB, *subGpuC);
}
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, 1);
outputCheck->copyFrom(*gpuA);
MatrixCheckErr(*cpuA, *outputCheck);
}
void testMatrixBinaryClassificationError(int height, int width) {
MatrixPtr cpuA = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuA = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuB = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuC = std::make_shared<GpuMatrix>(height, width);
MatrixPtr cpuA2 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuB2 = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuC2 = std::make_shared<CpuMatrix>(height, width);
cpuA->randomizeUniform();
cpuB->randomizeUniform();
cpuC->randomizeUniform();
gpuA->copyFrom(*cpuA);
gpuB->copyFrom(*cpuB);
gpuC->copyFrom(*cpuC);
cpuA2->copyFrom(*cpuA);
cpuB2->copyFrom(*cpuB);
cpuC2->copyFrom(*cpuC);
real scale = 0.5;
int columnOffset = rand() % width; // NOLINT
cpuA->binaryClassificationError(columnOffset, *cpuB, *cpuC, scale);
gpuA->binaryClassificationError(columnOffset, *gpuB, *gpuC, scale);
cpuA2->binaryClassificationError2(columnOffset, *cpuB2, *cpuC2, scale);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width); outputCheck->mul(cpu, cpuI);
outputCheck->copyFrom(*gpuA); cpu->setDiag(1.0);
MatrixCheckErr(*cpuA, *outputCheck); TensorCheckErr(*cpu, *outputCheck);
MatrixCheckErr(*cpuA, *cpuA2);
}
TEST(Matrix, aggregate) {
for (auto height : {1, 11, 16, 32, 64, 73, 128, 200, 1024, 2345}) {
for (auto width : {1, 9, 16, 32, 64, 100, 512, 1000, 1024, 2453}) {
VLOG(3) << " height=" << height << " width=" << width;
testMatrixRowSum(height, width);
testMatrixRowMax(height, width);
testMatrixColSum(height, width);
testMatrixColMax(height, width);
testMatrixCollectBias(height, width);
testMatrixTernaryRowDotMul(height, width);
testMatrixAddDotMulVMM(height, width);
testMatrixSumOfSquares(height, width);
testMatrixBinaryClassificationError(height, width);
}
}
} }
TEST(Matrix, aggregate2) { TEST(Matrix, unary) {
for (auto height : {16, 32, 128, 512, 1024}) { for (auto height : {1, 3, 11, 73, 128, 200, 330}) {
for (auto width : for (auto width : {1, 3, 32, 100, 512, 1000, 3210}) {
{16, 32, 64, 128, 256, 512, 768, 1024, 2048, 3072, 4096}) {
VLOG(3) << " height=" << height << " width=" << width; VLOG(3) << " height=" << height << " width=" << width;
int endCol = rand() % width; // NOLINT testMatrixDeepSwap(height, width);
testMatrixRowMax(height, width, endCol); testMatrixZeroAtOffset(height, width);
testMatrixSumOfSquares(height, width, endCol); testMatrixGetSum(height, width);
testMatrixColSum(height, width, endCol); testMatrixTranspose(height, width);
testMatrixColMax(height, width, endCol);
testMatrixAddDotMulVMM(height, width, endCol);
} }
// inverse
testMatrixInverse(height);
} }
} }
void testMatrixAddAtOffset(int height, int width1, int width2) { void testMatrixSoftmax(int height, int width) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width1); MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width2); MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width1); MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width2); MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width);
cpuInput->randomizeUniform(); cpuInput->randomizeUniform();
gpuInput->copyFrom(*cpuInput); gpuInput->copyFrom(*cpuInput);
cpuOutput->randomizeUniform(); cpuOutput->zero();
gpuOutput->copyFrom(*cpuOutput); gpuOutput->zero();
cpuInput->softmax(*cpuOutput);
int columnOffset = 0; gpuInput->softmax(*gpuOutput);
int offset = std::abs(width1 - width2);
if (offset) {
columnOffset = rand() % offset; // NOLINT
}
cpuOutput->addAtOffset(*cpuInput, columnOffset);
gpuOutput->addAtOffset(*gpuInput, columnOffset);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width2); TensorCheckErr(*cpuOutput, *gpuOutput);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckEqual(*cpuOutput, *outputCheck);
} }
void testMatrixAssignAtOffset(int height, int width1, int width2) { void testSequenceSoftmax(int batchSize) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width1); // forward
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width2); int inputDim = 1;
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width1); MatrixPtr cpuInput = std::make_shared<CpuMatrix>(batchSize, inputDim);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width2); MatrixPtr gpuInput = std::make_shared<GpuMatrix>(batchSize, inputDim);
cpuInput->randomizeUniform(); cpuInput->randomizeUniform();
gpuInput->copyFrom(*cpuInput); gpuInput->copyFrom(*cpuInput);
cpuOutput->randomizeUniform();
gpuOutput->copyFrom(*cpuOutput);
int columnOffset = 0; IVectorPtr cpuSequence;
int offset = std::abs(width1 - width2); generateSequenceStartPositions(batchSize, cpuSequence);
if (offset) { IVectorPtr gpuSequence = IVector::create(cpuSequence->getSize(), true);
columnOffset = rand() % offset; // NOLINT gpuSequence->copyFrom(*cpuSequence);
}
cpuOutput->assignAtOffset(*cpuInput, columnOffset);
gpuOutput->assignAtOffset(*gpuInput, columnOffset);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width2); cpuInput->sequenceSoftmax(*cpuInput, *cpuSequence);
outputCheck->copyFrom(*gpuOutput); gpuInput->sequenceSoftmax(*gpuInput, *gpuSequence);
MatrixCheckEqual(*cpuOutput, *outputCheck);
TensorCheckErr(*cpuInput, *gpuInput);
} }
TEST(Matrix, AtOffset) { void testMatrixSoftmaxThreshold(int height, int width) {
for (auto height : {1, 11, 73, 128, 200}) { MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width);
for (auto width1 : {1, 32, 100, 512, 1000}) { MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width);
for (auto width2 : {1, 32, 100, 512, 1000}) { MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width);
VLOG(3) << " height=" << height << " width1=" << width1 MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width);
<< " width2=" << width2;
testMatrixAddAtOffset(height, width1, width2); cpuInput->randomizeUniform();
testMatrixAssignAtOffset(height, width1, width2); cpuInput->getData()[0] = 100.0;
} gpuInput->copyFrom(*cpuInput);
cpuOutput->zero();
gpuOutput->zero();
cpuInput->softmax(*cpuOutput);
gpuInput->softmax(*gpuOutput);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(height, width);
outputCheck->copyFrom(*gpuOutput);
// check output zero
int cpuCount = 0;
int gpuCount = 0;
auto zeroNum = [](MatrixPtr out, int& count) {
for (size_t i = 0; i < out->getHeight(); i++) {
for (size_t j = 0; j < out->getWidth(); j++) {
if (out->getElement(i, j) == 0) count++;
} }
} }
};
zeroNum(cpuOutput, cpuCount);
zeroNum(outputCheck, gpuCount);
EXPECT_EQ(cpuCount, 0) << "Cpu softmax output value 0";
EXPECT_EQ(gpuCount, 0) << "Gpu softmax output value 0";
} }
void testMatrixSelectRows(int numSamples, int tableSize, int inputDim) { void testMatrixSoftmaxBp(int height, int width) {
MatrixPtr cpuTable = std::make_shared<CpuMatrix>(tableSize, inputDim); MatrixPtr cpuInput = std::make_shared<CpuMatrix>(height, width);
MatrixPtr gpuTable = std::make_shared<GpuMatrix>(tableSize, inputDim); MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(height, width);
cpuTable->randomizeUniform(); MatrixPtr gpuInput = std::make_shared<GpuMatrix>(height, width);
gpuTable->copyFrom(*cpuTable); MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(height, width);
IVectorPtr cpuIds;
IVectorPtr gpuIds;
cpuIds = VectorT<int>::create(numSamples, false);
gpuIds = VectorT<int>::create(numSamples, true);
cpuIds->rand(tableSize);
gpuIds->copyFrom(*cpuIds);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(numSamples, inputDim); cpuInput->randomizeUniform();
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(numSamples, inputDim); gpuInput->copyFrom(*cpuInput);
cpuOutput->randomizeUniform(); cpuOutput->randomizeUniform();
gpuOutput->copyFrom(*cpuOutput); gpuOutput->copyFrom(*cpuOutput);
gpuOutput->softmaxBackward(*gpuInput);
cpuOutput->selectRows(*cpuTable, *cpuIds); MatrixPtr sftMaxSum = std::make_shared<CpuMatrix>(height, 1);
gpuOutput->selectRows(*gpuTable, *gpuIds); MatrixPtr sftMaxDot = std::make_shared<CpuMatrix>(height, width);
sftMaxDot->dotMul(*cpuOutput, *cpuInput);
sftMaxSum->colMerge(*sftMaxDot);
cpuOutput->softmaxDerivative(*cpuInput, *sftMaxSum);
// check TensorCheckErr(*cpuOutput, *gpuOutput);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(numSamples, inputDim); }
outputCheck->copyFrom(*gpuOutput);
MatrixCheckEqual(*cpuOutput, *outputCheck); TEST(Matrix, softmax) {
for (auto height : {1, 11, 73, 128, 200}) {
for (auto width : {1, 32, 100, 512, 1000}) {
VLOG(3) << " height=" << height << " width=" << width;
testMatrixSoftmax(height, width);
testMatrixSoftmaxBp(height, width);
testMatrixSoftmaxThreshold(height, width);
}
testSequenceSoftmax(height);
}
} }
void testMatrixAddToRows(int numSamples, int tableSize, int inputDim) { void testMatrixAddToRows(int numSamples, int tableSize, int inputDim) {
...@@ -1322,10 +470,7 @@ void testMatrixAddToRows(int numSamples, int tableSize, int inputDim) { ...@@ -1322,10 +470,7 @@ void testMatrixAddToRows(int numSamples, int tableSize, int inputDim) {
cpuOutput->addToRows(*cpuTable, *cpuIds); cpuOutput->addToRows(*cpuTable, *cpuIds);
gpuOutput->addToRows(*gpuTable, *gpuIds); gpuOutput->addToRows(*gpuTable, *gpuIds);
// check TensorCheckErr(*cpuTable, *gpuTable);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(tableSize, inputDim);
outputCheck->copyFrom(*gpuTable);
MatrixCheckErr(*cpuTable, *outputCheck);
} }
TEST(Matrix, tableProjection) { TEST(Matrix, tableProjection) {
...@@ -1334,7 +479,6 @@ TEST(Matrix, tableProjection) { ...@@ -1334,7 +479,6 @@ TEST(Matrix, tableProjection) {
for (auto inputDim : {20, 50}) { for (auto inputDim : {20, 50}) {
VLOG(3) << " numSamples=" << numSamples << " tableSize=" << tableSize VLOG(3) << " numSamples=" << numSamples << " tableSize=" << tableSize
<< " inputDim=" << inputDim; << " inputDim=" << inputDim;
testMatrixSelectRows(numSamples, tableSize, inputDim);
testMatrixAddToRows(numSamples, tableSize, inputDim); testMatrixAddToRows(numSamples, tableSize, inputDim);
} }
} }
...@@ -1368,9 +512,7 @@ void testMatrixMul(bool transa, bool transb, int dimM, int dimN, int dimK) { ...@@ -1368,9 +512,7 @@ void testMatrixMul(bool transa, bool transb, int dimM, int dimN, int dimK) {
cpuC->mul(cpuA, cpuB, alpha, beta); cpuC->mul(cpuA, cpuB, alpha, beta);
gpuC->mul(gpuA, gpuB, alpha, beta); gpuC->mul(gpuA, gpuB, alpha, beta);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(heightC, widthC); TensorCheckErr(*cpuC, *gpuC);
outputCheck->copyFrom(*gpuC);
MatrixCheckErr(*cpuC, *outputCheck);
} }
void testSubMatrixMul(bool transa, bool transb, int dimM, int dimN, int dimK) { void testSubMatrixMul(bool transa, bool transb, int dimM, int dimN, int dimK) {
...@@ -1442,9 +584,7 @@ void testSubMatrixMul(bool transa, bool transb, int dimM, int dimN, int dimK) { ...@@ -1442,9 +584,7 @@ void testSubMatrixMul(bool transa, bool transb, int dimM, int dimN, int dimK) {
subCpuC->mul(subCpuA, subCpuB, alpha, beta); subCpuC->mul(subCpuA, subCpuB, alpha, beta);
subGpuC->mul(subGpuA, subGpuB, alpha, beta); subGpuC->mul(subGpuA, subGpuB, alpha, beta);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(heightC, widthC); TensorCheckErr(*cpuC, *gpuC);
outputCheck->copyFrom(*gpuC);
MatrixCheckErr(*cpuC, *outputCheck);
} }
TEST(Matrix, mul) { TEST(Matrix, mul) {
...@@ -1498,9 +638,7 @@ void testVectorReset(int size) { ...@@ -1498,9 +638,7 @@ void testVectorReset(int size) {
cpu->reset(value); cpu->reset(value);
gpu->reset(value); gpu->reset(value);
std::shared_ptr<CpuVectorT<T>> out = std::make_shared<CpuVectorT<T>>(size); TensorCheckEqual(*cpu, *gpu);
out->copyFrom(*gpu);
VectorCheckEqual(*cpu, *out);
} }
template <class T> template <class T>
...@@ -1526,9 +664,7 @@ void testVecortSelectFrom(int size) { ...@@ -1526,9 +664,7 @@ void testVecortSelectFrom(int size) {
cpuDst->selectFrom(*cpuSrc, *cpuIds); cpuDst->selectFrom(*cpuSrc, *cpuIds);
gpuDst->selectFrom(*gpuSrc, *gpuIds); gpuDst->selectFrom(*gpuSrc, *gpuIds);
std::shared_ptr<CpuVectorT<T>> out = std::make_shared<CpuVectorT<T>>(size); TensorCheckEqual(*cpuDst, *gpuDst);
out->copyFrom(*gpuDst);
VectorCheckEqual(*cpuDst, *out);
} }
template <class T> template <class T>
...@@ -1539,9 +675,7 @@ void testVecotrZeroMem(int size) { ...@@ -1539,9 +675,7 @@ void testVecotrZeroMem(int size) {
cpu->zeroMem(); cpu->zeroMem();
gpu->zeroMem(); gpu->zeroMem();
std::shared_ptr<CpuVectorT<T>> out = std::make_shared<CpuVectorT<T>>(size); TensorCheckEqual(*cpu, *gpu);
out->copyFrom(*gpu);
VectorCheckEqual(*cpu, *out);
} }
template <class T> template <class T>
...@@ -1562,9 +696,7 @@ void testVectorIsEqual(int size) { ...@@ -1562,9 +696,7 @@ void testVectorIsEqual(int size) {
cpuA->isEqualTo(*cpuB, value); cpuA->isEqualTo(*cpuB, value);
gpuA->isEqualTo(*gpuB, value); gpuA->isEqualTo(*gpuB, value);
std::shared_ptr<CpuVectorT<T>> out = std::make_shared<CpuVectorT<T>>(size); TensorCheckEqual(*cpuA, *gpuA);
out->copyFrom(*gpuA);
VectorCheckEqual(*cpuA, *out);
} }
TEST(Vector, Equal) { TEST(Vector, Equal) {
...@@ -1595,9 +727,7 @@ void testMatrixTopK(int samples, int dim, int beamSize) { ...@@ -1595,9 +727,7 @@ void testMatrixTopK(int samples, int dim, int beamSize) {
cpuSrc->rowMax(*cpuIds, *cpuVal); cpuSrc->rowMax(*cpuIds, *cpuVal);
gpuSrc->rowMax(*gpuIds, *gpuVal); gpuSrc->rowMax(*gpuIds, *gpuVal);
MatrixPtr outVal = std::make_shared<CpuMatrix>(samples, beamSize); TensorCheckEqual(*cpuVal, *gpuVal);
outVal->copyFrom(*gpuVal);
MatrixCheckEqual(*cpuVal, *outVal);
} }
TEST(Matrix, topK) { TEST(Matrix, topK) {
...@@ -1633,9 +763,7 @@ void testSMatrixTopK(int samples, int dim, int beamSize, real ratio) { ...@@ -1633,9 +763,7 @@ void testSMatrixTopK(int samples, int dim, int beamSize, real ratio) {
cpuSrc->rowMax(*cpuIds, *cpuVal); cpuSrc->rowMax(*cpuIds, *cpuVal);
gpuSrc->rowMax(*gpuIds, *gpuVal); gpuSrc->rowMax(*gpuIds, *gpuVal);
MatrixPtr outCheckMaxVal = std::make_shared<CpuMatrix>(samples, beamSize); TensorCheckEqual(*cpuVal, *gpuVal);
outCheckMaxVal->copyFrom(*gpuVal);
MatrixCheckEqual(*cpuVal, *outCheckMaxVal);
IVectorPtr outCheckIds = std::make_shared<CpuIVector>(samples * beamSize); IVectorPtr outCheckIds = std::make_shared<CpuIVector>(samples * beamSize);
outCheckIds->copyFrom(*gpuIds); outCheckIds->copyFrom(*gpuIds);
...@@ -1665,42 +793,6 @@ TEST(SMatrix, topK) { ...@@ -1665,42 +793,6 @@ TEST(SMatrix, topK) {
} }
} }
void testMatrixCopyByRowIndex(int outHeight, int inHeight, int width) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(inHeight, width);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(inHeight, width);
cpuInput->randomizeUniform();
gpuInput->copyFrom(*cpuInput);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(outHeight, width);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(outHeight, width);
cpuOutput->zero();
gpuOutput->zero();
IVectorPtr cpuRowIndex = IVector::create(outHeight, false);
IVectorPtr gpuRowIndex = IVector::create(outHeight, true);
cpuRowIndex->rand(inHeight);
gpuRowIndex->copyFrom(*cpuRowIndex);
cpuOutput->copyByRowIndex(*cpuInput, *cpuRowIndex);
gpuOutput->copyByRowIndex(*gpuInput, *gpuRowIndex);
// check
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(outHeight, width);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckEqual(*cpuOutput, *outputCheck);
}
TEST(Matrix, copyByRowIndex) {
for (auto outHeight : {31, 500, 1000}) {
for (auto inHeight : {17, 257, 500, 1200}) {
for (auto width : {512, 1024}) {
VLOG(3) << outHeight << " " << inHeight << " " << width;
testMatrixCopyByRowIndex(outHeight, inHeight, width);
}
}
}
}
void testMatrixSequenceAvgForward(int batchSize, int inputDim, int mode) { void testMatrixSequenceAvgForward(int batchSize, int inputDim, int mode) {
MatrixPtr cpuInput = std::make_shared<CpuMatrix>(batchSize, inputDim); MatrixPtr cpuInput = std::make_shared<CpuMatrix>(batchSize, inputDim);
MatrixPtr gpuInput = std::make_shared<GpuMatrix>(batchSize, inputDim); MatrixPtr gpuInput = std::make_shared<GpuMatrix>(batchSize, inputDim);
...@@ -1721,10 +813,7 @@ void testMatrixSequenceAvgForward(int batchSize, int inputDim, int mode) { ...@@ -1721,10 +813,7 @@ void testMatrixSequenceAvgForward(int batchSize, int inputDim, int mode) {
cpuOutput->sequenceAvgForward(*cpuInput, *cpuSequence, mode); cpuOutput->sequenceAvgForward(*cpuInput, *cpuSequence, mode);
gpuOutput->sequenceAvgForward(*gpuInput, *gpuSequence, mode); gpuOutput->sequenceAvgForward(*gpuInput, *gpuSequence, mode);
// check TensorCheckErr(*cpuOutput, *gpuOutput);
MatrixPtr outputCheck = std::make_shared<CpuMatrix>(newBatchSize, inputDim);
outputCheck->copyFrom(*gpuOutput);
MatrixCheckErr(*cpuOutput, *outputCheck);
} }
TEST(Matrix, sequenceAvgForward) { TEST(Matrix, sequenceAvgForward) {
...@@ -1739,45 +828,6 @@ TEST(Matrix, sequenceAvgForward) { ...@@ -1739,45 +828,6 @@ TEST(Matrix, sequenceAvgForward) {
} }
} }
void testCosSim(int heightX, int heightY, int width, real scale) {
MatrixPtr prevOutX = CpuMatrix::create(heightX, width, false, false);
MatrixPtr prevOutY = CpuMatrix::create(heightY, width, false, false);
MatrixPtr output = CpuMatrix::create(heightX, 1, false, false);
prevOutX->randomizeUniform();
prevOutY->randomizeUniform();
prevOutX->add(-0.5);
prevOutY->add(-0.5);
output->randomizeUniform();
MatrixPtr prevOutXGpu = GpuMatrix::create(heightX, width, false, true);
MatrixPtr prevOutYGpu = GpuMatrix::create(heightY, width, false, true);
MatrixPtr outputGpu = GpuMatrix::create(heightX, 1, false, true);
prevOutXGpu->copyFrom(*prevOutX);
prevOutYGpu->copyFrom(*prevOutY);
outputGpu->copyFrom(*output);
output->cosSim(*prevOutX, *prevOutY, scale);
outputGpu->cosSim(*prevOutXGpu, *prevOutYGpu, scale);
MatrixPtr outputCheck = CpuMatrix::create(heightX, 1, false, false);
outputCheck->copyFrom(*outputGpu);
MatrixCheckErr(*output, *outputCheck);
}
TEST(Matrix, cosSim) {
for (auto heightX : {10, 100, 1000}) {
for (auto heightY : {1, heightX}) {
for (auto width : {10, 100, 1000}) {
for (auto scale : {1.0, 2.0}) {
testCosSim(heightX, heightY, width, scale);
}
}
}
}
}
void testCosSimDerivate(int heightX, int heightY, int width, real scale) { void testCosSimDerivate(int heightX, int heightY, int width, real scale) {
MatrixPtr prevOutX = CpuMatrix::create(heightX, width, false, false); MatrixPtr prevOutX = CpuMatrix::create(heightX, width, false, false);
MatrixPtr prevOutY = CpuMatrix::create(heightY, width, false, false); MatrixPtr prevOutY = CpuMatrix::create(heightY, width, false, false);
...@@ -1817,12 +867,8 @@ void testCosSimDerivate(int heightX, int heightY, int width, real scale) { ...@@ -1817,12 +867,8 @@ void testCosSimDerivate(int heightX, int heightY, int width, real scale) {
*prevGradYGpu, *prevGradYGpu,
scale); scale);
MatrixPtr prevGradXCheck = CpuMatrix::create(heightX, width, false, false); TensorCheckErr(*prevGradX, *prevGradXGpu);
MatrixPtr prevGradYCheck = CpuMatrix::create(heightY, width, false, false); TensorCheckErr(*prevGradY, *prevGradYGpu);
prevGradXCheck->copyFrom(*prevGradXGpu);
prevGradYCheck->copyFrom(*prevGradYGpu);
MatrixCheckErr(*prevGradX, *prevGradXCheck);
MatrixCheckErr(*prevGradY, *prevGradYCheck);
} }
TEST(Matrix, cosSimDerivate) { TEST(Matrix, cosSimDerivate) {
...@@ -1837,80 +883,6 @@ TEST(Matrix, cosSimDerivate) { ...@@ -1837,80 +883,6 @@ TEST(Matrix, cosSimDerivate) {
} }
} }
void testParamReluForward(int height, int width, int w_height, int w_width) {
MatrixPtr output = CpuMatrix::create(height, width, false, false);
MatrixPtr input = CpuMatrix::create(height, width, false, false);
MatrixPtr w = CpuMatrix::create(w_height, w_width, false, false);
output->randomizeUniform();
input->randomizeUniform();
w->randomizeUniform();
input->add(-0.5);
MatrixPtr outputGpu = GpuMatrix::create(height, width, false, true);
MatrixPtr inputGpu = GpuMatrix::create(height, width, false, true);
MatrixPtr wGpu = GpuMatrix::create(w_height, w_width, false, true);
inputGpu->copyFrom(*input);
wGpu->copyFrom(*w);
output->paramReluForward(*input, *w);
outputGpu->paramReluForward(*inputGpu, *wGpu);
MatrixPtr outputCheck = CpuMatrix::create(height, width, false, false);
outputCheck->copyFrom(*outputGpu);
MatrixCheckEqual(*output, *outputCheck);
}
TEST(Matrix, paramReluForward) {
for (auto height : {10, 100}) {
for (auto width : {10, 100}) {
for (auto w_height : {1, 2}) {
for (auto w_width : {1, 2}) {
testParamReluForward(height, width, w_height, w_width);
}
}
}
}
}
void testParamReluBackwardW(int height, int width, int w_height, int w_width) {
MatrixPtr oGrad = CpuMatrix::create(height, width, false, false);
MatrixPtr input = CpuMatrix::create(height, width, false, false);
MatrixPtr w = CpuMatrix::create(w_height, w_width, false, false);
oGrad->randomizeUniform();
input->randomizeUniform();
w->randomizeUniform();
input->add(-0.5);
MatrixPtr oGradGpu = GpuMatrix::create(height, width, false, true);
MatrixPtr inputGpu = GpuMatrix::create(height, width, false, true);
MatrixPtr wGpu = GpuMatrix::create(w_height, w_width, false, true);
oGradGpu->copyFrom(*oGrad);
inputGpu->copyFrom(*input);
wGpu->copyFrom(*w);
w->paramReluBackwardW(*oGrad, *input);
wGpu->paramReluBackwardW(*oGradGpu, *inputGpu);
MatrixPtr wCheck = CpuMatrix::create(w_height, w_width, false, false);
wCheck->copyFrom(*wGpu);
MatrixCheckErr(*w, *wCheck);
}
TEST(Matrix, paramReluBackwardW) {
for (auto height : {10, 100}) {
for (auto width : {10, 100}) {
for (auto w_height : {1, 2}) {
for (auto w_width : {1, 2}) {
testParamReluBackwardW(height, width, w_height, w_width);
}
}
}
}
}
void testParamReluBackwardDiff(int height, void testParamReluBackwardDiff(int height,
int width, int width,
int w_height, int w_height,
...@@ -1939,9 +911,7 @@ void testParamReluBackwardDiff(int height, ...@@ -1939,9 +911,7 @@ void testParamReluBackwardDiff(int height,
diff->paramReluBackwardDiff(*oGrad, *input, *w); diff->paramReluBackwardDiff(*oGrad, *input, *w);
diffGpu->paramReluBackwardDiff(*oGradGpu, *inputGpu, *wGpu); diffGpu->paramReluBackwardDiff(*oGradGpu, *inputGpu, *wGpu);
MatrixPtr diffCheck = CpuMatrix::create(height, width, false, false); TensorCheckErr(*diff, *diffGpu);
diffCheck->copyFrom(*diffGpu);
MatrixCheckErr(*diff, *diffCheck);
} }
TEST(Matrix, paramReluBackwardDiff) { TEST(Matrix, paramReluBackwardDiff) {
...@@ -1972,9 +942,7 @@ void testClassificationError(int numSamples, int dim) { ...@@ -1972,9 +942,7 @@ void testClassificationError(int numSamples, int dim) {
cpuError->classificationError(cpuOutput, cpuLabel); cpuError->classificationError(cpuOutput, cpuLabel);
gpuError->classificationError(gpuOutput, gpuLabel); gpuError->classificationError(gpuOutput, gpuLabel);
MatrixPtr check = std::make_shared<CpuMatrix>(numSamples, 1); TensorCheckEqual(*cpuError, *gpuError);
check->copyFrom(*gpuError);
MatrixCheckEqual(*cpuError, *check);
} }
TEST(Matrix, classificationError) { TEST(Matrix, classificationError) {
...@@ -2139,9 +1107,8 @@ void testAvgPoolFwdBwd(int numSamples, ...@@ -2139,9 +1107,8 @@ void testAvgPoolFwdBwd(int numSamples,
outW, outW,
padH, padH,
padW); padW);
MatrixPtr targetCheck = CpuMatrix::create(numSamples, outWidth, false, false);
targetCheck->copyFrom(*targetGpu); TensorCheckErr(*target, *targetGpu);
MatrixCheckErr(*target, *targetCheck);
MatrixPtr inputGrad = CpuMatrix::create(numSamples, inWidth, false, false); MatrixPtr inputGrad = CpuMatrix::create(numSamples, inWidth, false, false);
MatrixPtr inputGpuGrad = GpuMatrix::create(numSamples, inWidth, false, true); MatrixPtr inputGpuGrad = GpuMatrix::create(numSamples, inWidth, false, true);
...@@ -2180,10 +1147,8 @@ void testAvgPoolFwdBwd(int numSamples, ...@@ -2180,10 +1147,8 @@ void testAvgPoolFwdBwd(int numSamples,
1.0, 1.0,
padH, padH,
padW); padW);
MatrixPtr targetBwdCheck =
CpuMatrix::create(numSamples, inWidth, false, false); TensorCheckErr(*inputGrad, *inputGpuGrad);
targetBwdCheck->copyFrom(*inputGpuGrad);
MatrixCheckErr(*inputGrad, *targetBwdCheck);
} }
TEST(Matrix, PoolFwdBwd) { TEST(Matrix, PoolFwdBwd) {
...@@ -2248,11 +1213,9 @@ void testMaxOutFwdBwd( ...@@ -2248,11 +1213,9 @@ void testMaxOutFwdBwd(
MatrixPtr target = CpuMatrix::create(numSamples, outWidth, false, false); MatrixPtr target = CpuMatrix::create(numSamples, outWidth, false, false);
MatrixPtr targetGpu = GpuMatrix::create(numSamples, outWidth, false, true); MatrixPtr targetGpu = GpuMatrix::create(numSamples, outWidth, false, true);
MatrixPtr targetCheck = CpuMatrix::create(numSamples, outWidth, false, false);
IVectorPtr id = CpuIVector::create(numSamples * outWidth, false); IVectorPtr id = CpuIVector::create(numSamples * outWidth, false);
IVectorPtr idGpu = GpuIVector::create(numSamples * outWidth, true); IVectorPtr idGpu = GpuIVector::create(numSamples * outWidth, true);
IVectorPtr idCheck = CpuIVector::create(numSamples * outWidth, false);
input->randomizeUniform(); input->randomizeUniform();
inputGpu->copyFrom(*input); inputGpu->copyFrom(*input);
...@@ -2260,11 +1223,8 @@ void testMaxOutFwdBwd( ...@@ -2260,11 +1223,8 @@ void testMaxOutFwdBwd(
target->maxoutForward(*input, *id, outChannels, groups); target->maxoutForward(*input, *id, outChannels, groups);
targetGpu->maxoutForward(*inputGpu, *idGpu, outChannels, groups); targetGpu->maxoutForward(*inputGpu, *idGpu, outChannels, groups);
// check TensorCheckErr(*target, *targetGpu);
targetCheck->copyFrom(*targetGpu); TensorCheckEqual(*id, *idGpu);
MatrixCheckErr(*target, *targetCheck);
idCheck->copyFrom(*idGpu);
VectorCheckEqual(*id, *idCheck);
// backward // backward
MatrixPtr inputGrad = CpuMatrix::create(numSamples, inWidth, false, false); MatrixPtr inputGrad = CpuMatrix::create(numSamples, inWidth, false, false);
...@@ -2273,8 +1233,6 @@ void testMaxOutFwdBwd( ...@@ -2273,8 +1233,6 @@ void testMaxOutFwdBwd(
MatrixPtr targetGrad = CpuMatrix::create(numSamples, outWidth, false, false); MatrixPtr targetGrad = CpuMatrix::create(numSamples, outWidth, false, false);
MatrixPtr targetGpuGrad = MatrixPtr targetGpuGrad =
GpuMatrix::create(numSamples, outWidth, false, true); GpuMatrix::create(numSamples, outWidth, false, true);
MatrixPtr targetCheckGrad =
CpuMatrix::create(numSamples, inWidth, false, false);
inputGrad->randomizeUniform(); inputGrad->randomizeUniform();
targetGrad->randomizeUniform(); targetGrad->randomizeUniform();
...@@ -2284,9 +1242,7 @@ void testMaxOutFwdBwd( ...@@ -2284,9 +1242,7 @@ void testMaxOutFwdBwd(
inputGrad->maxoutBackward(*targetGrad, *id, outChannels, groups); inputGrad->maxoutBackward(*targetGrad, *id, outChannels, groups);
inputGpuGrad->maxoutBackward(*targetGpuGrad, *idGpu, outChannels, groups); inputGpuGrad->maxoutBackward(*targetGpuGrad, *idGpu, outChannels, groups);
// check TensorCheckErr(*inputGrad, *inputGpuGrad);
targetCheckGrad->copyFrom(*inputGpuGrad);
MatrixCheckErr(*inputGrad, *targetCheckGrad);
} }
TEST(Matrix, MaxOutFwdBwd) { TEST(Matrix, MaxOutFwdBwd) {
...@@ -2306,113 +1262,6 @@ TEST(Matrix, MaxOutFwdBwd) { ...@@ -2306,113 +1262,6 @@ TEST(Matrix, MaxOutFwdBwd) {
} }
} }
void testAddSharedBias(int numSamples, int dim, int channel) {
MatrixPtr cpuData = std::make_shared<CpuMatrix>(numSamples, dim);
MatrixPtr gpuData = std::make_shared<GpuMatrix>(numSamples, dim);
MatrixPtr cpuBias = std::make_shared<CpuMatrix>(1, channel);
MatrixPtr gpuBias = std::make_shared<GpuMatrix>(1, channel);
cpuData->randomizeUniform();
gpuData->copyFrom(*cpuData);
cpuBias->randomizeUniform();
gpuBias->copyFrom(*cpuBias);
cpuData->addSharedBias(*cpuBias, 1.0);
gpuData->addSharedBias(*gpuBias, 1.0);
MatrixPtr check = std::make_shared<CpuMatrix>(numSamples, dim);
check->copyFrom(*gpuData);
MatrixCheckErr(*cpuData, *check);
}
void testCollectSharedBias(int numSamples, int dim, int channel) {
MatrixPtr cpuData = std::make_shared<CpuMatrix>(numSamples, dim);
MatrixPtr gpuData = std::make_shared<GpuMatrix>(numSamples, dim);
MatrixPtr cpuBias = std::make_shared<CpuMatrix>(1, channel);
MatrixPtr gpuBias = std::make_shared<GpuMatrix>(1, channel);
cpuData->randomizeUniform();
gpuData->copyFrom(*cpuData);
cpuBias->randomizeUniform();
gpuBias->copyFrom(*cpuBias);
cpuBias->collectSharedBias(*cpuData, 1.0);
gpuBias->collectSharedBias(*gpuData, 1.0);
MatrixPtr check = std::make_shared<CpuMatrix>(1, channel);
check->copyFrom(*gpuBias);
MatrixCheckErr(*cpuBias, *check);
}
TEST(Matrix, sharedBias) {
for (auto numSamples : {1, 100, 520}) {
for (auto dim : {100 * 16, 100 * 32}) {
for (auto channel : {8, 16}) {
VLOG(3) << " numSamples=" << numSamples << " dim=" << dim
<< " channel=" << channel;
testAddSharedBias(numSamples, dim, channel);
testCollectSharedBias(numSamples, dim, channel);
}
}
}
}
void testMultiBinaryLabelCrossEntropy(int numSamples, int dim) {
MatrixPtr output = std::make_shared<CpuMatrix>(numSamples, dim);
MatrixPtr cpuOutput = std::make_shared<CpuMatrix>(numSamples, dim);
MatrixPtr gpuOutput = std::make_shared<GpuMatrix>(numSamples, dim);
MatrixPtr cpuEntropy = std::make_shared<CpuMatrix>(numSamples, 1);
MatrixPtr gpuEntropy = std::make_shared<GpuMatrix>(numSamples, 1);
MatrixPtr cpuGrad = std::make_shared<CpuMatrix>(numSamples, dim);
MatrixPtr gpuGrad = std::make_shared<GpuMatrix>(numSamples, dim);
MatrixPtr cpuLabel = std::make_shared<CpuSparseMatrix>(
numSamples, dim, numSamples, NO_VALUE, SPARSE_CSR, false);
MatrixPtr gpuLabel = std::make_shared<GpuSparseMatrix>(
numSamples, dim, numSamples, NO_VALUE, SPARSE_CSR, false);
for (int i = 0; i < numSamples; i++) {
const unsigned int id = rand() % dim; // NOLINT
cpuLabel->setRow(i, 1, &id, nullptr);
gpuLabel->setRow(i, 1, &id, nullptr);
}
output->randomizeUniform();
cpuOutput->zeroMem();
output->softmax(*cpuOutput);
gpuOutput->copyFrom(*cpuOutput);
cpuEntropy->zeroMem();
gpuEntropy->zeroMem();
cpuEntropy->multiBinaryLabelCrossEntropy(*cpuOutput, *cpuLabel);
gpuEntropy->multiBinaryLabelCrossEntropy(*gpuOutput, *gpuLabel);
MatrixPtr check1 = std::make_shared<CpuMatrix>(numSamples, 1);
check1->copyFrom(*gpuEntropy);
MatrixCheckErr(*cpuEntropy, *check1);
cpuGrad->zeroMem();
gpuGrad->zeroMem();
cpuGrad->multiBinaryLabelCrossEntropyBp(*cpuOutput, *cpuLabel);
gpuGrad->multiBinaryLabelCrossEntropyBp(*gpuOutput, *gpuLabel);
MatrixPtr check2 = std::make_shared<CpuMatrix>(numSamples, dim);
check2->copyFrom(*gpuGrad);
MatrixCheckErr(*cpuGrad, *check2);
}
TEST(Matrix, multiBinaryCrossEntropy) {
for (auto numSamples : {100, 1000, 10000}) {
for (auto dim : {100, 1000, 10000}) {
VLOG(3) << " numSamples=" << numSamples << " dim=" << dim;
testMultiBinaryLabelCrossEntropy(numSamples, dim);
}
}
}
int main(int argc, char** argv) { int main(int argc, char** argv) {
testing::InitGoogleTest(&argc, argv); testing::InitGoogleTest(&argc, argv);
initMain(argc, argv); initMain(argc, argv);
......
paddle/parameter/Argument.cpp
浏览文件 @ 021b3a44
...@@ -225,6 +225,8 @@ void Argument::resizeAndCopyFrom(const Argument& src, ...@@ -225,6 +225,8 @@ void Argument::resizeAndCopyFrom(const Argument& src,
} }
resizeAndCopy(udp, src.udp, useGpu, stream); resizeAndCopy(udp, src.udp, useGpu, stream);
resizeAndCopy(strs, src.strs, useGpu, stream); resizeAndCopy(strs, src.strs, useGpu, stream);
frameWidth = src.frameWidth;
frameHeight = src.frameHeight;
} }
int32_t Argument::resizeAndCopyFrom(const Argument& src, int32_t Argument::resizeAndCopyFrom(const Argument& src,
......
paddle/py_paddle/util.py
浏览文件 @ 021b3a44
...@@ -559,10 +559,10 @@ def __monkey_patch_trainer__(): ...@@ -559,10 +559,10 @@ def __monkey_patch_trainer__():
def monkeypatches(): def monkeypatches():
patches = [ patches = [__monkeypatch_init_paddle__,
__monkeypatch_init_paddle__, __monkeypatch_gradient_machine__, __monkeypatch_gradient_machine__,
__monkey_patch_protobuf_objects__, __monkey_patch_parameter__, __monkey_patch_protobuf_objects__,
__monkey_patch_trainer__ __monkey_patch_parameter__,
] __monkey_patch_trainer__]
for patch in patches: for patch in patches:
patch() patch()
paddle/scripts/docker/Dockerfile 0 → 100644
浏览文件 @ 021b3a44
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Authors <paddle-dev@baidu.com>
RUN apt-get update && \
apt-get install -y cmake libprotobuf-dev protobuf-compiler git \
libgoogle-glog-dev libgflags-dev libatlas-dev libatlas3-base g++ m4 python-pip \
python-protobuf python-numpy python-dev swig openssh-server \
wget unzip python-matplotlib tar xz-utils bzip2 gzip coreutils \
sed grep graphviz libjpeg-dev zlib1g-dev doxygen && \
apt-get clean -y
RUN pip install BeautifulSoup docopt PyYAML pillow \
'sphinx>=1.4.0' sphinx_rtd_theme breathe recommonmark
ARG WITH_AVX
ENV WITH_AVX=${WITH_AVX:-ON}
ENV WITH_GPU=OFF
RUN mkdir /paddle
COPY . /paddle/
RUN /paddle/paddle/scripts/docker/build.sh
RUN echo 'export LD_LIBRARY_PATH=/usr/lib64:${LD_LIBRARY_PATH}' >> /etc/profile
RUN pip install /usr/local/opt/paddle/share/wheels/*.whl
RUN paddle version # print version after build
# Configure OpenSSH server. c.f. https://docs.docker.com/engine/examples/running_ssh_service
RUN mkdir /var/run/sshd
RUN echo 'root:root' | chpasswd
RUN sed -ri 's/^PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config
RUN sed -ri 's/UsePAM yes/#UsePAM yes/g' /etc/ssh/sshd_config
EXPOSE 22
CMD ["/usr/sbin/sshd", "-D"]
paddle/scripts/docker/Dockerfile.cpu 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=OFF
ENV IS_DEVEL=OFF
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=ON
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.cpu-demo 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=OFF
ENV IS_DEVEL=ON
ENV WITH_DEMO=ON
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=ON
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.cpu-devel 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=OFF
ENV IS_DEVEL=ON
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=ON
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.cpu-noavx 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=OFF
ENV IS_DEVEL=OFF
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=OFF
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.cpu-noavx-demo 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=OFF
ENV IS_DEVEL=ON
ENV WITH_DEMO=ON
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=OFF
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.cpu-noavx-devel 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM ubuntu:14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=OFF
ENV IS_DEVEL=ON
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=OFF
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.gpu
浏览文件 @ 021b3a44
FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com> MAINTAINER PaddlePaddle Authors <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0 RUN apt-get update && \
apt-get install -y cmake libprotobuf-dev protobuf-compiler git \
libgoogle-glog-dev libgflags-dev libatlas-dev libatlas3-base g++ m4 python-pip \
python-protobuf python-numpy python-dev swig openssh-server \
wget unzip python-matplotlib tar xz-utils bzip2 gzip coreutils \
sed grep graphviz libjpeg-dev zlib1g-dev doxygen && \
apt-get clean -y
RUN pip install BeautifulSoup docopt PyYAML pillow \
'sphinx>=1.4.0' sphinx_rtd_theme breathe recommonmark
ARG WITH_AVX
ENV WITH_AVX=${WITH_AVX:-ON}
ENV WITH_GPU=ON ENV WITH_GPU=ON
ENV IS_DEVEL=OFF
ENV WITH_DEMO=OFF RUN mkdir /paddle
ENV PIP_INSTALL_ARGS "" COPY . /paddle/
ENV PIP_GENERAL_ARGS "" RUN /paddle/paddle/scripts/docker/build.sh
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=ON RUN echo 'export LD_LIBRARY_PATH=/usr/lib64:${LD_LIBRARY_PATH}' >> /etc/profile
RUN cd /root/ && bash build.sh RUN pip install /usr/local/opt/paddle/share/wheels/*.whl
RUN paddle version # print version after build
# Configure OpenSSH server. c.f. https://docs.docker.com/engine/examples/running_ssh_service
RUN mkdir /var/run/sshd
RUN echo 'root:root' | chpasswd
RUN sed -ri 's/^PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config
RUN sed -ri 's/UsePAM yes/#UsePAM yes/g' /etc/ssh/sshd_config
EXPOSE 22
CMD ["/usr/sbin/sshd", "-D"]
paddle/scripts/docker/Dockerfile.gpu-demo 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=ON
ENV IS_DEVEL=ON
ENV WITH_DEMO=ON
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=ON
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.gpu-devel 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=ON
ENV IS_DEVEL=ON
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=ON
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.gpu-noavx 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=ON
ENV IS_DEVEL=OFF
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=OFF
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.gpu-noavx-demo 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=ON
ENV IS_DEVEL=ON
ENV WITH_DEMO=ON
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=OFF
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.gpu-noavx-devel 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=ON
ENV IS_DEVEL=ON
ENV WITH_DEMO=OFF
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=OFF
RUN cd /root/ && bash build.sh
paddle/scripts/docker/Dockerfile.m4 已删除 100644 → 0
浏览文件 @ 4f1bf30d
FROM PADDLE_BASE_IMAGE
MAINTAINER PaddlePaddle Dev Team <paddle-dev@baidu.com>
COPY build.sh /root/
ENV GIT_CHECKOUT=v0.9.0a0
ENV WITH_GPU=PADDLE_WITH_GPU
ENV IS_DEVEL=PADDLE_IS_DEVEL
ENV WITH_DEMO=PADDLE_WITH_DEMO
ENV PIP_INSTALL_ARGS ""
ENV PIP_GENERAL_ARGS ""
ENV USE_UBUNTU_MIRROR OFF
ENV WITH_AVX=PADDLE_WITH_AVX
RUN cd /root/ && bash build.sh
paddle/scripts/docker/build.sh 100644 → 100755
浏览文件 @ 021b3a44
...@@ -7,43 +7,21 @@ function abort(){ ...@@ -7,43 +7,21 @@ function abort(){
trap 'abort' 0 trap 'abort' 0
set -e set -e
if [ ${USE_UBUNTU_MIRROR} == "ON" ]; then
sed -i 's#http://archive\.ubuntu\.com/ubuntu/#mirror://mirrors\.ubuntu\.com/mirrors\.txt#g'\
/etc/apt/sources.list
fi
apt-get update
apt-get install -y cmake libprotobuf-dev protobuf-compiler git \
libgoogle-glog-dev libgflags-dev libatlas-dev libatlas3-base g++ m4 python-pip\
python-protobuf python-numpy python-dev swig
if [ ${WITH_GPU} == 'ON' ]; then if [ ${WITH_GPU} == 'ON' ]; then
ln -s /usr/lib/x86_64-linux-gnu/libcudnn.so /usr/lib/libcudnn.so ln -s /usr/lib/x86_64-linux-gnu/libcudnn.so /usr/lib/libcudnn.so
fi fi
cd ~ mkdir -p /paddle/build # -p means no error if exists
git clone https://github.com/baidu/Paddle.git paddle cd /paddle/build
cd paddle cmake .. \
git checkout ${GIT_CHECKOUT} -DWITH_DOC=ON \
mkdir build -DWITH_GPU=${WITH_GPU} \
cd build -DWITH_AVX=${WITH_AVX} \
cmake .. -DWITH_DOC=OFF -DWITH_GPU=${WITH_GPU} -DWITH_SWIG_PY=ON\ -DWITH_SWIG_PY=ON \
-DCUDNN_ROOT=/usr/ -DWITH_STYLE_CHECK=OFF -DWITH_AVX=${WITH_AVX} -DCUDNN_ROOT=/usr/ \
-DWITH_STYLE_CHECK=OFF
make -j `nproc` make -j `nproc`
# because durning make install, there are several warning, so set +e, do not cause abort
make install make install
echo 'export LD_LIBRARY_PATH=/usr/lib64:${LD_LIBRARY_PATH}' >> /etc/profile
pip ${PIP_GENERAL_ARGS} install ${PIP_INSTALL_ARGS} /usr/local/opt/paddle/share/wheels/*.whl
paddle version # print version after build
if [ ${WITH_DEMO} == "ON" ]; then
apt-get install -y wget unzip perl python-matplotlib tar xz-utils bzip2 gzip coreutils\
sed grep graphviz libjpeg-dev zlib1g-dev
pip ${PIP_GENERAL_ARGS} install ${PIP_INSTALL_ARGS} BeautifulSoup docopt \
PyYAML pillow
fi
if [ ${IS_DEVEL} == "OFF" ]; then # clean build packages.
cd ~
rm -rf paddle
fi
apt-get clean -y
trap : 0 trap : 0
paddle/scripts/docker/generate.sh 已删除 100644 → 0
浏览文件 @ 4f1bf30d
#!/bin/bash
set -e
cd `dirname $0`
m4 -DPADDLE_WITH_GPU=OFF -DPADDLE_IS_DEVEL=OFF -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=ubuntu:14.04 -DPADDLE_WITH_AVX=ON\
Dockerfile.m4 > Dockerfile.cpu
m4 -DPADDLE_WITH_GPU=OFF -DPADDLE_IS_DEVEL=OFF -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=ubuntu:14.04 -DPADDLE_WITH_AVX=OFF\
Dockerfile.m4 > Dockerfile.cpu-noavx
m4 -DPADDLE_WITH_GPU=OFF -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=ubuntu:14.04 -DPADDLE_WITH_AVX=OFF\
Dockerfile.m4 > Dockerfile.cpu-noavx-devel
m4 -DPADDLE_WITH_GPU=OFF -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=ubuntu:14.04 -DPADDLE_WITH_AVX=ON\
Dockerfile.m4 > Dockerfile.cpu-devel
m4 -DPADDLE_WITH_GPU=OFF -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=ON \
-DPADDLE_BASE_IMAGE=ubuntu:14.04 -DPADDLE_WITH_AVX=ON\
Dockerfile.m4 > Dockerfile.cpu-demo
m4 -DPADDLE_WITH_GPU=OFF -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=ON \
-DPADDLE_BASE_IMAGE=ubuntu:14.04 -DPADDLE_WITH_AVX=OFF\
Dockerfile.m4 > Dockerfile.cpu-noavx-demo
m4 -DPADDLE_WITH_GPU=ON -DPADDLE_IS_DEVEL=OFF -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 \
-DPADDLE_WITH_AVX=ON \
Dockerfile.m4 > Dockerfile.gpu
m4 -DPADDLE_WITH_GPU=ON -DPADDLE_IS_DEVEL=OFF -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 \
-DPADDLE_WITH_AVX=OFF \
Dockerfile.m4 > Dockerfile.gpu-noavx
m4 -DPADDLE_WITH_GPU=ON -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 \
-DPADDLE_WITH_AVX=ON \
Dockerfile.m4 > Dockerfile.gpu-devel
m4 -DPADDLE_WITH_GPU=ON -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=OFF \
-DPADDLE_BASE_IMAGE=nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 \
-DPADDLE_WITH_AVX=OFF \
Dockerfile.m4 > Dockerfile.gpu-noavx-devel
m4 -DPADDLE_WITH_GPU=ON -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=ON \
-DPADDLE_BASE_IMAGE=nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 \
-DPADDLE_WITH_AVX=ON \
Dockerfile.m4 > Dockerfile.gpu-demo
m4 -DPADDLE_WITH_GPU=ON -DPADDLE_IS_DEVEL=ON -DPADDLE_WITH_DEMO=ON \
-DPADDLE_BASE_IMAGE=nvidia/cuda:7.5-cudnn5-devel-ubuntu14.04 \
-DPADDLE_WITH_AVX=OFF \
Dockerfile.m4 > Dockerfile.gpu-noavx-demo
paddle/scripts/submit_local.sh.in
浏览文件 @ 021b3a44
...@@ -29,6 +29,7 @@ function version(){ ...@@ -29,6 +29,7 @@ function version(){
} }
function ver2num() { function ver2num() {
set -e
# convert version to number. # convert version to number.
if [ -z "$1" ]; then # empty argument if [ -z "$1" ]; then # empty argument
printf "%03d%03d%03d%03d%03d" 0 printf "%03d%03d%03d%03d%03d" 0
...@@ -41,6 +42,7 @@ function ver2num() { ...@@ -41,6 +42,7 @@ function ver2num() {
printf "%03d%03d%03d%03d%03d" $VERN printf "%03d%03d%03d%03d%03d" $VERN
fi fi
fi fi
set +e
} }
PADDLE_CONF_HOME="$HOME/.config/paddle" PADDLE_CONF_HOME="$HOME/.config/paddle"
......
paddle/trainer/Trainer.cpp
浏览文件 @ 021b3a44
...@@ -205,7 +205,7 @@ void Trainer::init(const std::shared_ptr<TrainerConfigHelper>& config, ...@@ -205,7 +205,7 @@ void Trainer::init(const std::shared_ptr<TrainerConfigHelper>& config,
(!IGradientMachineMode::dataMustInCpu(mode_, FLAGS_trainer_count)); (!IGradientMachineMode::dataMustInCpu(mode_, FLAGS_trainer_count));
dataProvider_ = dataProvider; dataProvider_ = dataProvider;
if (!dataProvider_ && config_->hasDataConfig()) { if (!dataProvider_ && config_->hasDataConfig() && !testing_) {
dataProvider_.reset(DataProvider::create(*config_, *config_, gpuData)); dataProvider_.reset(DataProvider::create(*config_, *config_, gpuData));
} }
if (!testDataProvider_) { if (!testDataProvider_) {
......
paddle/trainer/tests/test_config.conf
浏览文件 @ 021b3a44
...@@ -59,7 +59,6 @@ pool = img_pool_layer(input=fc2, ...@@ -59,7 +59,6 @@ pool = img_pool_layer(input=fc2,
padding_y=2, padding_y=2,
stride=2, stride=2,
stride_y=3, stride_y=3,
img_width=3,
pool_type=CudnnAvgPooling()) pool_type=CudnnAvgPooling())
concat = concat_layer(input=[fc3, fc4]) concat = concat_layer(input=[fc3, fc4])
......
proto/CMakeLists.txt
浏览文件 @ 021b3a44
...@@ -11,8 +11,12 @@ set(real_proto_files) ...@@ -11,8 +11,12 @@ set(real_proto_files)
# TODO(yuyang18): Some internal proto will also be depended on. # TODO(yuyang18): Some internal proto will also be depended on.
# Find a way to automatically calculate all depends. # Find a way to automatically calculate all depends.
foreach(filename ${proto_filenames}) foreach(filename ${proto_filenames})
set(PROTOBUF_3_FLAGS "")
if (PROTOBUF_3)
set(PROTOBUF_3_FLAGS "-Dproto3")
endif()
add_custom_command(OUTPUT ${filename} add_custom_command(OUTPUT ${filename}
COMMAND ${M4_EXECUTABLE} -Dreal=${ACCURACY} -I '${INTERNAL_PROTO_PATH}' COMMAND ${M4_EXECUTABLE} -Dreal=${ACCURACY} ${PROTOBUF_3_FLAGS} -I '${INTERNAL_PROTO_PATH}'
${PROJ_ROOT}/proto/${filename}.m4 > ${filename} ${PROJ_ROOT}/proto/${filename}.m4 > ${filename}
DEPENDS ${PROJ_ROOT}/proto/${filename}.m4 DEPENDS ${PROJ_ROOT}/proto/${filename}.m4
COMMENT "Generate ${filename}") COMMENT "Generate ${filename}")
......
proto/DataConfig.proto.m4
浏览文件 @ 021b3a44
...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS, ...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
ifdef(`proto3', `syntax = "proto2";')
package paddle; package paddle;
......
proto/DataFormat.proto.m4
浏览文件 @ 021b3a44
...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS, ...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
ifdef(`proto3', `syntax = "proto2";')
package paddle; package paddle;
......
proto/ModelConfig.proto.m4
浏览文件 @ 021b3a44
...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS, ...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
ifdef(`proto3', `syntax = "proto2";')
import "ParameterConfig.proto"; import "ParameterConfig.proto";
...@@ -76,6 +77,12 @@ message ConvConfig { ...@@ -76,6 +77,12 @@ message ConvConfig {
required uint32 filter_size_y = 10; required uint32 filter_size_y = 10;
required uint32 padding_y = 11; required uint32 padding_y = 11;
required uint32 stride_y = 12; required uint32 stride_y = 12;
// if not set, use output_x
optional uint32 output_y = 13;
// if not set, use img_size
optional uint32 img_size_y = 14;
} }
message PoolConfig { message PoolConfig {
...@@ -121,11 +128,9 @@ message PoolConfig { ...@@ -121,11 +128,9 @@ message PoolConfig {
} }
message SppConfig { message SppConfig {
required string pool_type = 1; required ImageConfig image_conf = 1;
required uint32 pyramid_height = 2; required string pool_type = 2;
required uint32 channels = 3; required uint32 pyramid_height = 3;
required uint32 img_size = 4;
optional uint32 img_size_y = 5;
} }
message NormConfig { message NormConfig {
...@@ -155,6 +160,12 @@ message NormConfig { ...@@ -155,6 +160,12 @@ message NormConfig {
// fixed window: shared a fixed window for each value // fixed window: shared a fixed window for each value
// sliding window: have a different window for each value // sliding window: have a different window for each value
optional bool blocked = 8; optional bool blocked = 8;
// if not set, use output_x
optional uint32 output_y = 9;
// if not set, use img_size
optional uint32 img_size_y = 10;
} }
message BlockExpandConfig { message BlockExpandConfig {
...@@ -179,12 +190,8 @@ message BlockExpandConfig { ...@@ -179,12 +190,8 @@ message BlockExpandConfig {
} }
message MaxOutConfig { message MaxOutConfig {
required uint32 channels = 1; required ImageConfig image_conf = 1;
required uint32 groups = 2; required uint32 groups = 2;
// The size of input feature map.
required uint32 img_size_x = 3;
required uint32 img_size_y = 4;
} }
message ProjectionConfig { message ProjectionConfig {
...@@ -225,12 +232,10 @@ message OperatorConfig { ...@@ -225,12 +232,10 @@ message OperatorConfig {
message BilinearInterpConfig { message BilinearInterpConfig {
// The size of input feature map. // The size of input feature map.
optional uint32 img_size_x = 1; required ImageConfig image_conf = 1;
optional uint32 img_size_y = 2;
// The size of output feature map. // The size of output feature map.
required uint32 out_size_x = 3; required uint32 out_size_x = 2;
required uint32 out_size_y = 4; required uint32 out_size_y = 3;
required uint32 num_channels = 5;
} }
message ImageConfig { message ImageConfig {
...@@ -240,6 +245,7 @@ message ImageConfig { ...@@ -240,6 +245,7 @@ message ImageConfig {
// The size of input feature map. // The size of input feature map.
required uint32 img_size = 8; required uint32 img_size = 8;
required uint32 img_size_y = 9;
} }
message LayerInputConfig { message LayerInputConfig {
...@@ -413,6 +419,9 @@ sinclude(`ModelConfigLayer.proto.m4') ...@@ -413,6 +419,9 @@ sinclude(`ModelConfigLayer.proto.m4')
// to string and reinterpreted in the user's own layer implementation. // to string and reinterpreted in the user's own layer implementation.
optional string user_arg = 49; optional string user_arg = 49;
// to indicate rectangle image data
optional uint64 height = 50;
optional uint64 width = 51;
} }
message EvaluatorConfig { message EvaluatorConfig {
......
proto/ParameterConfig.proto.m4
浏览文件 @ 021b3a44
...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS, ...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
ifdef(`proto3', `syntax = "proto2";')
package paddle; package paddle;
......
proto/ParameterService.proto.m4
浏览文件 @ 021b3a44
...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS, ...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
ifdef(`proto3', `syntax = "proto2";')
import "ParameterConfig.proto"; import "ParameterConfig.proto";
import "TrainerConfig.proto"; import "TrainerConfig.proto";
...@@ -20,7 +21,6 @@ package paddle; ...@@ -20,7 +21,6 @@ package paddle;
/** /**
* Various structs for communicating with parameter server * Various structs for communicating with parameter server
*/ */
enum ParameterUpdateMode { enum ParameterUpdateMode {
// Set parameter // Set parameter
PSERVER_UPDATE_MODE_SET_PARAM = 0;//use local param PSERVER_UPDATE_MODE_SET_PARAM = 0;//use local param
......
proto/TrainerConfig.proto.m4
浏览文件 @ 021b3a44
...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS, ...@@ -11,6 +11,7 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
ifdef(`proto3', `syntax = "proto2";')
import "DataConfig.proto"; import "DataConfig.proto";
import "ModelConfig.proto"; import "ModelConfig.proto";
......
python/paddle/trainer/config_parser.py
浏览文件 @ 021b3a44
...@@ -138,7 +138,14 @@ def init_config_environment( ...@@ -138,7 +138,14 @@ def init_config_environment(
g_root_submodel=None, g_root_submodel=None,
g_submodel_map={}, g_submodel_map={},
g_submodel_stack=[], g_submodel_stack=[],
g_add_submodel_suffix=False, ): g_add_submodel_suffix=False,
# Whether current layer needs to pass the image height and width.
# Default value is true, but if it encounters recurrent_layer_group,
# it will be false. The reason is that image is converted to be sequence,
# image height will be sequence length, and image width will be feature
# length of each timestep.
g_pass_height_width=True, ):
for k, v in locals().iteritems(): for k, v in locals().iteritems():
globals()[k] = copy.deepcopy(v) globals()[k] = copy.deepcopy(v)
...@@ -686,9 +693,9 @@ class ConvProjection(Projection): ...@@ -686,9 +693,9 @@ class ConvProjection(Projection):
parse_conv(conv_conf, input_layer_name, self.proj_conf.conv_conf, parse_conv(conv_conf, input_layer_name, self.proj_conf.conv_conf,
num_filters) num_filters)
# TODO: support rectangle input self.proj_conf.output_size = self.proj_conf.conv_conf.output_x * \
self.proj_conf.output_size = (self.proj_conf.conv_conf.output_x self.proj_conf.conv_conf.output_y * \
**2) * num_filters num_filters
def calc_output_size(self, input_layer_config): def calc_output_size(self, input_layer_config):
return self.proj_conf.output_size return self.proj_conf.output_size
...@@ -698,7 +705,8 @@ class ConvProjection(Projection): ...@@ -698,7 +705,8 @@ class ConvProjection(Projection):
ci = self.proj_conf.conv_conf.channels ci = self.proj_conf.conv_conf.channels
fh = self.proj_conf.conv_conf.filter_size fh = self.proj_conf.conv_conf.filter_size
fw = self.proj_conf.conv_conf.filter_size_y fw = self.proj_conf.conv_conf.filter_size_y
return co * ci * fh * fw gr = self.proj_conf.conv_conf.groups
return co * ci * fh * fw / gr
def calc_bias_size(self): def calc_bias_size(self):
return self.proj_conf.num_filters return self.proj_conf.num_filters
...@@ -763,8 +771,9 @@ class ConvOperator(Operator): ...@@ -763,8 +771,9 @@ class ConvOperator(Operator):
parse_conv(conv_conf, parse_conv(conv_conf,
MakeLayerNameInSubmodel(input_layer_names[0]), MakeLayerNameInSubmodel(input_layer_names[0]),
self.operator_conf.conv_conf, num_filters) self.operator_conf.conv_conf, num_filters)
self.operator_conf.output_size = (self.operator_conf.conv_conf.output_x self.operator_conf.output_size = self.operator_conf.conv_conf.output_x * \
**2) * num_filters self.operator_conf.conv_conf.output_y * \
num_filters
config_assert(len(input_layer_names) == 2, "Conv is binary operator") config_assert(len(input_layer_names) == 2, "Conv is binary operator")
...@@ -799,14 +808,12 @@ class Conv(Cfg): ...@@ -799,14 +808,12 @@ class Conv(Cfg):
config_assert(output_x <= 0) config_assert(output_x <= 0)
# please refer to the comments in proto/ModelConfig.proto
@config_class @config_class
class BilinearInterp(Cfg): class BilinearInterp(Cfg):
def __init__(self, out_size_x=None, out_size_y=None, num_channels=None): def __init__(self, out_size_x=None, out_size_y=None, channels=None):
self.add_keys(locals()) self.add_keys(locals())
# please refer to the comments in proto/ModelConfig.proto
@config_class @config_class
class Pool(Cfg): class Pool(Cfg):
def __init__( def __init__(
...@@ -824,14 +831,12 @@ class Pool(Cfg): ...@@ -824,14 +831,12 @@ class Pool(Cfg):
self.add_keys(locals()) self.add_keys(locals())
# please refer to the comments in proto/ModelConfig.proto
@config_class @config_class
class SpatialPyramidPool(Cfg): class SpatialPyramidPool(Cfg):
def __init__(self, pool_type, pyramid_height, channels, img_width=None): def __init__(self, pool_type, pyramid_height, channels):
self.add_keys(locals()) self.add_keys(locals())
# please refer to the comments in proto/ModelConfig.proto
@config_class @config_class
class Norm(Cfg): class Norm(Cfg):
def __init__(self, def __init__(self,
...@@ -846,7 +851,6 @@ class Norm(Cfg): ...@@ -846,7 +851,6 @@ class Norm(Cfg):
self.add_keys(locals()) self.add_keys(locals())
# please refer to the comments in proto/ModelConfig.proto
@config_class @config_class
class Image(Cfg): class Image(Cfg):
def __init__(self, channels, img_size=None): def __init__(self, channels, img_size=None):
...@@ -1053,18 +1057,8 @@ def TestData(data_config, async_load_data=None): ...@@ -1053,18 +1057,8 @@ def TestData(data_config, async_load_data=None):
g_config.test_data_config.async_load_data = async_load_data g_config.test_data_config.async_load_data = async_load_data
def parse_bilinear(bilinear, input_layer_name, bilinear_conf): #caffe_mode: compute the output size using floor instead of ceil,
bilinear_conf.out_size_x = bilinear.out_size_x # which is consistent of caffe and CuDNN's convention.
bilinear_conf.out_size_y = bilinear.out_size_y
bilinear_conf.num_channels = bilinear.num_channels
'''
caffe_mode: compute the output size using floor instead of ceil,
which is consistent of caffe and CuDNN's convention.
'''
def cnn_output_size(img_size, filter_size, padding, stride, caffe_mode): def cnn_output_size(img_size, filter_size, padding, stride, caffe_mode):
output = (2 * padding + img_size - filter_size) / float(stride) output = (2 * padding + img_size - filter_size) / float(stride)
if caffe_mode: if caffe_mode:
...@@ -1073,20 +1067,34 @@ def cnn_output_size(img_size, filter_size, padding, stride, caffe_mode): ...@@ -1073,20 +1067,34 @@ def cnn_output_size(img_size, filter_size, padding, stride, caffe_mode):
return 1 + int(math.ceil(output)) return 1 + int(math.ceil(output))
''' #calcualte image_size based on output_size for de-convolution (ConvTransLayer).
calcualte image_size based on output_size for convolution. #It is the reverse function of cnn_output_size
It is the reverse function of cnn_output_size
'''
def cnn_image_size(output_size, filter_size, padding, stride, caffe_mode): def cnn_image_size(output_size, filter_size, padding, stride, caffe_mode):
if caffe_mode:
img_size = (output_size - 1) * stride + filter_size - 2 * padding img_size = (output_size - 1) * stride + filter_size - 2 * padding
else: if not caffe_mode:
img_size = (output_size - 2) * stride + filter_size - 2 * padding + 1 img_size = img_size + 1
return img_size return img_size
def get_img_size(input_layer_name, channels):
input = g_layer_map[input_layer_name]
img_pixels = input.size / channels
img_size = input.width if input.width > 0 else int(img_pixels**0.5)
img_size_y = input.height if input.height > 0 else int(img_pixels /
img_size)
config_assert(
img_size * img_size_y == img_pixels,
"Input layer %s: Incorrect input image size %d * %d for input image pixels %d"
% (input_layer_name, img_size, img_size_y, img_pixels))
return img_size, img_size_y
def parse_bilinear(bilinear, input_layer_name, bilinear_conf):
parse_image(bilinear, input_layer_name, bilinear_conf.image_conf)
bilinear_conf.out_size_x = bilinear.out_size_x
bilinear_conf.out_size_y = bilinear.out_size_y
def parse_pool(pool, input_layer_name, pool_conf): def parse_pool(pool, input_layer_name, pool_conf):
pool_conf.pool_type = pool.pool_type pool_conf.pool_type = pool.pool_type
config_assert(pool.pool_type in [ config_assert(pool.pool_type in [
...@@ -1102,14 +1110,8 @@ def parse_pool(pool, input_layer_name, pool_conf): ...@@ -1102,14 +1110,8 @@ def parse_pool(pool, input_layer_name, pool_conf):
pool_conf.size_y = default(pool.size_y, pool_conf.size_x) pool_conf.size_y = default(pool.size_y, pool_conf.size_x)
pool_conf.stride_y = default(pool.stride_y, pool_conf.stride) pool_conf.stride_y = default(pool.stride_y, pool_conf.stride)
img_pixels = g_layer_map[input_layer_name].size / pool.channels pool_conf.img_size, pool_conf.img_size_y = \
# the img_width may be removed, get_img_size(input_layer_name, pool.channels)
# and it can be calculated automatically later.
pool_conf.img_size = default(pool.img_width, int(img_pixels**0.5))
pool_conf.img_size_y = img_pixels / pool_conf.img_size
config_assert(pool_conf.img_size * pool_conf.img_size_y == img_pixels,
"Incorrect input image size %d for input image pixels %d" %
(pool_conf.img_size, img_pixels))
config_assert(not pool.start, "start is deprecated in pooling.") config_assert(not pool.start, "start is deprecated in pooling.")
...@@ -1125,29 +1127,18 @@ def parse_pool(pool, input_layer_name, pool_conf): ...@@ -1125,29 +1127,18 @@ def parse_pool(pool, input_layer_name, pool_conf):
def parse_spp(spp, input_layer_name, spp_conf): def parse_spp(spp, input_layer_name, spp_conf):
parse_image(spp, input_layer_name, spp_conf.image_conf)
spp_conf.pool_type = spp.pool_type spp_conf.pool_type = spp.pool_type
config_assert(spp.pool_type in ['max-projection', 'avg-projection'], config_assert(spp.pool_type in ['max-projection', 'avg-projection'],
"pool-type %s is not in " "pool-type %s is not in "
"['max-projection', 'avg-projection']" % spp.pool_type) "['max-projection', 'avg-projection']" % spp.pool_type)
spp_conf.pyramid_height = spp.pyramid_height spp_conf.pyramid_height = spp.pyramid_height
spp_conf.channels = spp.channels
img_pixels = g_layer_map[input_layer_name].size / spp_conf.channels
spp_conf.img_size = default(spp.img_width, int(img_pixels**0.5))
spp_conf.img_size_y = img_pixels / spp_conf.img_size
config_assert(spp_conf.img_size * spp_conf.img_size_y == img_pixels,
"Incorrect input image size %d for input image pixels %d" %
(spp_conf.img_size, img_pixels))
def parse_image(image, input_layer_name, image_conf): def parse_image(image, input_layer_name, image_conf):
image_conf.channels = image.channels image_conf.channels = image.channels
image_pixels = g_layer_map[input_layer_name].size / image_conf.channels image_conf.img_size, image_conf.img_size_y = \
image_conf.img_size = int(image_pixels**0.5) get_img_size(input_layer_name, image_conf.channels)
config_assert((image_conf.img_size**2) == image_pixels,
"Incorrect input image size %d for input image pixels %d" %
(image_conf.img_size, image_pixels))
def parse_norm(norm, input_layer_name, norm_conf): def parse_norm(norm, input_layer_name, norm_conf):
...@@ -1161,24 +1152,18 @@ def parse_norm(norm, input_layer_name, norm_conf): ...@@ -1161,24 +1152,18 @@ def parse_norm(norm, input_layer_name, norm_conf):
norm_conf.pow = norm.pow norm_conf.pow = norm.pow
norm_conf.blocked = norm.blocked norm_conf.blocked = norm.blocked
img_pixels = g_layer_map[input_layer_name].size / norm.channels norm_conf.img_size, norm_conf.img_size_y = \
norm_conf.img_size = int(img_pixels**0.5) get_img_size(input_layer_name, norm.channels)
config_assert((norm_conf.img_size**2) == img_pixels,
"Incorrect input image size %d for input image pixels %d" %
(norm_conf.img_size, img_pixels))
norm_conf.output_x = norm_conf.img_size norm_conf.output_x = norm_conf.img_size
norm_conf.output_y = norm_conf.img_size_y
if norm.norm_type in ['cmrnorm-projection']: if norm.norm_type in ['cmrnorm-projection']:
norm_conf.scale /= norm.size norm_conf.scale /= norm.size
else: else:
norm_conf.scale /= norm.size**2 norm_conf.scale /= norm.size**2
''' #caffe_mode: compute the output size using floor instead of ceil,
caffe_mode: compute the output size using floor instead of ceil, # which is consistent of caffe and CuDNN's convention.
which is consistent of caffe and CuDNN's convention.
'''
def parse_conv(conv, input_layer_name, conv_conf, num_filters, trans=False): def parse_conv(conv, input_layer_name, conv_conf, num_filters, trans=False):
conv_conf.filter_size = conv.filter_size conv_conf.filter_size = conv.filter_size
conv_conf.filter_size_y = conv.filter_size_y conv_conf.filter_size_y = conv.filter_size_y
...@@ -1192,33 +1177,24 @@ def parse_conv(conv, input_layer_name, conv_conf, num_filters, trans=False): ...@@ -1192,33 +1177,24 @@ def parse_conv(conv, input_layer_name, conv_conf, num_filters, trans=False):
if not trans: if not trans:
conv_conf.filter_channels = conv.channels / conv.groups conv_conf.filter_channels = conv.channels / conv.groups
conv_conf.img_size, conv_conf.img_size_y = \
img_pixels = g_layer_map[input_layer_name].size / conv.channels get_img_size(input_layer_name, conv.channels)
print('channels=%d size=%d' % (conv.channels,
g_layer_map[input_layer_name].size))
conv_conf.img_size = int(img_pixels**0.5)
config_assert((conv_conf.img_size**2) == img_pixels, (
"Input layer %s: Incorrect input image size %d for input " +
"image pixels %d") %
(input_layer_name, conv_conf.img_size, img_pixels))
conv_conf.output_x = cnn_output_size( conv_conf.output_x = cnn_output_size(
conv_conf.img_size, conv_conf.filter_size, conv_conf.padding, conv_conf.img_size, conv_conf.filter_size, conv_conf.padding,
conv_conf.stride, conv_conf.caffe_mode) conv_conf.stride, conv_conf.caffe_mode)
conv_conf.output_y = cnn_output_size(
conv_conf.img_size_y, conv_conf.filter_size_y, conv_conf.padding_y,
conv_conf.stride_y, conv_conf.caffe_mode)
else: else:
conv_conf.filter_channels = num_filters / conv.groups conv_conf.filter_channels = num_filters / conv.groups
conv_conf.output_x, conv_conf.output_y = \
outputSize = g_layer_map[input_layer_name].size / conv.channels get_img_size(input_layer_name, conv.channels)
print('channels=%d size=%d' % (conv.channels,
g_layer_map[input_layer_name].size))
conv_conf.output_x = int(outputSize**0.5)
config_assert((conv_conf.output_x**2) == outputSize, (
"Input layer %s: Incorrect input image size %d for input " +
"image pixels %d") %
(input_layer_name, conv_conf.output_x, outputSize))
conv_conf.img_size = cnn_image_size( conv_conf.img_size = cnn_image_size(
conv_conf.output_x, conv_conf.filter_size, conv_conf.padding, conv_conf.output_x, conv_conf.filter_size, conv_conf.padding,
conv_conf.stride, conv_conf.caffe_mode) conv_conf.stride, conv_conf.caffe_mode)
conv_conf.img_size_y = cnn_image_size(
conv_conf.output_y, conv_conf.filter_size_y, conv_conf.padding_y,
conv_conf.stride_y, conv_conf.caffe_mode)
def parse_block_expand(block_expand, input_layer_name, block_expand_conf): def parse_block_expand(block_expand, input_layer_name, block_expand_conf):
...@@ -1247,10 +1223,8 @@ def parse_block_expand(block_expand, input_layer_name, block_expand_conf): ...@@ -1247,10 +1223,8 @@ def parse_block_expand(block_expand, input_layer_name, block_expand_conf):
def parse_maxout(maxout, input_layer_name, maxout_conf): def parse_maxout(maxout, input_layer_name, maxout_conf):
maxout_conf.channels = maxout.channels parse_image(maxout, input_layer_name, maxout_conf.image_conf)
maxout_conf.groups = maxout.groups maxout_conf.groups = maxout.groups
maxout_conf.img_size_x = maxout.img_size_x
maxout_conf.img_size_y = maxout.img_size_y
# Define an evaluator # Define an evaluator
...@@ -1377,6 +1351,12 @@ class LayerBase(object): ...@@ -1377,6 +1351,12 @@ class LayerBase(object):
g_current_submodel.layer_names.append(self.config.name) g_current_submodel.layer_names.append(self.config.name)
if self.config.type != 'data' and g_pass_height_width:
height = self.get_input_layer(0).height
width = self.get_input_layer(0).width
if height and width:
self.set_layer_height_width(height, width)
def get_input_layer(self, input_index): def get_input_layer(self, input_index):
return g_layer_map[self.config.inputs[input_index].input_layer_name] return g_layer_map[self.config.inputs[input_index].input_layer_name]
...@@ -1494,6 +1474,23 @@ class LayerBase(object): ...@@ -1494,6 +1474,23 @@ class LayerBase(object):
'Different inputs result in' + 'Different inputs result in' +
'different layer size at layer %s' % self.config.name) 'different layer size at layer %s' % self.config.name)
def set_layer_height_width(self, height, width):
self.config.height = height
self.config.width = width
def set_cnn_layer(self,
input_layer_name,
height,
width,
channels,
is_print=True):
size = height * width * channels
self.set_layer_size(size)
self.set_layer_height_width(height, width)
if is_print:
print("output for %s: c = %d, h = %d, w = %d, size = %d" %
(input_layer_name, channels, height, width, size))
@config_layer('multi_class_cross_entropy_with_selfnorm') @config_layer('multi_class_cross_entropy_with_selfnorm')
class MultiClassCrossEntropySelfNormCostLayer(LayerBase): class MultiClassCrossEntropySelfNormCostLayer(LayerBase):
...@@ -1583,9 +1580,11 @@ class PrintLayer(LayerBase): ...@@ -1583,9 +1580,11 @@ class PrintLayer(LayerBase):
@config_layer('data') @config_layer('data')
class DataLayer(LayerBase): class DataLayer(LayerBase):
def __init__(self, name, size, device=None): def __init__(self, name, size, height=None, width=None, device=None):
super(DataLayer, self).__init__( super(DataLayer, self).__init__(
name, 'data', size, inputs=[], device=device) name, 'data', size, inputs=[], device=device)
if height and width:
self.set_layer_height_width(height, width)
''' '''
...@@ -1684,14 +1683,13 @@ class ConvLayerBase(LayerBase): ...@@ -1684,14 +1683,13 @@ class ConvLayerBase(LayerBase):
for input_index in xrange(len(self.inputs)): for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index) input_layer = self.get_input_layer(input_index)
parse_conv(self.inputs[input_index].conv, input_layer.name,
self.config.inputs[input_index].conv_conf, num_filters)
conv_conf = self.config.inputs[input_index].conv_conf conv_conf = self.config.inputs[input_index].conv_conf
parse_conv(self.inputs[input_index].conv, input_layer.name,
conv_conf, num_filters)
psize = self.calc_parameter_size(conv_conf) psize = self.calc_parameter_size(conv_conf)
print("output size for %s is %d " % (name, conv_conf.output_x))
self.create_input_parameter(input_index, psize) self.create_input_parameter(input_index, psize)
self.set_layer_size( self.set_cnn_layer(name, conv_conf.output_y, conv_conf.output_x,
(conv_conf.output_x**2) * self.config.num_filters) self.config.num_filters)
psize = self.config.size psize = self.config.size
if shared_biases: if shared_biases:
...@@ -1778,10 +1776,11 @@ class NormLayer(LayerBase): ...@@ -1778,10 +1776,11 @@ class NormLayer(LayerBase):
name, 'norm', 0, inputs=inputs, device=device) name, 'norm', 0, inputs=inputs, device=device)
for input_index in xrange(len(self.inputs)): for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index) input_layer = self.get_input_layer(input_index)
parse_norm(self.inputs[input_index].norm, input_layer.name,
self.config.inputs[input_index].norm_conf)
norm_conf = self.config.inputs[input_index].norm_conf norm_conf = self.config.inputs[input_index].norm_conf
self.set_layer_size((norm_conf.output_x**2) * norm_conf.channels) parse_norm(self.inputs[input_index].norm, input_layer.name,
norm_conf)
self.set_cnn_layer(name, norm_conf.output_y, norm_conf.output_x,
norm_conf.channels, False)
@config_layer('pool') @config_layer('pool')
...@@ -1791,13 +1790,11 @@ class PoolLayer(LayerBase): ...@@ -1791,13 +1790,11 @@ class PoolLayer(LayerBase):
name, 'pool', 0, inputs=inputs, device=device) name, 'pool', 0, inputs=inputs, device=device)
for input_index in xrange(len(self.inputs)): for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index) input_layer = self.get_input_layer(input_index)
parse_pool(self.inputs[input_index].pool, input_layer.name,
self.config.inputs[input_index].pool_conf)
pool_conf = self.config.inputs[input_index].pool_conf pool_conf = self.config.inputs[input_index].pool_conf
print("output size for %s is %d*%d " % (name, pool_conf.output_y, parse_pool(self.inputs[input_index].pool, input_layer.name,
pool_conf.output_x)) pool_conf)
self.set_layer_size( self.set_cnn_layer(name, pool_conf.output_y, pool_conf.output_x,
(pool_conf.output_x * pool_conf.output_y) * pool_conf.channels) pool_conf.channels)
@config_layer('spp') @config_layer('spp')
...@@ -1807,12 +1804,10 @@ class SpatialPyramidPoolLayer(LayerBase): ...@@ -1807,12 +1804,10 @@ class SpatialPyramidPoolLayer(LayerBase):
name, 'spp', 0, inputs=inputs, device=device) name, 'spp', 0, inputs=inputs, device=device)
for input_index in xrange(len(self.inputs)): for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index) input_layer = self.get_input_layer(input_index)
parse_spp(self.inputs[input_index].spp, input_layer.name,
self.config.inputs[input_index].spp_conf)
spp_conf = self.config.inputs[input_index].spp_conf spp_conf = self.config.inputs[input_index].spp_conf
output_size = (pow(4, spp_conf.pyramid_height) - 1) / (4 - 1) parse_spp(self.inputs[input_index].spp, input_layer.name, spp_conf)
print("output size for %s is %d " % (name, output_size)) output_x = (pow(4, spp_conf.pyramid_height) - 1) / (4 - 1)
self.set_layer_size(output_size * spp_conf.channels) self.set_cnn_layer(name, 1, output_x, spp_conf.image_conf.channels)
@config_layer('batch_norm') @config_layer('batch_norm')
...@@ -1874,10 +1869,10 @@ class BatchNormLayer(LayerBase): ...@@ -1874,10 +1869,10 @@ class BatchNormLayer(LayerBase):
self.config.moving_average_fraction = moving_average_fraction self.config.moving_average_fraction = moving_average_fraction
input_layer = self.get_input_layer(0) input_layer = self.get_input_layer(0)
parse_image(self.inputs[0].image, input_layer.name,
self.config.inputs[0].image_conf)
image_conf = self.config.inputs[0].image_conf image_conf = self.config.inputs[0].image_conf
self.set_layer_size((image_conf.img_size**2) * image_conf.channels) parse_image(self.inputs[0].image, input_layer.name, image_conf)
self.set_cnn_layer(name, image_conf.img_size_y, image_conf.img_size,
image_conf.channels)
psize = self.calc_parameter_size(image_conf) psize = self.calc_parameter_size(image_conf)
dims = [1, psize] dims = [1, psize]
...@@ -1935,11 +1930,11 @@ class MaxOutLayer(LayerBase): ...@@ -1935,11 +1930,11 @@ class MaxOutLayer(LayerBase):
super(MaxOutLayer, self).__init__( super(MaxOutLayer, self).__init__(
name, 'maxout', 0, inputs=inputs, **xargs) name, 'maxout', 0, inputs=inputs, **xargs)
input_layer = self.get_input_layer(0) input_layer = self.get_input_layer(0)
parse_maxout(self.inputs[0].maxout, input_layer.name,
self.config.inputs[0].maxout_conf)
maxout_conf = self.config.inputs[0].maxout_conf maxout_conf = self.config.inputs[0].maxout_conf
self.set_layer_size(g_layer_map[input_layer.name].size / parse_maxout(self.inputs[0].maxout, input_layer.name, maxout_conf)
maxout_conf.groups) out_channels = maxout_conf.image_conf.channels / maxout_conf.groups
self.set_cnn_layer(name, g_layer_map[input_layer.name].height,
g_layer_map[input_layer.name].width, out_channels)
# key: cost type # key: cost type
...@@ -2519,11 +2514,10 @@ class BilinearInterpLayer(LayerBase): ...@@ -2519,11 +2514,10 @@ class BilinearInterpLayer(LayerBase):
super(BilinearInterpLayer, self).__init__( super(BilinearInterpLayer, self).__init__(
name, 'bilinear_interp', 0, inputs=inputs, **xargs) name, 'bilinear_interp', 0, inputs=inputs, **xargs)
input_layer = self.get_input_layer(0) input_layer = self.get_input_layer(0)
parse_bilinear(self.inputs[0].bilinear_interp, input_layer.name, conf = self.config.inputs[0].bilinear_interp_conf
self.config.inputs[0].bilinear_interp_conf) parse_bilinear(self.inputs[0].bilinear_interp, input_layer.name, conf)
conf = self.inputs[0].bilinear_interp self.set_cnn_layer(name, conf.out_size_y, conf.out_size_x,
self.set_layer_size(conf.out_size_x * conf.out_size_y * conf.image_conf.channels)
conf.num_channels)
@config_layer('sum_to_one_norm') @config_layer('sum_to_one_norm')
...@@ -2996,6 +2990,8 @@ class CTCLayer(LayerBase): ...@@ -2996,6 +2990,8 @@ class CTCLayer(LayerBase):
@config_layer('recurrent_layer_group') @config_layer('recurrent_layer_group')
class RecurrentLayerGroup(LayerBase): class RecurrentLayerGroup(LayerBase):
def __init__(self, name, device=None): def __init__(self, name, device=None):
global g_pass_height_width
g_pass_height_width = False
super(RecurrentLayerGroup, self).__init__( super(RecurrentLayerGroup, self).__init__(
name, 'recurrent_layer_group', 0, inputs=[], device=device) name, 'recurrent_layer_group', 0, inputs=[], device=device)
......
python/paddle/trainer_config_helpers/layers.py
浏览文件 @ 021b3a44
...@@ -129,6 +129,9 @@ class LayerType(object): ...@@ -129,6 +129,9 @@ class LayerType(object):
HSIGMOID = 'hsigmoid' HSIGMOID = 'hsigmoid'
CONV_LAYER = "conv" CONV_LAYER = "conv"
CONVTRANS_LAYER = "convt" CONVTRANS_LAYER = "convt"
EXCONV_LAYER = "exconv"
EXCONVTRANS_LAYER = "exconvt"
CUDNNCONV_LAYER = "cudnn_conv"
POOL_LAYER = "pool" POOL_LAYER = "pool"
BATCH_NORM_LAYER = 'batch_norm' BATCH_NORM_LAYER = 'batch_norm'
NORM_LAYER = 'norm' NORM_LAYER = 'norm'
...@@ -763,7 +766,7 @@ def mixed_layer(size=0, ...@@ -763,7 +766,7 @@ def mixed_layer(size=0,
@layer_support() @layer_support()
def data_layer(name, size, layer_attr=None): def data_layer(name, size, height=None, width=None, layer_attr=None):
""" """
Define DataLayer For NeuralNetwork. Define DataLayer For NeuralNetwork.
...@@ -778,6 +781,10 @@ def data_layer(name, size, layer_attr=None): ...@@ -778,6 +781,10 @@ def data_layer(name, size, layer_attr=None):
:type name: basestring :type name: basestring
:param size: Size of this data layer. :param size: Size of this data layer.
:type size: int :type size: int
:param height: Height of this data layer, used for image
:type size: int|None
:param width: Width of this data layer, used for image
:type size: int|None
:param layer_attr: Extra Layer Attribute. :param layer_attr: Extra Layer Attribute.
:type layer_attr: ExtraLayerAttribute. :type layer_attr: ExtraLayerAttribute.
:return: LayerOutput object. :return: LayerOutput object.
...@@ -787,6 +794,8 @@ def data_layer(name, size, layer_attr=None): ...@@ -787,6 +794,8 @@ def data_layer(name, size, layer_attr=None):
type=LayerType.DATA, type=LayerType.DATA,
name=name, name=name,
size=size, size=size,
height=height,
width=width,
**ExtraLayerAttribute.to_kwargs(layer_attr)) **ExtraLayerAttribute.to_kwargs(layer_attr))
return LayerOutput(name, LayerType.DATA, size=size) return LayerOutput(name, LayerType.DATA, size=size)
...@@ -1480,7 +1489,7 @@ def bilinear_interp_layer(input, ...@@ -1480,7 +1489,7 @@ def bilinear_interp_layer(input,
bilinear_interp=BilinearInterp( bilinear_interp=BilinearInterp(
out_size_x=out_size_x, out_size_x=out_size_x,
out_size_y=out_size_y, out_size_y=out_size_y,
num_channels=num_channels)), channels=num_channels)),
type=LayerType.BILINEAR_INTERP_LAYER, type=LayerType.BILINEAR_INTERP_LAYER,
**ExtraLayerAttribute.to_kwargs(layer_attr)) **ExtraLayerAttribute.to_kwargs(layer_attr))
return LayerOutput( return LayerOutput(
...@@ -1762,7 +1771,8 @@ def img_conv_layer(input, ...@@ -1762,7 +1771,8 @@ def img_conv_layer(input,
filter_size_y=None, filter_size_y=None,
stride_y=None, stride_y=None,
padding_y=None, padding_y=None,
trans=False): trans=False,
layer_type=None):
""" """
Convolution layer for image. Paddle only support square input currently and Convolution layer for image. Paddle only support square input currently and
thus input image's width equals height. thus input image's width equals height.
...@@ -1829,6 +1839,10 @@ def img_conv_layer(input, ...@@ -1829,6 +1839,10 @@ def img_conv_layer(input,
:type layer_attr: ExtraLayerAttribute :type layer_attr: ExtraLayerAttribute
:param trans: true if it is a convTransLayer, false if it is a convLayer :param trans: true if it is a convTransLayer, false if it is a convLayer
:type trans: bool :type trans: bool
:param layer_type: specify the layer_type, default is None. If trans=True,
layer_type has to be "exconvt", otherwise layer_type
has to be either "exconv" or "cudnn_conv"
:type layer_type: String
:return: LayerOutput object. :return: LayerOutput object.
:rtype: LayerOutput :rtype: LayerOutput
""" """
...@@ -1865,6 +1879,13 @@ def img_conv_layer(input, ...@@ -1865,6 +1879,13 @@ def img_conv_layer(input,
param_attr.attr["initial_strategy"] = 0 param_attr.attr["initial_strategy"] = 0
param_attr.attr["initial_smart"] = False param_attr.attr["initial_smart"] = False
if layer_type:
if trans:
assert layer_type in ["exconvt"]
else:
assert layer_type in ["exconv", "cudnn_conv"]
lt = layer_type
else:
lt = LayerType.CONVTRANS_LAYER if trans else LayerType.CONV_LAYER lt = LayerType.CONVTRANS_LAYER if trans else LayerType.CONV_LAYER
l = Layer( l = Layer(
...@@ -1908,8 +1929,7 @@ def img_pool_layer(input, ...@@ -1908,8 +1929,7 @@ def img_pool_layer(input,
layer_attr=None, layer_attr=None,
pool_size_y=None, pool_size_y=None,
stride_y=None, stride_y=None,
padding_y=None, padding_y=None):
img_width=None):
""" """
Image pooling Layer. Image pooling Layer.
...@@ -1940,9 +1960,6 @@ def img_pool_layer(input, ...@@ -1940,9 +1960,6 @@ def img_pool_layer(input,
:type stride_y: int|None :type stride_y: int|None
:param layer_attr: Extra Layer attribute. :param layer_attr: Extra Layer attribute.
:type layer_attr: ExtraLayerAttribute :type layer_attr: ExtraLayerAttribute
:param img_width: the width of input feature map. If it is None, the input feature
map should be square.
:type img_width: int|None
:return: LayerOutput object. :return: LayerOutput object.
:rtype: LayerOutput :rtype: LayerOutput
""" """
...@@ -1978,8 +1995,7 @@ def img_pool_layer(input, ...@@ -1978,8 +1995,7 @@ def img_pool_layer(input,
padding=padding, padding=padding,
size_y=pool_size_y, size_y=pool_size_y,
stride_y=stride_y, stride_y=stride_y,
padding_y=padding_y, padding_y=padding_y))
img_width=img_width))
], ],
**ExtraLayerAttribute.to_kwargs(layer_attr)) **ExtraLayerAttribute.to_kwargs(layer_attr))
return LayerOutput( return LayerOutput(
...@@ -1997,7 +2013,6 @@ def spp_layer(input, ...@@ -1997,7 +2013,6 @@ def spp_layer(input,
num_channels=None, num_channels=None,
pool_type=None, pool_type=None,
pyramid_height=None, pyramid_height=None,
img_width=None,
layer_attr=None): layer_attr=None):
""" """
Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition.
...@@ -2014,9 +2029,6 @@ def spp_layer(input, ...@@ -2014,9 +2029,6 @@ def spp_layer(input,
:type scale: BasePoolingType :type scale: BasePoolingType
:param pyramid_height: pyramid height. :param pyramid_height: pyramid height.
:type pyramid_height: int :type pyramid_height: int
:param img_width: the width of input feature map. If it is None, the input feature
map should be square.
:type img_width: int|None
:param layer_attr: Extra Layer Attribute. :param layer_attr: Extra Layer Attribute.
:type layer_attr: ExtraLayerAttribute :type layer_attr: ExtraLayerAttribute
:return: LayerOutput object. :return: LayerOutput object.
...@@ -2043,8 +2055,7 @@ def spp_layer(input, ...@@ -2043,8 +2055,7 @@ def spp_layer(input,
spp=SpatialPyramidPool( spp=SpatialPyramidPool(
pool_type=type_name, pool_type=type_name,
channels=num_channels, channels=num_channels,
pyramid_height=pyramid_height, pyramid_height=pyramid_height)),
img_width=img_width)),
**ExtraLayerAttribute.to_kwargs(layer_attr)) **ExtraLayerAttribute.to_kwargs(layer_attr))
return LayerOutput( return LayerOutput(
name, name,
......
python/paddle/trainer_config_helpers/tests/CMakeLists.txt
浏览文件 @ 021b3a44
...@@ -4,7 +4,17 @@ add_test(NAME layers_test ...@@ -4,7 +4,17 @@ add_test(NAME layers_test
python ${PROJ_ROOT}/python/paddle/trainer_config_helpers/tests/layers_test.py python ${PROJ_ROOT}/python/paddle/trainer_config_helpers/tests/layers_test.py
WORKING_DIRECTORY ${PROJ_ROOT}/python/paddle) WORKING_DIRECTORY ${PROJ_ROOT}/python/paddle)
add_test(NAME test_layerHelpers if (PROTOBUF_3)
add_paddle_exe(protobuf_equal
ProtobufEqualMain.cpp)
add_test(NAME test_layerHelpers
COMMAND COMMAND
${PROJ_ROOT}/python/paddle/trainer_config_helpers/tests/configs/run_tests.sh ${PROJ_ROOT}/python/paddle/trainer_config_helpers/tests/configs/run_tests.sh
) ${CMAKE_CURRENT_BINARY_DIR}/protobuf_equal
)
else()
add_test(NAME test_layerHelpers
COMMAND
${PROJ_ROOT}/python/paddle/trainer_config_helpers/tests/configs/run_tests.sh
)
endif()
python/paddle/trainer_config_helpers/tests/ProtobufEqualMain.cpp 0 → 100644
浏览文件 @ 021b3a44
/* Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this 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 <google/protobuf/text_format.h>
#include <google/protobuf/util/message_differencer.h>
#include <fstream>
#include <iostream>
#include "TrainerConfig.pb.h"
bool loadPb(google::protobuf::Message* conf, const std::string& filename) {
std::ifstream fin;
fin.open(filename.c_str());
if (fin.is_open()) {
std::string str((std::istreambuf_iterator<char>(fin)),
std::istreambuf_iterator<char>());
bool ok = google::protobuf::TextFormat::ParseFromString(str, conf);
fin.close();
return ok;
} else {
return false;
}
}
int main(int argc, char** argv) {
std::unique_ptr<google::protobuf::Message> config1;
std::unique_ptr<google::protobuf::Message> config2;
if (argc == 3) {
config1.reset(new paddle::ModelConfig());
config2.reset(new paddle::ModelConfig());
} else if (argc == 4) {
config1.reset(new paddle::TrainerConfig());
config2.reset(new paddle::TrainerConfig());
}
if (!config1 || !config2) {
return 1;
} else if (!loadPb(config1.get(), argv[1])) {
return 2;
} else if (!loadPb(config2.get(), argv[2])) {
return 3;
} else {
if (google::protobuf::util::MessageDifferencer::ApproximatelyEquals(
*config1, *config2)) {
return 0;
} else {
return 4;
}
}
}
python/paddle/trainer_config_helpers/tests/configs/file_list.sh 0 → 100755
浏览文件 @ 021b3a44
#!/bin/bash
export configs=(test_fc layer_activations projections test_print_layer
test_sequence_pooling test_lstmemory_layer test_grumemory_layer
last_first_seq test_expand_layer test_ntm_layers test_hsigmoid
img_layers img_trans_layers util_layers simple_rnn_layers unused_layers test_cost_layers
test_rnn_group shared_fc shared_lstm test_cost_layers_with_weight
test_spp_layer test_bilinear_interp test_maxout test_bi_grumemory math_ops)
export whole_configs=(test_split_datasource)
python/paddle/trainer_config_helpers/tests/configs/generate_protostr.sh
浏览文件 @ 021b3a44
...@@ -5,24 +5,16 @@ cd `dirname $0` ...@@ -5,24 +5,16 @@ cd `dirname $0`
export PYTHONPATH=$PWD/../../../../ export PYTHONPATH=$PWD/../../../../
protostr=$PWD/protostr protostr=$PWD/protostr
. file_list.sh
configs=(test_fc layer_activations projections test_print_layer
test_sequence_pooling test_lstmemory_layer test_grumemory_layer
last_first_seq test_expand_layer test_ntm_layers test_hsigmoid
img_layers img_trans_layers util_layers simple_rnn_layers unused_layers test_cost_layers
test_rnn_group shared_fc shared_lstm test_cost_layers_with_weight
test_spp_layer test_bilinear_interp test_maxout test_bi_grumemory math_ops)
whole_configs=(test_split_datasource)
for conf in ${configs[*]} for conf in ${configs[*]}
do do
echo "Generating " $conf echo "Generating " $conf
python -m paddle.utils.dump_config $conf.py > $protostr/$conf.protostr.unitest python -m paddle.utils.dump_config $conf.py > $protostr/$conf.protostr.unittest
done done
for conf in ${whole_configs[*]} for conf in ${whole_configs[*]}
do do
echo "Generating " $conf echo "Generating " $conf
python -m paddle.utils.dump_config $conf.py "" --whole > $protostr/$conf.protostr.unitest python -m paddle.utils.dump_config $conf.py "" --whole > $protostr/$conf.protostr.unittest
done done
python/paddle/trainer_config_helpers/tests/configs/protostr/img_layers.protostr
浏览文件 @ 021b3a44
...@@ -26,11 +26,15 @@ layers { ...@@ -26,11 +26,15 @@ layers {
filter_size_y: 32 filter_size_y: 32
padding_y: 1 padding_y: 1
stride_y: 1 stride_y: 1
output_y: 227
img_size_y: 256
} }
} }
bias_parameter_name: "___conv_0__.wbias" bias_parameter_name: "___conv_0__.wbias"
num_filters: 64 num_filters: 64
shared_biases: true shared_biases: true
height: 227
width: 227
} }
layers { layers {
name: "__batch_norm_0__" name: "__batch_norm_0__"
...@@ -43,6 +47,7 @@ layers { ...@@ -43,6 +47,7 @@ layers {
image_conf { image_conf {
channels: 64 channels: 64
img_size: 227 img_size: 227
img_size_y: 227
} }
} }
inputs { inputs {
...@@ -55,6 +60,8 @@ layers { ...@@ -55,6 +60,8 @@ layers {
} }
bias_parameter_name: "___batch_norm_0__.wbias" bias_parameter_name: "___batch_norm_0__.wbias"
moving_average_fraction: 0.9 moving_average_fraction: 0.9
height: 227
width: 227
} }
layers { layers {
name: "__crmnorm_0__" name: "__crmnorm_0__"
...@@ -72,8 +79,12 @@ layers { ...@@ -72,8 +79,12 @@ layers {
output_x: 227 output_x: 227
img_size: 227 img_size: 227
blocked: false blocked: false
output_y: 227
img_size_y: 227
} }
} }
height: 227
width: 227
} }
layers { layers {
name: "__pool_0__" name: "__pool_0__"
...@@ -97,6 +108,8 @@ layers { ...@@ -97,6 +108,8 @@ layers {
padding_y: 0 padding_y: 0
} }
} }
height: 196
width: 196
} }
parameters { parameters {
name: "___conv_0__.w0" name: "___conv_0__.w0"
......
python/paddle/trainer_config_helpers/tests/configs/protostr/img_trans_layers.protostr
浏览文件 @ 021b3a44
...@@ -26,6 +26,8 @@ layers { ...@@ -26,6 +26,8 @@ layers {
filter_size_y: 32 filter_size_y: 32
padding_y: 1 padding_y: 1
stride_y: 1 stride_y: 1
output_y: 227
img_size_y: 256
} }
} }
bias_parameter_name: "___conv_0__.wbias" bias_parameter_name: "___conv_0__.wbias"
...@@ -43,6 +45,7 @@ layers { ...@@ -43,6 +45,7 @@ layers {
image_conf { image_conf {
channels: 64 channels: 64
img_size: 256 img_size: 256
img_size_y: 256
} }
} }
inputs { inputs {
...@@ -55,6 +58,8 @@ layers { ...@@ -55,6 +58,8 @@ layers {
} }
bias_parameter_name: "___batch_norm_0__.wbias" bias_parameter_name: "___batch_norm_0__.wbias"
moving_average_fraction: 0.9 moving_average_fraction: 0.9
height: 256
width: 256
} }
layers { layers {
name: "__crmnorm_0__" name: "__crmnorm_0__"
...@@ -72,8 +77,12 @@ layers { ...@@ -72,8 +77,12 @@ layers {
output_x: 256 output_x: 256
img_size: 256 img_size: 256
blocked: false blocked: false
output_y: 256
img_size_y: 256
} }
} }
height: 256
width: 256
} }
layers { layers {
name: "__pool_0__" name: "__pool_0__"
...@@ -97,6 +106,8 @@ layers { ...@@ -97,6 +106,8 @@ layers {
padding_y: 0 padding_y: 0
} }
} }
height: 225
width: 225
} }
parameters { parameters {
name: "___conv_0__.w0" name: "___conv_0__.w0"
......
python/paddle/trainer_config_helpers/tests/configs/protostr/projections.protostr
浏览文件 @ 021b3a44
...@@ -177,6 +177,8 @@ layers { ...@@ -177,6 +177,8 @@ layers {
filter_size_y: 3 filter_size_y: 3
padding_y: 0 padding_y: 0
stride_y: 1 stride_y: 1
output_y: 30
img_size_y: 32
} }
num_filters: 64 num_filters: 64
} }
......
python/paddle/trainer_config_helpers/tests/configs/protostr/test_bilinear_interp.protostr
浏览文件 @ 021b3a44
...@@ -26,11 +26,15 @@ layers { ...@@ -26,11 +26,15 @@ layers {
filter_size_y: 3 filter_size_y: 3
padding_y: 1 padding_y: 1
stride_y: 1 stride_y: 1
output_y: 48
img_size_y: 48
} }
} }
bias_parameter_name: "___conv_0__.wbias" bias_parameter_name: "___conv_0__.wbias"
num_filters: 16 num_filters: 16
shared_biases: true shared_biases: true
height: 48
width: 48
} }
layers { layers {
name: "__bilinear_interp_layer_0__" name: "__bilinear_interp_layer_0__"
...@@ -40,11 +44,17 @@ layers { ...@@ -40,11 +44,17 @@ layers {
inputs { inputs {
input_layer_name: "__conv_0__" input_layer_name: "__conv_0__"
bilinear_interp_conf { bilinear_interp_conf {
image_conf {
channels: 16
img_size: 48
img_size_y: 48
}
out_size_x: 64 out_size_x: 64
out_size_y: 64 out_size_y: 64
num_channels: 16
} }
} }
height: 64
width: 64
} }
layers { layers {
name: "__pool_0__" name: "__pool_0__"
...@@ -55,19 +65,21 @@ layers { ...@@ -55,19 +65,21 @@ layers {
input_layer_name: "__bilinear_interp_layer_0__" input_layer_name: "__bilinear_interp_layer_0__"
pool_conf { pool_conf {
pool_type: "max-projection" pool_type: "max-projection"
channels: 4 channels: 16
size_x: 2 size_x: 2
stride: 2 stride: 2
output_x: 64 output_x: 32
img_size: 128 img_size: 64
padding: 0 padding: 0
size_y: 2 size_y: 2
stride_y: 2 stride_y: 2
output_y: 64 output_y: 32
img_size_y: 128 img_size_y: 64
padding_y: 0 padding_y: 0
} }
} }
height: 32
width: 32
} }
layers { layers {
name: "__fc_layer_0__" name: "__fc_layer_0__"
...@@ -78,6 +90,8 @@ layers { ...@@ -78,6 +90,8 @@ layers {
input_layer_name: "__pool_0__" input_layer_name: "__pool_0__"
input_parameter_name: "___fc_layer_0__.w0" input_parameter_name: "___fc_layer_0__.w0"
} }
height: 32
width: 32
} }
parameters { parameters {
name: "___conv_0__.w0" name: "___conv_0__.w0"
......
python/paddle/trainer_config_helpers/tests/configs/protostr/test_maxout.protostr
浏览文件 @ 021b3a44
...@@ -4,6 +4,8 @@ layers { ...@@ -4,6 +4,8 @@ layers {
type: "data" type: "data"
size: 2304 size: 2304
active_type: "" active_type: ""
height: 48
width: 48
} }
layers { layers {
name: "__conv_0__" name: "__conv_0__"
...@@ -26,11 +28,15 @@ layers { ...@@ -26,11 +28,15 @@ layers {
filter_size_y: 3 filter_size_y: 3
padding_y: 1 padding_y: 1
stride_y: 1 stride_y: 1
output_y: 48
img_size_y: 48
} }
} }
bias_parameter_name: "___conv_0__.wbias" bias_parameter_name: "___conv_0__.wbias"
num_filters: 16 num_filters: 16
shared_biases: true shared_biases: true
height: 48
width: 48
} }
layers { layers {
name: "__maxout_layer_0__" name: "__maxout_layer_0__"
...@@ -40,12 +46,16 @@ layers { ...@@ -40,12 +46,16 @@ layers {
inputs { inputs {
input_layer_name: "__conv_0__" input_layer_name: "__conv_0__"
maxout_conf { maxout_conf {
image_conf {
channels: 16 channels: 16
img_size: 48
img_size_y: 48
}
groups: 2 groups: 2
img_size_x: 0
img_size_y: 0
} }
} }
height: 48
width: 48
} }
layers { layers {
name: "__pool_0__" name: "__pool_0__"
...@@ -69,48 +79,58 @@ layers { ...@@ -69,48 +79,58 @@ layers {
padding_y: 0 padding_y: 0
} }
} }
height: 24
width: 24
} }
layers { layers {
name: "__conv_1__" name: "__conv_1__"
type: "exconv" type: "exconv"
size: 18432 size: 73728
active_type: "" active_type: ""
inputs { inputs {
input_layer_name: "__pool_0__" input_layer_name: "__pool_0__"
input_parameter_name: "___conv_1__.w0" input_parameter_name: "___conv_1__.w0"
conv_conf { conv_conf {
filter_size: 3 filter_size: 3
channels: 32 channels: 8
stride: 1 stride: 1
padding: 1 padding: 1
groups: 1 groups: 1
filter_channels: 32 filter_channels: 8
output_x: 12 output_x: 24
img_size: 12 img_size: 24
caffe_mode: true caffe_mode: true
filter_size_y: 3 filter_size_y: 3
padding_y: 1 padding_y: 1
stride_y: 1 stride_y: 1
output_y: 24
img_size_y: 24
} }
} }
bias_parameter_name: "___conv_1__.wbias" bias_parameter_name: "___conv_1__.wbias"
num_filters: 128 num_filters: 128
shared_biases: true shared_biases: true
height: 24
width: 24
} }
layers { layers {
name: "__maxout_layer_1__" name: "__maxout_layer_1__"
type: "maxout" type: "maxout"
size: 9216 size: 18432
active_type: "" active_type: ""
inputs { inputs {
input_layer_name: "__conv_0__" input_layer_name: "__conv_1__"
maxout_conf { maxout_conf {
image_conf {
channels: 128 channels: 128
img_size: 24
img_size_y: 24
}
groups: 4 groups: 4
img_size_x: 0
img_size_y: 0
} }
} }
height: 24
width: 24
} }
layers { layers {
name: "__block_expand_layer_0__" name: "__block_expand_layer_0__"
...@@ -118,7 +138,7 @@ layers { ...@@ -118,7 +138,7 @@ layers {
size: 192 size: 192
active_type: "" active_type: ""
inputs { inputs {
input_layer_name: "__maxout_layer_0__" input_layer_name: "__maxout_layer_1__"
block_expand_conf { block_expand_conf {
channels: 32 channels: 32
stride_x: 1 stride_x: 1
...@@ -133,6 +153,8 @@ layers { ...@@ -133,6 +153,8 @@ layers {
img_size_y: 0 img_size_y: 0
} }
} }
height: 24
width: 24
} }
layers { layers {
name: "__fc_layer_0__" name: "__fc_layer_0__"
...@@ -143,6 +165,8 @@ layers { ...@@ -143,6 +165,8 @@ layers {
input_layer_name: "__block_expand_layer_0__" input_layer_name: "__block_expand_layer_0__"
input_parameter_name: "___fc_layer_0__.w0" input_parameter_name: "___fc_layer_0__.w0"
} }
height: 24
width: 24
} }
parameters { parameters {
name: "___conv_0__.w0" name: "___conv_0__.w0"
...@@ -164,9 +188,9 @@ parameters { ...@@ -164,9 +188,9 @@ parameters {
} }
parameters { parameters {
name: "___conv_1__.w0" name: "___conv_1__.w0"
size: 36864 size: 9216
initial_mean: 0.0 initial_mean: 0.0
initial_std: 0.0833333333333 initial_std: 0.166666666667
initial_strategy: 0 initial_strategy: 0
initial_smart: false initial_smart: false
} }
......
python/paddle/trainer_config_helpers/tests/configs/protostr/test_spp_layer.protostr
浏览文件 @ 021b3a44
...@@ -4,6 +4,8 @@ layers { ...@@ -4,6 +4,8 @@ layers {
type: "data" type: "data"
size: 3200 size: 3200
active_type: "" active_type: ""
height: 20
width: 10
} }
layers { layers {
name: "__spp_0__" name: "__spp_0__"
...@@ -13,13 +15,17 @@ layers { ...@@ -13,13 +15,17 @@ layers {
inputs { inputs {
input_layer_name: "data" input_layer_name: "data"
spp_conf { spp_conf {
pool_type: "max-projection" image_conf {
pyramid_height: 2
channels: 16 channels: 16
img_size: 10 img_size: 10
img_size_y: 20 img_size_y: 20
} }
pool_type: "max-projection"
pyramid_height: 2
}
} }
height: 1
width: 5
} }
input_layer_names: "data" input_layer_names: "data"
output_layer_names: "__spp_0__" output_layer_names: "__spp_0__"
......
python/paddle/trainer_config_helpers/tests/configs/run_tests.sh
浏览文件 @ 021b3a44
...@@ -5,13 +5,31 @@ set -e ...@@ -5,13 +5,31 @@ set -e
protostr=`dirname $0`/protostr protostr=`dirname $0`/protostr
files=`ls $protostr | grep -v "unitest"` files=`ls $protostr | grep -v "unittest"`
./generate_protostr.sh ./generate_protostr.sh
for file in $files . ./file_list.sh
do
if [ -z $1 ]; then
for file in $files
do
base_protostr=$protostr/$file base_protostr=$protostr/$file
new_protostr=$protostr/$file.unitest new_protostr=$protostr/$file.unittest
diff $base_protostr $new_protostr -u diff $base_protostr $new_protostr -u
done done
else
for file in ${configs[*]}
do
if ! $1 $protostr/$file.protostr $protostr/$file.protostr.unittest; then
diff $protostr/$file.protostr $protostr/$file.protostr.unittest -u
fi
done
for file in ${whole_configs[*]}
do
if ! $1 $protostr/$file.protostr $protostr/$file.protostr.unittest --whole; then
diff $protostr/$file.protostr $protostr/$file.protostr.unittest -u
fi
done
fi
python/paddle/trainer_config_helpers/tests/configs/test_bilinear_interp.py
浏览文件 @ 021b3a44
...@@ -17,7 +17,7 @@ bilinear = bilinear_interp_layer(input=conv, out_size_x=64, out_size_y=64) ...@@ -17,7 +17,7 @@ bilinear = bilinear_interp_layer(input=conv, out_size_x=64, out_size_y=64)
pool = img_pool_layer( pool = img_pool_layer(
input=bilinear, input=bilinear,
num_channels=4, num_channels=16,
pool_size=2, pool_size=2,
stride=2, stride=2,
pool_type=MaxPooling()) pool_type=MaxPooling())
......
python/paddle/trainer_config_helpers/tests/configs/test_maxout.py
浏览文件 @ 021b3a44
...@@ -2,7 +2,7 @@ from paddle.trainer_config_helpers import * ...@@ -2,7 +2,7 @@ from paddle.trainer_config_helpers import *
settings(batch_size=1000, learning_rate=1e-5) settings(batch_size=1000, learning_rate=1e-5)
data = data_layer(name='data', size=2304) data = data_layer(name='data', size=2304, height=48, width=48)
conv = img_conv_layer( conv = img_conv_layer(
input=data, input=data,
...@@ -21,16 +21,21 @@ pool = img_pool_layer( ...@@ -21,16 +21,21 @@ pool = img_pool_layer(
conv2 = img_conv_layer( conv2 = img_conv_layer(
input=pool, input=pool,
filter_size=3, filter_size=3,
num_channels=32, num_channels=8,
num_filters=128, num_filters=128,
padding=1, padding=1,
act=LinearActivation(), act=LinearActivation(),
bias_attr=True) bias_attr=True)
maxout2 = maxout_layer(input=conv, num_channels=128, groups=4) maxout2 = maxout_layer(input=conv2, num_channels=128, groups=4)
block = block_expand_layer( block = block_expand_layer(
input=maxout, num_channels=32, stride_x=1, stride_y=1, block_x=1, block_y=6) input=maxout2,
num_channels=32,
stride_x=1,
stride_y=1,
block_x=1,
block_y=6)
fc = fc_layer(input=block, size=384, bias_attr=False) fc = fc_layer(input=block, size=384, bias_attr=False)
......
python/paddle/trainer_config_helpers/tests/configs/test_spp_layer.py
浏览文件 @ 021b3a44
...@@ -2,13 +2,9 @@ from paddle.trainer_config_helpers import * ...@@ -2,13 +2,9 @@ from paddle.trainer_config_helpers import *
settings(batch_size=100, learning_rate=1e-5) settings(batch_size=100, learning_rate=1e-5)
data = data_layer(name='data', size=3200) data = data_layer(name='data', size=3200, height=20, width=10)
spp = spp_layer( spp = spp_layer(
input=data, input=data, pyramid_height=2, num_channels=16, pool_type=MaxPooling())
pyramid_height=2,
num_channels=16,
pool_type=MaxPooling(),
img_width=10)
outputs(spp) outputs(spp)
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或