Skip to content

P
PaddleDetection

PaddlePaddle / PaddleDetection
大约 1 年 前同步成功

通知 694
Star 11112
Fork 2696
P
PaddleDetection

体验新版 GitCode，发现更多精彩内容 >>

提交 b23d99dd 编写于 作者: Y Yu Yang
浏览文件
操作

Merge branch 'develop' of github.com:baidu/Paddle into feature/add_sum_cost_in_args

上级 343d9973 748229aa
展开全部 隐藏空白更改
内联 并排
Showing with 5896 addition and 1987 deletion
.gitignore
浏览文件 @ b23d99dd
......@@ -9,6 +9,8 @@ build/
.pydevproject
Makefile
.test_env/
third_party/
*~
bazel-*
third_party/
.gitmodules 已删除 100644 → 0
浏览文件 @ 343d9973
[submodule "warp-ctc"]
path = warp-ctc
url = https://github.com/baidu-research/warp-ctc.git
.pre-commit-config.yaml
浏览文件 @ b23d99dd
......@@ -2,7 +2,7 @@
sha: c25201a00e6b0514370501050cf2a8538ac12270
hooks:
- id: remove-crlf
files: (?!.*warp-ctc)^.*$
files: (?!.*third_party)^.*$
- repo: https://github.com/reyoung/mirrors-yapf.git
sha: v0.13.2
hooks:
......@@ -15,7 +15,7 @@
- id: check-merge-conflict
- id: check-symlinks
- id: detect-private-key
files: (?!.*warp-ctc)^.*$
files: (?!.*third_party)^.*$
- id: end-of-file-fixer
- repo: https://github.com/PaddlePaddle/clang-format-pre-commit-hook.git
sha: 28c0ea8a67a3e2dbbf4822ef44e85b63a0080a29
......
.travis.yml
浏览文件 @ b23d99dd
......@@ -21,23 +21,17 @@ addons:
packages:
- gcc-4.8
- g++-4.8
- wget
- git
- build-essential
- libatlas-base-dev
- python
- python-pip
- python2.7-dev
- m4
- python-numpy
- python-wheel
- libgoogle-glog-dev
- libgflags-dev
- libgtest-dev
- curl
- lcov
- graphviz
- swig
- graphviz
- clang-format-3.8
- automake
- libtool
......@@ -53,10 +47,9 @@ before_install:
fi
fi
fi
- if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then sudo paddle/scripts/travis/before_install.linux.sh; fi
- if [[ "$TRAVIS_OS_NAME" == "osx" ]]; then paddle/scripts/travis/before_install.osx.sh; fi
- if [[ "$JOB" == "PRE_COMMIT" ]]; then sudo ln -s /usr/bin/clang-format-3.8 /usr/bin/clang-format; fi
- pip install wheel protobuf sphinx recommonmark virtualenv numpy sphinx_rtd_theme pre-commit requests==2.9.2 LinkChecker
- pip install numpy wheel protobuf sphinx recommonmark sphinx_rtd_theme virtualenv pre-commit requests==2.9.2 LinkChecker
script:
- paddle/scripts/travis/main.sh
notifications:
......
CMakeLists.txt
浏览文件 @ b23d99dd
cmake_minimum_required(VERSION 2.8)
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License
cmake_minimum_required(VERSION 3.0)
project(paddle CXX C)
set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake")
set(PROJ_ROOT ${CMAKE_SOURCE_DIR})
include(package)
find_package(SWIG 2.0)
find_package(CUDA QUIET)
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(PythonInterp 2.7 REQUIRED)
find_package(ZLIB REQUIRED)
find_package(NumPy REQUIRED)
find_package(Threads REQUIRED)
find_package(AVX QUIET)
find_package(Glog REQUIRED)
find_package(Gflags REQUIRED)
find_package(GTest)
find_package(Sphinx)
find_package(Doxygen)
include(cblas)
find_program(M4_EXECUTABLE m4)
###################### Configurations ###########################
option(WITH_DSO "Compile PaddlePaddle with dynamic linked libraries" ON)
option(WITH_GPU "Compile PaddlePaddle with gpu" ${CUDA_FOUND})
option(WITH_DOUBLE "Compile PaddlePaddle with double precision, otherwise use single precision" OFF)
option(WITH_AVX "Compile PaddlePaddle with avx intrinsics" ${AVX_FOUND})
option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON)
option(WITH_STYLE_CHECK "Style Check for PaddlePaddle" ${PYTHONINTERP_FOUND})
option(WITH_RDMA "Compile PaddlePaddle with rdma support" OFF)
option(WITH_TIMER "Compile PaddlePaddle use timer" OFF)
option(WITH_PROFILER "Compile PaddlePaddle use gpu profiler" OFF)
option(WITH_TESTING "Compile and run unittest for PaddlePaddle" ${GTEST_FOUND})
option(WITH_DOC "Compile PaddlePaddle with documentation" OFF)
option(WITH_SWIG_PY "Compile PaddlePaddle with py PaddlePaddle prediction api" ${SWIG_FOUND})
option(ON_TRAVIS "Running test on travis-ci or not." OFF)
option(ON_COVERALLS "Generating code coverage data on coveralls or not." OFF)
option(COVERALLS_UPLOAD "Uploading the generated coveralls json." ON)
include(cpplint)
include(ccache)
if(WITH_RDMA)
include(rdma)
endif()
include(util)
include(flags)
include(cudnn)
include(FindPythonModule)
include(check_packages)
include(swig)
include(coveralls)
# Set PaddlePaddle version to Git tag name or Git commit ID.
find_package(CUDA QUIET)
find_package(Git REQUIRED)
# version.cmake will get the current PADDLE_VERSION
include(version)
add_definitions(-DPADDLE_VERSION=${PADDLE_VERSION})
if(NOT WITH_GPU)
add_definitions(-DPADDLE_ONLY_CPU)
add_definitions(-DHPPL_STUB_FUNC)
list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
else()
if(${CUDA_VERSION_MAJOR} VERSION_LESS 7)
message(FATAL_ERROR "Paddle need CUDA >= 7.0 to compile")
endif()
if(NOT CUDNN_FOUND)
message(FATAL_ERROR "Paddle need cudnn to compile")
endif()
if(WITH_AVX)
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${AVX_FLAG}")
else(WITH_AVX)
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SSE3_FLAG}")
endif(WITH_AVX)
# Include cuda and cudnn
include_directories(${CUDNN_INCLUDE_DIR})
include_directories(${CUDA_TOOLKIT_INCLUDE})
endif(NOT WITH_GPU)
if(WITH_DSO)
add_definitions(-DPADDLE_USE_DSO)
endif(WITH_DSO)
if(WITH_DOUBLE)
add_definitions(-DPADDLE_TYPE_DOUBLE)
set(ACCURACY double)
else(WITH_DOUBLE)
set(ACCURACY float)
endif(WITH_DOUBLE)
if(NOT WITH_TIMER)
add_definitions(-DPADDLE_DISABLE_TIMER)
endif(NOT WITH_TIMER)
if(NOT WITH_PROFILER)
add_definitions(-DPADDLE_DISABLE_PROFILER)
endif(NOT WITH_PROFILER)
if(WITH_AVX)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${AVX_FLAG}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${AVX_FLAG}")
else(WITH_AVX)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${SSE3_FLAG}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SSE3_FLAG}")
endif(WITH_AVX)
if(WITH_PYTHON)
include_directories(${PYTHON_INCLUDE_DIR})
include_directories(${PYTHON_NUMPY_INCLUDE_DIR})
else(WITH_PYTHON)
add_definitions(-DPADDLE_NO_PYTHON)
endif(WITH_PYTHON)
if(WITH_RDMA)
include_directories("${RDMA_INC_DIR}")
else(WITH_RDMA)
add_definitions(-DPADDLE_DISABLE_RDMA)
endif(WITH_RDMA)
# glog
include_directories(${LIBGLOG_INCLUDE_DIR})
#gflags
add_definitions(-DGFLAGS_NS=${GFLAGS_NAMESPACE})
include_directories(${GFLAGS_INCLUDE_DIRS})
find_package(Threads REQUIRED)
if(WITH_TESTING)
enable_testing()
include_directories(${GTEST_INCLUDE_DIRS})
endif()
include(system)
include(simd)
################################ Configurations #######################################
option(WITH_GPU "Compile PaddlePaddle with NVIDIA GPU" ${CUDA_FOUND})
option(WITH_AVX "Compile PaddlePaddle with AVX intrinsics" ${AVX_FOUND})
option(WITH_DSO "Compile PaddlePaddle with dynamic linked CUDA" OFF)
option(WITH_TESTING "Compile PaddlePaddle with unit testing" ON)
option(WITH_SWIG_PY "Compile PaddlePaddle with inference api" ON)
option(WITH_STYLE_CHECK "Compile PaddlePaddle with style check" ON)
option(WITH_PYTHON "Compile PaddlePaddle with python interpreter" ON)
option(WITH_DOUBLE "Compile PaddlePaddle with double precision" OFF)
option(WITH_RDMA "Compile PaddlePaddle with RDMA support" OFF)
option(WITH_TIMER "Compile PaddlePaddle with stats timer" OFF)
option(WITH_PROFILER "Compile PaddlePaddle with GPU profiler" OFF)
option(WITH_DOC "Compile PaddlePaddle with documentation" OFF)
option(ON_COVERALLS "Compile PaddlePaddle with code coverage" OFF)
option(COVERALLS_UPLOAD "Package code coverage data to coveralls" OFF)
option(ON_TRAVIS "Exclude special unit test on Travis CI" OFF)
########################################################################################
include(external/zlib) # download, build, install zlib
include(external/gflags) # download, build, install gflags
include(external/glog) # download, build, install glog
include(external/gtest) # download, build, install gtest
include(external/protobuf) # download, build, install protobuf
include(external/python) # download, build, install python
include(external/openblas) # download, build, install openblas
include(external/swig) # download, build, install swig
include(external/warpctc) # download, build, install warpctc
include(package) # set paddle packages
include(cpplint) # set paddle c++ style
include(ccache) # set ccache for compilation
include(util) # set unittest and link libs
include(rdma) # set rdma libraries
include(flags) # set paddle compile flags
include(cudnn) # set cudnn libraries
include(version) # set PADDLE_VERSION
include(coveralls) # set code coverage
include(configure) # add paddle env configuration
include_directories("${CBLAS_INC_DIR}")
include_directories("${PROJ_ROOT}")
include_directories("${PROJ_ROOT}/paddle/cuda/include")
include_directories(${PROTOBUF_INCLUDE_DIRS})
include_directories("${CMAKE_CURRENT_BINARY_DIR}/proto")
if(EXISTS "${PROJ_ROOT}/paddle/internals/CMakeLists.txt")
set(PADDLE_WITH_INTERNAL ON)
include(paddle/internals/CMakeLists.txt)
else()
set(PADDLE_WITH_INTERNAL OFF)
set(INTERNAL_PROTO_PATH "")
endif()
set(EXTERNAL_LIBS
# have not include gtest here.
${GFLAGS_LIBRARIES}
${GLOG_LIBRARIES}
${CBLAS_LIBRARIES}
${PROTOBUF_LIBRARY}
${ZLIB_LIBRARIES}
)
add_subdirectory(proto)
add_subdirectory(paddle)
add_subdirectory(python)
if(WITH_DOC)
add_subdirectory(doc)
endif()
cmake/cblas.cmake
浏览文件 @ b23d99dd
......@@ -13,6 +13,7 @@
# system paths.
#
set(CBLAS_FOUND OFF)
## Find MKL First.
set(MKL_ROOT $ENV{MKL_ROOT} CACHE PATH "Folder contains MKL")
......@@ -35,11 +36,12 @@ find_library(MKL_INTEL_LP64 NAMES mkl_intel_lp64 PATHS
if(MKL_INCLUDE_DIR AND MKL_CORE_LIB AND MKL_SEQUENTIAL_LIB AND MKL_INTEL_LP64)
set(CBLAS_PROVIDER MKL)
set(CBLAS_INC_DIR ${MKL_INCLUDE_DIR})
set(CBLAS_LIBS ${MKL_INTEL_LP64}
set(CBLAS_LIBRARIES ${MKL_INTEL_LP64}
${MKL_SEQUENTIAL_LIB}
${MKL_CORE_LIB})
add_definitions(-DPADDLE_USE_MKL)
message(STATUS "Found MKL (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBS})")
message(STATUS "Found MKL (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBRARIES})")
set(CBLAS_FOUND ON)
return() # return file.
endif()
......@@ -68,9 +70,10 @@ find_library(ATLAS_LIB NAMES lapack_atlas liblapack_atlas.so.3
if(ATLAS_INC_DIR AND ATLAS_CBLAS_LIB AND ATLAS_LIB)
set(CBLAS_PROVIDER ATLAS)
set(CBLAS_INC_DIR ${ATLAS_INC_DIR} ${ATLAS_CLAPACK_INC_DIR})
set(CBLAS_LIBS ${ATLAS_LIB} ${ATLAS_CBLAS_LIB})
set(CBLAS_LIBRARIES ${ATLAS_LIB} ${ATLAS_CBLAS_LIB})
add_definitions(-DPADDLE_USE_ATLAS)
message(STATUS "Found Atlas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBS})")
message(STATUS "Found Atlas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBRARIES})")
set(CBLAS_FOUND ON)
return()
endif()
......@@ -98,8 +101,9 @@ find_library(OPENBLAS_LIB NAMES openblas
if(OPENBLAS_INC_DIR AND OPENBLAS_LIB)
set(CBLAS_PROVIDER OPENBLAS)
set(CBLAS_INC_DIR ${OPENBLAS_INC_DIR})
set(CBLAS_LIBS ${OPENBLAS_LIB})
message(STATUS "Found OpenBlas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBS})")
set(CBLAS_LIBRARIES ${OPENBLAS_LIB})
message(STATUS "Found OpenBlas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBRARIES})")
set(CBLAS_FOUND ON)
return()
endif()
......@@ -130,9 +134,7 @@ find_library(REFERENCE_CBLAS_LIBRARY NAMES cblas PATHS
if (REFERENCE_CBLAS_INCLUDE_DIR AND REFERENCE_CBLAS_LIBRARY)
set(CBLAS_PROVIDER REFERENCE)
set(CBLAS_INC_DIR ${REFERENCE_CBLAS_INCLUDE_DIR})
set(CBLAS_LIBS ${REFERENCE_CBLAS_LIBRARY})
return()
set(CBLAS_LIBRARIES ${REFERENCE_CBLAS_LIBRARY})
message(STATUS "Found reference-cblas (include: ${CBLAS_INC_DIR}, library: ${CBLAS_LIBS})")
set(CBLAS_FOUND ON)
endif()
message(FATAL_ERROR "CBlas must be set. Paddle support MKL, ATLAS, OpenBlas, reference-cblas."
" Try set MKL_ROOT, ATLAS_ROOT, OPENBLAS_ROOT or REFERENCE_CBLAS_ROOT.")
cmake/check_packages.cmake 已删除 100644 → 0
浏览文件 @ 343d9973
# Check package for each cmake option
if(WITH_GPU)
find_package(CUDA REQUIRED) # CUDA is required when use gpu
endif()
if(WITH_PYTHON)
find_package(PythonLibs 2.6 REQUIRED)
find_package(PythonInterp REQUIRED)
find_package(NumPy REQUIRED)
endif()
if(WITH_STYLE_CHECK)
find_package(PythonInterp REQUIRED)
endif()
find_package(Glog REQUIRED)
find_package(Gflags REQUIRED)
if(WITH_TESTING)
find_package(GTest REQUIRED)
endif()
if(WITH_DOC)
find_package(Sphinx REQUIRED)
find_python_module(recommonmark REQUIRED)
endif()
if(WITH_SWIG_PY)
if(NOT SWIG_FOUND)
message(FATAL_ERROR "SWIG is not found. Please install swig or disable WITH_SWIG_PY")
endif()
find_python_module(wheel REQUIRED) # package wheel
endif()
if(NOT M4_EXECUTABLE)
message(FATAL_ERROR "Paddle need m4 to generate proto file.")
endif()
cmake/configure.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
if(NOT WITH_PYTHON)
add_definitions(-DPADDLE_NO_PYTHON)
endif(NOT WITH_PYTHON)
if(WITH_DSO)
add_definitions(-DPADDLE_USE_DSO)
endif(WITH_DSO)
if(WITH_DOUBLE)
add_definitions(-DPADDLE_TYPE_DOUBLE)
endif(WITH_DOUBLE)
if(NOT WITH_TIMER)
add_definitions(-DPADDLE_DISABLE_TIMER)
endif(NOT WITH_TIMER)
if(NOT WITH_PROFILER)
add_definitions(-DPADDLE_DISABLE_PROFILER)
endif(NOT WITH_PROFILER)
if(NOT WITH_GPU)
add_definitions(-DPADDLE_ONLY_CPU)
add_definitions(-DHPPL_STUB_FUNC)
list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
else()
FIND_PACKAGE(CUDA REQUIRED)
if(${CUDA_VERSION_MAJOR} VERSION_LESS 7)
message(FATAL_ERROR "Paddle need CUDA >= 7.0 to compile")
endif()
if(NOT CUDNN_FOUND)
message(FATAL_ERROR "Paddle need cudnn to compile")
endif()
if(WITH_AVX)
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${AVX_FLAG}")
else(WITH_AVX)
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} "-Xcompiler ${SSE3_FLAG}")
endif(WITH_AVX)
# Include cuda and cudnn
include_directories(${CUDNN_INCLUDE_DIR})
include_directories(${CUDA_TOOLKIT_INCLUDE})
endif(NOT WITH_GPU)
if(WITH_AVX)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${AVX_FLAG}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${AVX_FLAG}")
else(WITH_AVX)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${SSE3_FLAG}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${SSE3_FLAG}")
endif(WITH_AVX)
cmake/cpplint.cmake
浏览文件 @ b23d99dd
......@@ -53,7 +53,7 @@ macro(add_style_check_target TARGET_NAME)
if(LINT MATCHES ON)
add_custom_command(TARGET ${TARGET_NAME}
PRE_BUILD
COMMAND "${PYTHON_EXECUTABLE}" "${PROJ_ROOT}/paddle/scripts/cpplint.py"
COMMAND env ${py_env} "${PYTHON_EXECUTABLE}" "${PROJ_ROOT}/paddle/scripts/cpplint.py"
"--filter=${STYLE_FILTER}" ${filename}
WORKING_DIRECTORY ${CMAKE_CURRENT_LIST_DIR})
endif()
......
cmake/external/gflags.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(ExternalProject)
SET(GFLAGS_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/gflags)
SET(GFLAGS_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/gflags)
SET(GFLAGS_INCLUDE_DIR "${GFLAGS_INSTALL_DIR}/include" CACHE PATH "gflags include directory." FORCE)
IF(WIN32)
set(GFLAGS_LIBRARIES "${GFLAGS_INSTALL_DIR}/lib/gflags.lib" CACHE FILEPATH "GFLAGS_LIBRARIES" FORCE)
ELSE(WIN32)
set(GFLAGS_LIBRARIES "${GFLAGS_INSTALL_DIR}/lib/libgflags.a" CACHE FILEPATH "GFLAGS_LIBRARIES" FORCE)
ENDIF(WIN32)
INCLUDE_DIRECTORIES(${GFLAGS_INCLUDE_DIR})
ExternalProject_Add(
gflags
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/gflags/gflags.git"
PREFIX ${GFLAGS_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${GFLAGS_INSTALL_DIR}
CMAKE_ARGS -DCMAKE_POSITION_INDEPENDENT_CODE=ON
CMAKE_ARGS -DBUILD_TESTING=OFF
)
LIST(APPEND external_project_dependencies gflags)
cmake/external/glog.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(ExternalProject)
SET(GLOG_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/glog)
SET(GLOG_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/glog)
SET(GLOG_INCLUDE_DIR "${GLOG_INSTALL_DIR}/include" CACHE PATH "glog include directory." FORCE)
IF(WIN32)
SET(GLOG_LIBRARIES "${GLOG_INSTALL_DIR}/lib/libglog.lib" CACHE FILEPATH "glog library." FORCE)
ELSE(WIN32)
SET(GLOG_LIBRARIES "${GLOG_INSTALL_DIR}/lib/libglog.a" CACHE FILEPATH "glog library." FORCE)
ENDIF(WIN32)
INCLUDE_DIRECTORIES(${GLOG_INCLUDE_DIR})
ExternalProject_Add(
glog
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/google/glog.git"
PREFIX ${GLOG_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${GLOG_INSTALL_DIR}
CMAKE_ARGS -DCMAKE_POSITION_INDEPENDENT_CODE=ON
CMAKE_ARGS -DWITH_GFLAGS=OFF
CMAKE_ARGS -DBUILD_TESTING=OFF
)
LIST(APPEND external_project_dependencies glog)
cmake/external/gtest.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
IF(WITH_TESTING)
ENABLE_TESTING()
INCLUDE(ExternalProject)
SET(GTEST_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/gtest)
SET(GTEST_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/gtest)
SET(GTEST_INCLUDE_DIR "${GTEST_INSTALL_DIR}/include" CACHE PATH "gtest include directory." FORCE)
INCLUDE_DIRECTORIES(${GTEST_INCLUDE_DIR})
IF(WIN32)
set(GTEST_LIBRARIES
"${GTEST_INSTALL_DIR}/lib/gtest.lib" CACHE FILEPATH "gtest libraries." FORCE)
set(GTEST_MAIN_LIBRARIES
"${GTEST_INSTALL_DIR}/lib/gtest_main.lib" CACHE FILEPATH "gtest main libraries." FORCE)
ELSE(WIN32)
set(GTEST_LIBRARIES
"${GTEST_INSTALL_DIR}/lib/libgtest.a" CACHE FILEPATH "gtest libraries." FORCE)
set(GTEST_MAIN_LIBRARIES
"${GTEST_INSTALL_DIR}/lib/libgtest_main.a" CACHE FILEPATH "gtest main libraries." FORCE)
ENDIF(WIN32)
ExternalProject_Add(
gtest
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/google/googletest.git"
GIT_TAG "release-1.8.0"
PREFIX ${GTEST_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${GTEST_INSTALL_DIR}
CMAKE_ARGS -DCMAKE_POSITION_INDEPENDENT_CODE=ON
CMAKE_ARGS -DBUILD_GMOCK=ON
CMAKE_ARGS -Dgtest_disable_pthreads=ON
CMAKE_ARGS -Dgtest_force_shared_crt=ON
)
LIST(APPEND external_project_dependencies gtest)
ENDIF(WITH_TESTING)
cmake/external/openblas.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(cblas)
IF(NOT ${CBLAS_FOUND})
MESSAGE(FATAL_ERROR "Please install OpenBlas, MKL or ATLAS.")
INCLUDE(ExternalProject)
SET(CBLAS_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/openblas)
SET(CBLAS_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/openblas)
SET(CBLAS_INC_DIR "${CBLAS_INSTALL_DIR}/include" CACHE PATH "openblas include directory." FORCE)
IF(WIN32)
SET(CBLAS_LIBRARIES "${CBLAS_INSTALL_DIR}/lib/openblas.lib" CACHE FILEPATH "openblas library." FORCE)
ELSE(WIN32)
SET(CBLAS_LIBRARIES "${CBLAS_INSTALL_DIR}/lib/libopenblas.a" CACHE FILEPATH "openblas library" FORCE)
ENDIF(WIN32)
ExternalProject_Add(
openblas
${EXTERNAL_PROJECT_LOG_ARGS}
URL "https://github.com/xianyi/OpenBLAS/archive/v0.2.19.tar.gz"
PREFIX ${CBLAS_SOURCES_DIR}
INSTALL_DIR ${CBLAS_INSTALL_DIR}
BUILD_IN_SOURCE 1
CONFIGURE_COMMAND ""
BUILD_COMMAND make CC=${CMAKE_C_COMPILER} FC=${CMAKE_Fortran_COMPILER}
INSTALL_COMMAND make install PREFIX=<INSTALL_DIR>
UPDATE_COMMAND ""
)
LIST(APPEND external_project_dependencies openblas)
ENDIF()
INCLUDE_DIRECTORIES(${CBLAS_INC_DIR})
cmake/external/protobuf.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(ExternalProject)
SET(PROTOBUF_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/protobuf)
SET(PROTOBUF_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/protobuf)
SET(PROTOBUF_INCLUDE_DIR "${PROTOBUF_INSTALL_DIR}/include" CACHE PATH "protobuf include directory." FORCE)
INCLUDE_DIRECTORIES(${PROTOBUF_INCLUDE_DIR})
IF(WIN32)
SET(PROTOBUF_LITE_LIBRARY
"${PROTOBUF_INSTALL_DIR}/lib/libprotobuf-lite.lib" CACHE FILEPATH "protobuf lite library." FORCE)
SET(PROTOBUF_LIBRARY
"${PROTOBUF_INSTALL_DIR}/lib/libprotobuf.lib" CACHE FILEPATH "protobuf library." FORCE)
SET(PROTOBUF_PROTOC_LIBRARY
"${PROTOBUF_INSTALL_DIR}/lib/libprotoc.lib" CACHE FILEPATH "protoc library." FORCE)
SET(PROTOBUF_PROTOC_EXECUTABLE "${PROTOBUF_INSTALL_DIR}/bin/protoc.exe" CACHE FILEPATH "protobuf executable." FORCE)
ELSE(WIN32)
IF(${HOST_SYSTEM} STREQUAL "centos")
SET(LIB "lib64")
ELSE()
SET(LIB "lib")
ENDIF()
SET(PROTOBUF_LITE_LIBRARY
"${PROTOBUF_INSTALL_DIR}/${LIB}/libprotobuf-lite.a" CACHE FILEPATH "protobuf lite library." FORCE)
SET(PROTOBUF_LIBRARY
"${PROTOBUF_INSTALL_DIR}/${LIB}/libprotobuf.a" CACHE FILEPATH "protobuf library." FORCE)
SET(PROTOBUF_PROTOC_LIBRARY
"${PROTOBUF_INSTALL_DIR}/${LIB}/libprotoc.a" CACHE FILEPATH "protoc library." FORCE)
SET(PROTOBUF_PROTOC_EXECUTABLE "${PROTOBUF_INSTALL_DIR}/bin/protoc" CACHE FILEPATH "protobuf executable." FORCE)
ENDIF(WIN32)
ExternalProject_Add(
protobuf
${EXTERNAL_PROJECT_LOG_ARGS}
PREFIX ${PROTOBUF_SOURCES_DIR}
UPDATE_COMMAND ""
DEPENDS zlib
GIT_REPOSITORY "https://github.com/google/protobuf.git"
GIT_TAG "9f75c5aa851cd877fb0d93ccc31b8567a6706546"
CONFIGURE_COMMAND
${CMAKE_COMMAND} ${PROTOBUF_SOURCES_DIR}/src/protobuf/cmake
-Dprotobuf_BUILD_TESTS=OFF
-DCMAKE_POSITION_INDEPENDENT_CODE=ON
-DCMAKE_BUILD_TYPE=Release
-DCMAKE_INSTALL_PREFIX=${PROTOBUF_INSTALL_DIR}
)
LIST(APPEND external_project_dependencies protobuf)
cmake/external/python.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(ExternalProject)
INCLUDE(python_module)
FIND_PACKAGE(PythonInterp 2.7)
FIND_PACKAGE(PythonLibs 2.7)
SET(py_env "")
IF(PYTHONLIBS_FOUND AND PYTHONINTERP_FOUND)
find_python_module(pip REQUIRED)
find_python_module(numpy REQUIRED)
find_python_module(wheel REQUIRED)
find_python_module(google.protobuf REQUIRED)
FIND_PACKAGE(NumPy REQUIRED)
ELSE(PYTHONLIBS_FOUND AND PYTHONINTERP_FOUND)
##################################### PYTHON ########################################
SET(PYTHON_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/python)
SET(PYTHON_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/python)
SET(_python_DIR ${PYTHON_INSTALL_DIR})
IF(UNIX)
SET(PYTHON_FOUND ON)
SET(PYTHON_INCLUDE_DIR "${PYTHON_INSTALL_DIR}/include/python2.7" CACHE PATH "Python include dir" FORCE)
SET(PYTHON_LIBRARIES "${PYTHON_INSTALL_DIR}/lib/libpython2.7.a" CACHE FILEPATH "Python library" FORCE)
SET(PYTHON_EXECUTABLE ${PYTHON_INSTALL_DIR}/bin/python CACHE FILEPATH "Python executable" FORCE)
SET(PY_SITE_PACKAGES_PATH "${PYTHON_INSTALL_DIR}/lib/python2.7/site-packages" CACHE PATH "Python site-packages path" FORCE)
ELSEIF(WIN32)
SET(PYTHON_FOUND ON)
SET(PYTHON_INCLUDE_DIR "${PYTHON_INSTALL_DIR}/include" CACHE PATH "Python include dir" FORCE)
SET(PYTHON_LIBRARIES "${PYTHON_INSTALL_DIR}/libs/python27.lib" CACHE FILEPATH "Python library" FORCE)
SET(PYTHON_EXECUTABLE "${PYTHON_INSTALL_DIR}/bin/python.exe" CACHE FILEPATH "Python executable" FORCE)
SET(PY_SITE_PACKAGES_PATH "${PYTHON_INSTALL_DIR}/Lib/site-packages" CACHE PATH "Python site-packages path" FORCE)
ELSE()
MESSAGE(FATAL_ERROR "Unknown system !")
ENDIF()
IF(APPLE)
LIST(APPEND EXTERNAL_PROJECT_OPTIONAL_CMAKE_ARGS
-DCMAKE_BUILD_WITH_INSTALL_RPATH:BOOL=ON
)
ENDIF()
SET(EXTERNAL_PROJECT_OPTIONAL_CMAKE_CACHE_ARGS)
# Force Python build to "Release".
IF(CMAKE_CONFIGURATION_TYPES)
SET(SAVED_CMAKE_CFG_INTDIR ${CMAKE_CFG_INTDIR})
SET(CMAKE_CFG_INTDIR "Release")
ELSE()
LIST(APPEND EXTERNAL_PROJECT_OPTIONAL_CMAKE_CACHE_ARGS
-DCMAKE_BUILD_TYPE:STRING=Release
)
ENDIF()
ExternalProject_Add(python
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/python-cmake-buildsystem/python-cmake-buildsystem.git"
PREFIX ${PYTHON_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DPYTHON_VERSION=2.7.12
CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
CMAKE_ARGS -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
CMAKE_CACHE_ARGS
-DCMAKE_INSTALL_PREFIX:PATH=${PYTHON_INSTALL_DIR}
-DBUILD_LIBPYTHON_SHARED:BOOL=OFF
-DUSE_SYSTEM_LIBRARIES:BOOL=OFF
-DZLIB_ROOT:FILEPATH=${ZLIB_ROOT}
-DZLIB_INCLUDE_DIR:PATH=${ZLIB_INCLUDE_DIR}
-DZLIB_LIBRARY:FILEPATH=${ZLIB_LIBRARIES}
-DDOWNLOAD_SOURCES:BOOL=ON
-DINSTALL_WINDOWS_TRADITIONAL:BOOL=OFF
${EXTERNAL_PROJECT_OPTIONAL_CMAKE_CACHE_ARGS}
${EXTERNAL_PROJECT_OPTIONAL_CMAKE_ARGS}
DEPENDS zlib
)
SET(py_env
PATH=${PYTHON_INSTALL_DIR}/bin
PYTHONHOME=${PYTHON_INSTALL_DIR}
PYTHONPATH=${PYTHON_INSTALL_DIR}/lib:${PYTHON_INSTALL_DIR}/lib/python2.7:${PY_SITE_PACKAGES_PATH})
####################################################################################
##################################### SETUPTOOLS ###################################
SET(SETUPTOOLS_SOURCES_DIR ${PYTHON_SOURCES_DIR}/setuptools)
ExternalProject_Add(setuptools
${EXTERNAL_PROJECT_LOG_ARGS}
PREFIX ${SETUPTOOLS_SOURCES_DIR}
URL "https://pypi.python.org/packages/source/s/setuptools/setuptools-18.3.2.tar.gz"
BUILD_IN_SOURCE 1
PATCH_COMMAND ""
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
INSTALL_COMMAND ""
BUILD_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py install
DEPENDS python zlib
)
#####################################################################################
##################################### SIX ###########################################
SET(SIX_SOURCES_DIR ${PYTHON_SOURCES_DIR}/six)
ExternalProject_Add(six
${EXTERNAL_PROJECT_LOG_ARGS}
PREFIX ${SIX_SOURCES_DIR}
URL https://pypi.python.org/packages/source/s/six/six-1.10.0.tar.gz
BUILD_IN_SOURCE 1
PATCH_COMMAND ""
UPDATE_COMMAND ""
CONFIGURE_COMMAND ""
INSTALL_COMMAND ""
BUILD_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py install
DEPENDS python setuptools
)
#####################################################################################
##################################### CYTHON ########################################
SET(CYTHON_SOURCES_DIR ${PYTHON_SOURCES_DIR}/cython)
ExternalProject_Add(cython
${EXTERNAL_PROJECT_LOG_ARGS}
PREFIX ${CYTHON_SOURCES_DIR}
URL https://github.com/cython/cython/archive/0.25.2.tar.gz
GIT_TAG 0.25.2
BUILD_IN_SOURCE 1
CONFIGURE_COMMAND ""
PATCH_COMMAND ""
UPDATE_COMMAND ""
INSTALL_COMMAND ""
BUILD_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py install
DEPENDS python
)
####################################################################################
##################################### NUMPY ########################################
SET(NUMPY_SOURCES_DIR ${PYTHON_SOURCES_DIR}/numpy)
SET(NUMPY_TAG_VERSION "v1.11.3")
SET(NUMPY_VERSION "1.11.3")
SET(EGG_NAME "")
SET(PYTHON_NUMPY_INCLUDE_DIR "")
IF(WIN32)
SET(EGG_NAME "numpy-${NUMPY_VERSION}-py2.7-${HOST_SYSTEM}.egg")
ELSE(WIN32)
IF(APPLE)
SET(EGG_NAME "numpy-${NUMPY_VERSION}-py2.7-${HOST_SYSTEM}-${MACOS_VERSION}")
ELSE(APPLE)
SET(EGG_NAME "numpy-${NUMPY_VERSION}-py2.7-linux")
SET(EGG_NAME "numpy-${NUMPY_VERSION}-py2.7-linux")
ENDIF(APPLE)
FOREACH(suffix x86_64 intel fat64 fat32 universal)
LIST(APPEND PYTHON_NUMPY_INCLUDE_DIR ${PY_SITE_PACKAGES_PATH}/${EGG_NAME}-${suffix}.egg/numpy/core/include)
ENDFOREACH()
ENDIF(WIN32)
ExternalProject_Add(numpy
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY https://github.com/numpy/numpy.git
GIT_TAG ${NUMPY_TAG_VERSION}
CONFIGURE_COMMAND ""
UPDATE_COMMAND ""
PREFIX ${NUMPY_SOURCES_DIR}
BUILD_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py build
INSTALL_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py install
BUILD_IN_SOURCE 1
DEPENDS python setuptools cython
)
####################################################################################
##################################### WHEEL ########################################
SET(WHEEL_SOURCES_DIR ${PYTHON_SOURCES_DIR}/wheel)
ExternalProject_Add(wheel
${EXTERNAL_PROJECT_LOG_ARGS}
URL https://pypi.python.org/packages/source/w/wheel/wheel-0.29.0.tar.gz
PREFIX ${WHEEL_SOURCES_DIR}
CONFIGURE_COMMAND ""
UPDATE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py install
BUILD_IN_SOURCE 1
DEPENDS python setuptools
)
####################################################################################
################################### PROTOBUF #######################################
SET(PY_PROTOBUF_SOURCES_DIR ${PYTHON_SOURCES_DIR}/protobuf)
ExternalProject_Add(python-protobuf
${EXTERNAL_PROJECT_LOG_ARGS}
URL https://pypi.python.org/packages/e0/b0/0a1b364fe8a7d177b4b7d4dca5b798500dc57a7273b93cca73931b305a6a/protobuf-3.1.0.post1.tar.gz
URL_MD5 38b5fb160c768d2f8444d0c6d637ff91
PREFIX ${PY_PROTOBUF_SOURCES_DIR}
BUILD_IN_SOURCE 1
PATCH_COMMAND ""
CONFIGURE_COMMAND ""
BUILD_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py build
INSTALL_COMMAND env ${py_env} ${PYTHON_EXECUTABLE} setup.py install
DEPENDS python setuptools six
)
####################################################################################
LIST(APPEND external_project_dependencies python setuptools six cython wheel python-protobuf numpy)
ENDIF(PYTHONLIBS_FOUND AND PYTHONINTERP_FOUND)
INCLUDE_DIRECTORIES(${PYTHON_INCLUDE_DIR})
INCLUDE_DIRECTORIES(${PYTHON_NUMPY_INCLUDE_DIR})
MESSAGE("[Paddle] Python Executable: ${PYTHON_EXECUTABLE}")
MESSAGE("[Paddle] Python Include: ${PYTHON_INCLUDE_DIRS}")
MESSAGE("[Paddle] Python Libraries: ${PYTHON_LIBRARIES}")
cmake/external/swig.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
FIND_PACKAGE(SWIG)
IF(NOT SWIG_FOUND)
# build swig as an external project
INCLUDE(ExternalProject)
SET(SWIG_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/swig)
SET(SWIG_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/swig)
SET(SWIG_TARGET_VERSION "3.0.2")
SET(SWIG_DOWNLOAD_SRC_MD5 "62f9b0d010cef36a13a010dc530d0d41")
SET(SWIG_DOWNLOAD_WIN_MD5 "3f18de4fc09ab9abb0d3be37c11fbc8f")
IF(WIN32)
# swig.exe available as pre-built binary on Windows:
ExternalProject_Add(swig
URL http://prdownloads.sourceforge.net/swig/swigwin-${SWIG_TARGET_VERSION}.zip
URL_MD5 ${SWIG_DOWNLOAD_WIN_MD5}
SOURCE_DIR ${SWIG_SOURCES_DIR}
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
UPDATE_COMMAND ""
)
SET(SWIG_DIR ${SWIG_SOURCES_DIR} CACHE FILEPATH "SWIG Directory" FORCE)
SET(SWIG_EXECUTABLE ${SWIG_SOURCES_DIR}/swig.exe CACHE FILEPATH "SWIG Executable" FORCE)
ELSE(WIN32)
# From PCRE configure
ExternalProject_Add(pcre
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY https://github.com/svn2github/pcre.git
PREFIX ${SWIG_SOURCES_DIR}/pcre
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${SWIG_INSTALL_DIR}/pcre
)
# swig uses bison find it by cmake and pass it down
FIND_PACKAGE(BISON)
# From SWIG configure
ExternalProject_Add(swig
GIT_REPOSITORY https://github.com/swig/swig.git
GIT_TAG rel-3.0.10
PREFIX ${SWIG_SOURCES_DIR}
CONFIGURE_COMMAND cd ${SWIG_SOURCES_DIR}/src/swig && ./autogen.sh
CONFIGURE_COMMAND cd ${SWIG_SOURCES_DIR}/src/swig &&
env "PCRE_LIBS=${SWIG_INSTALL_DIR}/pcre/lib/libpcre.a ${SWIG_INSTALL_DIR}/pcre/lib/libpcrecpp.a ${SWIG_INSTALL_DIR}/pcre/lib/libpcreposix.a"
./configure
--prefix=${SWIG_INSTALL_DIR}
--with-pcre-prefix=${SWIG_INSTALL_DIR}/pcre
BUILD_COMMAND cd ${SWIG_SOURCES_DIR}/src/swig && make
INSTALL_COMMAND cd ${SWIG_SOURCES_DIR}/src/swig && make install
UPDATE_COMMAND ""
DEPENDS pcre
)
SET(SWIG_DIR ${SWIG_INSTALL_DIR}/share/swig/${SWIG_TARGET_VERSION})
SET(SWIG_EXECUTABLE ${SWIG_INSTALL_DIR}/bin/swig)
ENDIF(WIN32)
LIST(APPEND external_project_dependencies swig)
ENDIF(NOT SWIG_FOUND)
cmake/external/warpctc.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(ExternalProject)
SET(WARPCTC_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/warpctc)
SET(WARPCTC_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/warpctc)
SET(WARPCTC_INCLUDE_DIR "${WARPCTC_INSTALL_DIR}/include" CACHE PATH "Warp-ctc Directory" FORCE)
INCLUDE_DIRECTORIES(${WARPCTC_INCLUDE_DIR})
SET(WARPCTC_LIB_DIR "${WARPCTC_INSTALL_DIR}/lib" CACHE PATH "Warp-ctc Library Directory" FORCE)
IF(WIN32)
SET(WARPCTC_LIBRARIES
"${WARPCTC_INSTALL_DIR}/lib/warpctc.dll" CACHE FILEPATH "Warp-ctc Library" FORCE)
ELSE(WIN32)
IF(APPLE)
SET(_warpctc_SHARED_SUFFIX dylib)
ELSE(APPLE)
SET(_warpctc_SHARED_SUFFIX so)
ENDIF(APPLE)
SET(WARPCTC_LIBRARIES
"${WARPCTC_INSTALL_DIR}/lib/libwarpctc.${_warpctc_SHARED_SUFFIX}" CACHE FILEPATH "Warp-ctc Library" FORCE)
ENDIF(WIN32)
IF(CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang" )
SET(USE_OMP OFF)
ELSE()
SET(USE_OMP ON)
ENDIF()
ExternalProject_Add(
warpctc
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/gangliao/warp-ctc.git"
PREFIX ${WARPCTC_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_CXX_COMPILER=${CMAKE_CXX_COMPILER}
CMAKE_ARGS -DCMAKE_C_COMPILER=${CMAKE_C_COMPILER}
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${WARPCTC_INSTALL_DIR}
CMAKE_ARGS -DWITH_GPU=${WITH_GPU}
CMAKE_ARGS -DWITH_OMP=${USE_OMP}
CMAKE_ARGS -DWITH_TORCH=OFF
CMAKE_ARGS -DBUILD_SHARED=ON
)
LIST(APPEND external_project_dependencies warpctc)
cmake/external/zlib.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
INCLUDE(ExternalProject)
SET(ZLIB_SOURCES_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/zlib)
SET(ZLIB_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third_party/install/zlib)
SET(ZLIB_ROOT ${ZLIB_INSTALL_DIR} CACHE FILEPATH "zlib root directory." FORCE)
SET(ZLIB_INCLUDE_DIR "${ZLIB_INSTALL_DIR}/include" CACHE PATH "zlib include directory." FORCE)
IF(WIN32)
SET(ZLIB_LIBRARIES "${ZLIB_INSTALL_DIR}/lib/zlibstatic.lib" CACHE FILEPATH "zlib library." FORCE)
ELSE(WIN32)
set(ZLIB_LIBRARIES "${ZLIB_INSTALL_DIR}/lib/libz.a" CACHE FILEPATH "zlib library." FORCE)
ENDIF(WIN32)
INCLUDE_DIRECTORIES(${ZLIB_INCLUDE_DIR})
ExternalProject_Add(
zlib
${EXTERNAL_PROJECT_LOG_ARGS}
GIT_REPOSITORY "https://github.com/madler/zlib.git"
GIT_TAG "v1.2.8"
PREFIX ${ZLIB_SOURCES_DIR}
UPDATE_COMMAND ""
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${ZLIB_INSTALL_DIR}
CMAKE_ARGS -DBUILD_SHARED_LIBS=OFF
CMAKE_ARGS -DCMAKE_POSITION_INDEPENDENT_CODE=ON
CMAKE_ARGS -DCMAKE_MACOSX_RPATH=ON
)
LIST(APPEND external_project_dependencies zlib)
cmake/rdma.cmake
浏览文件 @ b23d99dd
......@@ -5,72 +5,76 @@
# svn co https://svn.baidu.com/sys/ip/trunk/rdma/thirdparty rdma/
# we use static output in svn repositories to avoid implict bugs from not standard runtime env.
set(RDMA_ROOT $ENV{RDMA_ROOT} CACHE PATH "Folder contains RDMA sock library and thirdparty library")
if(WITH_RDMA)
set(RDMA_ROOT $ENV{RDMA_ROOT} CACHE PATH "Folder contains RDMA sock library and thirdparty library")
function(generate_rdma_links)
#redirect to current DIR to isolate the pollution from system runtime environment
#it can benifits unified control for different gcc environment.
#e.g, by default gcc48 did not refer /usr/lib64 which could contain low version
#runtime libraries that will crash process while loading it. That redirect trick
#can fix it.
execute_process(
COMMAND mkdir -p librdma
COMMAND ln -s -f /usr/lib64/libibverbs.so.1.0.0 librdma/libibverbs.so.1
COMMAND ln -s -f /usr/lib64/libibverbs.so.1.0.0 librdma/libibverbs.so
COMMAND ln -s -f /usr/lib64/librdmacm.so.1.0.0 librdma/librdmacm.so.1
COMMAND ln -s -f /usr/lib64/librdmacm.so.1.0.0 librdma/librdmacm.so
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
)
endfunction(generate_rdma_links)
#check and set headers
find_path(RDMA_INC_SXISOCK sxi_sock.h PATHS ${RDMA_ROOT}/sockrdmav1/output/include)
find_path(RDMA_INC_XIO libxio.h PATHS ${RDMA_ROOT}/thirdparty/output/accelio)
find_path(RDMA_INC_EVENT event2 PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_path(RDMA_INC_NUMA numa.h PATHS ${RDMA_ROOT}/thirdparty/output/libnuma)
#check and set libs
find_library(RDMA_LIB_SXISOCK NAMES sxisock PATHS ${RDMA_ROOT}/sockrdmav1/output)
find_library(RDMA_LIB_XIO NAMES xio PATHS ${RDMA_ROOT}/thirdparty/output/accelio)
find_library(RDMA_LIB_EVENT NAMES event PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_EVENT_CORE NAMES event_core PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_EVENT_EXTRA NAMES event_extra PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_EVENT_PTHREADS NAMES event_pthreads PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_NUMA NAMES numa PATHS ${RDMA_ROOT}/thirdparty/output/libnuma)
if(
RDMA_INC_SXISOCK AND
RDMA_INC_XIO AND
RDMA_INC_EVENT AND
RDMA_INC_NUMA AND
RDMA_LIB_SXISOCK AND
RDMA_LIB_XIO AND
RDMA_LIB_EVENT AND
RDMA_LIB_EVENT_CORE AND
RDMA_LIB_EVENT_EXTRA AND
RDMA_LIB_EVENT_PTHREADS AND
RDMA_LIB_NUMA
function(generate_rdma_links)
#redirect to current DIR to isolate the pollution from system runtime environment
#it can benifits unified control for different gcc environment.
#e.g, by default gcc48 did not refer /usr/lib64 which could contain low version
#runtime libraries that will crash process while loading it. That redirect trick
#can fix it.
execute_process(
COMMAND mkdir -p librdma
COMMAND ln -s -f /usr/lib64/libibverbs.so.1.0.0 librdma/libibverbs.so.1
COMMAND ln -s -f /usr/lib64/libibverbs.so.1.0.0 librdma/libibverbs.so
COMMAND ln -s -f /usr/lib64/librdmacm.so.1.0.0 librdma/librdmacm.so.1
COMMAND ln -s -f /usr/lib64/librdmacm.so.1.0.0 librdma/librdmacm.so
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
)
endfunction(generate_rdma_links)
set(RDMA_INC_DIR
${RDMA_INC_SXISOCK}
${RDMA_INC_XIO}
${RDMA_INC_EVENT}
${RDMA_INC_NUMA})
set(RDMA_LIBS
${RDMA_LIB_SXISOCK}
${RDMA_LIB_XIO}
${RDMA_LIB_EVENT}
${RDMA_LIB_EVENT_CORE}
${RDMA_LIB_EVENT_EXTRA}
${RDMA_LIB_EVENT_PTHREADS}
${RDMA_LIB_NUMA}
)
set(RDMA_LD_FLAGS "-L./librdma -libverbs -lrdmacm -Xlinker -rpath ./librdma")
return()
endif()
#check and set headers
find_path(RDMA_INC_SXISOCK sxi_sock.h PATHS ${RDMA_ROOT}/sockrdmav1/output/include)
find_path(RDMA_INC_XIO libxio.h PATHS ${RDMA_ROOT}/thirdparty/output/accelio)
find_path(RDMA_INC_EVENT event2 PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_path(RDMA_INC_NUMA numa.h PATHS ${RDMA_ROOT}/thirdparty/output/libnuma)
#check and set libs
find_library(RDMA_LIB_SXISOCK NAMES sxisock PATHS ${RDMA_ROOT}/sockrdmav1/output)
find_library(RDMA_LIB_XIO NAMES xio PATHS ${RDMA_ROOT}/thirdparty/output/accelio)
find_library(RDMA_LIB_EVENT NAMES event PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_EVENT_CORE NAMES event_core PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_EVENT_EXTRA NAMES event_extra PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_EVENT_PTHREADS NAMES event_pthreads PATHS ${RDMA_ROOT}/thirdparty/output/libevent)
find_library(RDMA_LIB_NUMA NAMES numa PATHS ${RDMA_ROOT}/thirdparty/output/libnuma)
#if this module is not called, RDMA_INC_DIR RDMA_LIBS will be null, so top module always refer this variable
if(
RDMA_INC_SXISOCK AND
RDMA_INC_XIO AND
RDMA_INC_EVENT AND
RDMA_INC_NUMA AND
RDMA_LIB_SXISOCK AND
RDMA_LIB_XIO AND
RDMA_LIB_EVENT AND
RDMA_LIB_EVENT_CORE AND
RDMA_LIB_EVENT_EXTRA AND
RDMA_LIB_EVENT_PTHREADS AND
RDMA_LIB_NUMA
)
message(FATAL_ERROR, "RDMA libraries are not found, try to set RDMA_ROOT or check all related libraries.")
set(RDMA_INC_DIR
${RDMA_INC_SXISOCK}
${RDMA_INC_XIO}
${RDMA_INC_EVENT}
${RDMA_INC_NUMA})
set(RDMA_LIBS
${RDMA_LIB_SXISOCK}
${RDMA_LIB_XIO}
${RDMA_LIB_EVENT}
${RDMA_LIB_EVENT_CORE}
${RDMA_LIB_EVENT_EXTRA}
${RDMA_LIB_EVENT_PTHREADS}
${RDMA_LIB_NUMA}
)
set(RDMA_LD_FLAGS "-L./librdma -libverbs -lrdmacm -Xlinker -rpath ./librdma")
include_directories("${RDMA_INC_DIR}")
else()
#if this module is not called, RDMA_INC_DIR RDMA_LIBS will be null, so top module always refer this variable
message(FATAL_ERROR, "RDMA libraries are not found, try to set RDMA_ROOT or check all related libraries.")
endif()
else(WITH_RDMA)
set(RDMA_LIBS "")
set(RDMA_LD_FLAGS "")
add_definitions(-DPADDLE_DISABLE_RDMA)
endif(WITH_RDMA)
cmake/swig.cmake 已删除 100644 → 0
浏览文件 @ 343d9973
function(generate_python_api target_name)
add_custom_command(OUTPUT ${PROJ_ROOT}/paddle/py_paddle/swig_paddle.py
${PROJ_ROOT}/paddle/Paddle_wrap.cxx
${PROJ_ROOT}/paddle/Paddle_wrap.h
COMMAND swig -python -c++ -outcurrentdir -I../ api/Paddle.swig
&& mv ${PROJ_ROOT}/paddle/swig_paddle.py ${PROJ_ROOT}/paddle/py_paddle/swig_paddle.py
DEPENDS ${PROJ_ROOT}/paddle/api/Paddle.swig
${PROJ_ROOT}/paddle/api/PaddleAPI.h
WORKING_DIRECTORY ${PROJ_ROOT}/paddle
COMMENT "Generate Python API from swig")
add_custom_target(${target_name} ALL DEPENDS
${PROJ_ROOT}/paddle/Paddle_wrap.cxx
${PROJ_ROOT}/paddle/Paddle_wrap.h
${PROJ_ROOT}/paddle/py_paddle/swig_paddle.py)
endfunction(generate_python_api)
cmake/system.cmake 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
IF(WIN32)
SET(HOST_SYSTEM "win32")
ELSE(WIN32)
IF(APPLE)
EXEC_PROGRAM (sw_vers ARGS -productVersion OUTPUT_VARIABLE MACOSX_VERSION)
STRING(REGEX MATCH "[0-9]+.[0-9]+" VERSION "${MACOSX_VERSION}")
SET(MACOS_VERSION ${VERSION})
SET(HOST_SYSTEM "macosx")
ELSE(APPLE)
IF(EXISTS "/etc/issue")
FILE(READ "/etc/issue" LINUX_ISSUE)
IF(LINUX_ISSUE MATCHES "CentOS")
SET(HOST_SYSTEM "centos")
ELSEIF(LINUX_ISSUE MATCHES "Debian")
SET(HOST_SYSTEM "debian")
ELSEIF(LINUX_ISSUE MATCHES "Ubuntu")
SET(HOST_SYSTEM "ubuntu")
ENDIF()
ENDIF(EXISTS "/etc/issue")
ENDIF(APPLE)
ENDIF(WIN32)
# query number of logical cores
CMAKE_HOST_SYSTEM_INFORMATION(RESULT CPU_CORES QUERY NUMBER_OF_LOGICAL_CORES)
MARK_AS_ADVANCED(HOST_SYSTEM CPU_CORES)
MESSAGE(STATUS "Found Paddle host system: ${HOST_SYSTEM}")
MESSAGE(STATUS "Found Paddle host system's CPU: ${CPU_CORES} cores")
# external dependencies log output
SET(EXTERNAL_PROJECT_LOG_ARGS
LOG_DOWNLOAD 0 # Wrap download in script to log output
LOG_UPDATE 1 # Wrap update in script to log output
LOG_CONFIGURE 1 # Wrap configure in script to log output
LOG_BUILD 1 # Wrap build in script to log output
LOG_TEST 1 # Wrap test in script to log output
LOG_INSTALL 1 # Wrap install in script to log output
)
cmake/util.cmake
浏览文件 @ b23d99dd
......@@ -24,7 +24,7 @@ function(target_circle_link_libraries TARGET_NAME)
list(APPEND libsInArgn ${arg})
endif()
endforeach()
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang" OR "${CMAKE_CXX_COMPILER_ID}" STREQUAL "AppleClang")
list(APPEND LIBS "-undefined dynamic_lookup")
endif()
list(REVERSE libsInArgn)
......@@ -81,18 +81,6 @@ function(link_paddle_exe TARGET_NAME)
set(METRIC_LIBS "")
endif()
if(PADDLE_WITH_INTERNAL)
set(INTERAL_LIBS paddle_internal_gserver paddle_internal_parameter)
target_circle_link_libraries(${TARGET_NAME}
ARCHIVE_START
paddle_internal_gserver
paddle_internal_owlqn
ARCHIVE_END
paddle_internal_parameter)
else()
set(INTERAL_LIBS "")
endif()
target_circle_link_libraries(${TARGET_NAME}
ARCHIVE_START
paddle_gserver
......@@ -107,26 +95,16 @@ function(link_paddle_exe TARGET_NAME)
paddle_parameter
paddle_proto
paddle_cuda
paddle_test_main
${METRIC_LIBS}
${PROTOBUF_LIBRARY}
${LIBGLOG_LIBRARY}
${GFLAGS_LIBRARIES}
${EXTERNAL_LIBS}
${CMAKE_THREAD_LIBS_INIT}
${CBLAS_LIBS}
${ZLIB_LIBRARIES}
${INTERAL_LIBS}
${CMAKE_DL_LIBS})
if(WITH_RDMA)
target_link_libraries(${TARGET_NAME}
${RDMA_LD_FLAGS}
${RDMA_LIBS})
endif()
${CMAKE_DL_LIBS}
${RDMA_LD_FLAGS}
${RDMA_LIBS})
if(WITH_PYTHON)
target_link_libraries(${TARGET_NAME}
${PYTHON_LIBRARIES})
${PYTHON_LIBRARIES} util)
endif()
if(WITH_GPU)
......@@ -143,10 +121,7 @@ function(link_paddle_exe TARGET_NAME)
endif()
endif()
if(NOT WITH_DSO)
target_link_libraries(${TARGET_NAME}
${WARPCTC_LIBRARY})
endif()
add_dependencies(${TARGET_NAME} ${external_project_dependencies})
endfunction()
# link_paddle_test
......@@ -155,8 +130,10 @@ endfunction()
# Rest Arguemnts: not used.
function(link_paddle_test TARGET_NAME)
link_paddle_exe(${TARGET_NAME})
target_link_libraries(${TARGET_NAME} ${GTEST_MAIN_LIBRARIES}
${GTEST_LIBRARIES})
target_link_libraries(${TARGET_NAME}
paddle_test_main
paddle_test_util
${GTEST_LIBRARIES})
endfunction()
# add_unittest_without_exec
......
cmake/version.cmake
浏览文件 @ b23d99dd
......@@ -21,4 +21,5 @@ while ("${PADDLE_VERSION}" STREQUAL "")
endif()
endwhile()
add_definitions(-DPADDLE_VERSION=${PADDLE_VERSION})
message(STATUS "Paddle version is ${PADDLE_VERSION}")
demo/mnist/.gitignore
浏览文件 @ b23d99dd
......@@ -4,3 +4,4 @@ mnist_vgg_model
plot.png
train.log
*pyc
.ipynb_checkpoints
demo/mnist/api_train.py 0 → 100644
浏览文件 @ b23d99dd
"""
A very basic example for how to use current Raw SWIG API to train mnist network.
Current implementation uses Raw SWIG, which means the API call is directly \
passed to C++ side of Paddle.
The user api could be simpler and carefully designed.
"""
import py_paddle.swig_paddle as api
from py_paddle import DataProviderConverter
import paddle.trainer.PyDataProvider2 as dp
import numpy as np
import random
from mnist_util import read_from_mnist
from paddle.trainer_config_helpers import *
def optimizer_config():
settings(
learning_rate=1e-4,
learning_method=AdamOptimizer(),
batch_size=1000,
model_average=ModelAverage(average_window=0.5),
regularization=L2Regularization(rate=0.5))
def network_config():
imgs = data_layer(name='pixel', size=784)
hidden1 = fc_layer(input=imgs, size=200)
hidden2 = fc_layer(input=hidden1, size=200)
inference = fc_layer(input=hidden2, size=10, act=SoftmaxActivation())
cost = classification_cost(
input=inference, label=data_layer(
name='label', size=10))
outputs(cost)
def init_parameter(network):
assert isinstance(network, api.GradientMachine)
for each_param in network.getParameters():
assert isinstance(each_param, api.Parameter)
array_size = len(each_param)
array = np.random.uniform(-1.0, 1.0, array_size).astype('float32')
each_param.getBuf(api.PARAMETER_VALUE).copyFromNumpyArray(array)
def generator_to_batch(generator, batch_size):
ret_val = list()
for each_item in generator:
ret_val.append(each_item)
if len(ret_val) == batch_size:
yield ret_val
ret_val = list()
if len(ret_val) != 0:
yield ret_val
class BatchPool(object):
def __init__(self, generator, batch_size):
self.data = list(generator)
self.batch_size = batch_size
def __call__(self):
random.shuffle(self.data)
for offset in xrange(0, len(self.data), self.batch_size):
limit = min(offset + self.batch_size, len(self.data))
yield self.data[offset:limit]
def input_order_converter(generator):
for each_item in generator:
yield each_item['pixel'], each_item['label']
def main():
api.initPaddle("-use_gpu=false", "-trainer_count=4") # use 4 cpu cores
# get enable_types for each optimizer.
# enable_types = [value, gradient, momentum, etc]
# For each optimizer(SGD, Adam), GradientMachine should enable different
# buffers.
opt_config_proto = parse_optimizer_config(optimizer_config)
opt_config = api.OptimizationConfig.createFromProto(opt_config_proto)
_temp_optimizer_ = api.ParameterOptimizer.create(opt_config)
enable_types = _temp_optimizer_.getParameterTypes()
# Create Simple Gradient Machine.
model_config = parse_network_config(network_config)
m = api.GradientMachine.createFromConfigProto(
model_config, api.CREATE_MODE_NORMAL, enable_types)
# This type check is not useful. Only enable type hint in IDE.
# Such as PyCharm
assert isinstance(m, api.GradientMachine)
# Initialize Parameter by numpy.
init_parameter(network=m)
# Create Local Updater. Local means not run in cluster.
# For a cluster training, here we can change to createRemoteUpdater
# in future.
updater = api.ParameterUpdater.createLocalUpdater(opt_config)
assert isinstance(updater, api.ParameterUpdater)
# Initialize ParameterUpdater.
updater.init(m)
# DataProvider Converter is a utility convert Python Object to Paddle C++
# Input. The input format is as same as Paddle's DataProvider.
converter = DataProviderConverter(
input_types=[dp.dense_vector(784), dp.integer_value(10)])
train_file = './data/raw_data/train'
test_file = './data/raw_data/t10k'
# start gradient machine.
# the gradient machine must be started before invoke forward/backward.
# not just for training, but also for inference.
m.start()
# evaluator can print error rate, etc. It is a C++ class.
batch_evaluator = m.makeEvaluator()
test_evaluator = m.makeEvaluator()
# Get Train Data.
# TrainData will stored in a data pool. Currently implementation is not care
# about memory, speed. Just a very naive implementation.
train_data_generator = input_order_converter(read_from_mnist(train_file))
train_data = BatchPool(train_data_generator, 512)
# outArgs is Neural Network forward result. Here is not useful, just passed
# to gradient_machine.forward
outArgs = api.Arguments.createArguments(0)
for pass_id in xrange(2): # we train 2 passes.
updater.startPass()
for batch_id, data_batch in enumerate(train_data()):
# data_batch is input images.
# here, for online learning, we could get data_batch from network.
# Start update one batch.
pass_type = updater.startBatch(len(data_batch))
# Start BatchEvaluator.
# batch_evaluator can be used between start/finish.
batch_evaluator.start()
# forwardBackward is a shortcut for forward and backward.
# It is sometimes faster than invoke forward/backward separately,
# because in GradientMachine, it may be async.
m.forwardBackward(converter(data_batch), outArgs, pass_type)
for each_param in m.getParameters():
updater.update(each_param)
# Get cost. We use numpy to calculate total cost for this batch.
cost_vec = outArgs.getSlotValue(0)
cost_vec = cost_vec.copyToNumpyMat()
cost = cost_vec.sum() / len(data_batch)
# Make evaluator works.
m.eval(batch_evaluator)
# Print logs.
print 'Pass id', pass_id, 'Batch id', batch_id, 'with cost=', \
cost, batch_evaluator
batch_evaluator.finish()
# Finish batch.
# * will clear gradient.
# * ensure all values should be updated.
updater.finishBatch(cost)
# testing stage. use test data set to test current network.
updater.apply()
test_evaluator.start()
test_data_generator = input_order_converter(read_from_mnist(test_file))
for data_batch in generator_to_batch(test_data_generator, 512):
# in testing stage, only forward is needed.
m.forward(converter(data_batch), outArgs, api.PASS_TEST)
m.eval(test_evaluator)
# print error rate for test data set
print 'Pass', pass_id, ' test evaluator: ', test_evaluator
test_evaluator.finish()
updater.restore()
updater.catchUpWith()
params = m.getParameters()
for each_param in params:
assert isinstance(each_param, api.Parameter)
value = each_param.getBuf(api.PARAMETER_VALUE)
value = value.copyToNumpyArray()
# Here, we could save parameter to every where you want
print each_param.getName(), value
updater.finishPass()
m.finish()
if __name__ == '__main__':
main()
demo/mnist/mnist_provider.py
浏览文件 @ b23d99dd
from paddle.trainer.PyDataProvider2 import *
import numpy
from mnist_util import read_from_mnist
# Define a py data provider
......@@ -8,27 +8,5 @@ import numpy
'label': integer_value(10)},
cache=CacheType.CACHE_PASS_IN_MEM)
def process(settings, filename): # settings is not used currently.
imgf = filename + "-images-idx3-ubyte"
labelf = filename + "-labels-idx1-ubyte"
f = open(imgf, "rb")
l = open(labelf, "rb")
f.read(16)
l.read(8)
# Define number of samples for train/test
if "train" in filename:
n = 60000
else:
n = 10000
images = numpy.fromfile(
f, 'ubyte', count=n * 28 * 28).reshape((n, 28 * 28)).astype('float32')
images = images / 255.0 * 2.0 - 1.0
labels = numpy.fromfile(l, 'ubyte', count=n).astype("int")
for i in xrange(n):
yield {"pixel": images[i, :], 'label': labels[i]}
f.close()
l.close()
for each in read_from_mnist(filename):
yield each
demo/mnist/mnist_util.py 0 → 100644
浏览文件 @ b23d99dd
import numpy
__all__ = ['read_from_mnist']
def read_from_mnist(filename):
imgf = filename + "-images-idx3-ubyte"
labelf = filename + "-labels-idx1-ubyte"
f = open(imgf, "rb")
l = open(labelf, "rb")
f.read(16)
l.read(8)
# Define number of samples for train/test
if "train" in filename:
n = 60000
else:
n = 10000
images = numpy.fromfile(
f, 'ubyte', count=n * 28 * 28).reshape((n, 28 * 28)).astype('float32')
images = images / 255.0 * 2.0 - 1.0
labels = numpy.fromfile(l, 'ubyte', count=n).astype("int")
for i in xrange(n):
yield {"pixel": images[i, :], 'label': labels[i]}
f.close()
l.close()
demo/model_zoo/embedding/pre_DictAndModel.sh
浏览文件 @ b23d99dd
......@@ -14,9 +14,19 @@
# limitations under the License.
set -e
set -x
BASE_URL='http://paddlepaddle.cdn.bcebos.com/model_zoo/embedding'
# download the dictionary and pretrained model
for file in baidu.dict model_32.emb model_64.emb model_128.emb model_256.emb
do
wget http://paddlepaddle.bj.bcebos.com/model_zoo/embedding/$file
DOWNLOAD_ITEMS=(baidu.dict model_32.emb model_64.emb model_128.emb model_256.emb)
ITEM_MD5=(fa03a12321eaab6c30a8fcc9442eaea3
f88c8325ee6da6187f1080e8fe66c1cd
927cf70f27f860aff1a5703ebf7f1584
a52e43655cd25d279777ed509a1ae27b
b92c67fe9ff70fea53596080e351ac80)
for ((i=0; i<${#ITEM_MD5[@]}; i++))
do
FILENAME=${DOWNLOAD_ITEMS[${i}]}
REAL_MD5=`wget ${BASE_URL}/${FILENAME} -O - | tee ${FILENAME} | md5sum | cut -d ' ' -f 1`
EXPECTED_MD5=${ITEM_MD5[${i}]}
[ "${EXPECTED_MD5}" = "${REAL_MD5}" ]
done
demo/quick_start/api_predict.sh
浏览文件 @ b23d99dd
......@@ -17,7 +17,7 @@ set -e
#Note the default model is pass-00002, you shold make sure the model path
#exists or change the mode path.
#only test on trainer_config.lr.py
model=output/pass-00001/
model=output/model/pass-00001/
config=trainer_config.lr.py
label=data/labels.list
dict=data/dict.txt
......
demo/quick_start/cluster/cluster_train.sh 0 → 100755
浏览文件 @ b23d99dd
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
# Should run pserver.sh before run this script.
bin_dir=$(cd `dirname $0`; pwd)
home_dir=$(cd "${bin_dir}/.."; pwd)
source "$bin_dir/env.sh"
model_dir="$bin_dir/output"
log_file="$bin_dir/train.log"
pushd "$home_dir"
cfg=trainer_config.lr.py
paddle train \
--config=$cfg \
--save_dir=${model_dir} \
--trainer_count=4 \
--local=0 \
--log_period=100 \
--num_passes=15 \
--use_gpu=false \
--show_parameter_stats_period=100 \
--test_all_data_in_one_period=1 \
--num_gradient_servers=1 \
--nics=`get_nics` \
--port=7164 \
--ports_num=1 \
--pservers="127.0.0.1" \
--comment="paddle_trainer" \
2>&1 | tee "$log_file"
popd
demo/quick_start/cluster/env.sh 0 → 100644
浏览文件 @ b23d99dd
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
function get_nics() {
machine=`uname -s`
local nics=""
if [ "$machine" == "Linux" ]; then
nics="lo"
elif [ "$machine" == "Darwin" ]; then
nics="lo0"
else
nics="unsupport"
fi
echo $nics
}
demo/quick_start/cluster/pserver.sh 0 → 100755
浏览文件 @ b23d99dd
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
bin_dir=$(cd `dirname $0`; pwd)
source "$bin_dir/env.sh"
paddle pserver \
--nics=`get_nics` \
--port=7164 \
--ports_num=1 \
--ports_num_for_sparse=1 \
--num_gradient_servers=1 \
--comment="paddle_pserver" \
2>&1 | tee 'pserver.log'
demo/recommendation/evaluate.py 0 → 100755
浏览文件 @ b23d99dd
#!/usr/bin/python
# Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import sys
import re
import math
def get_best_pass(log_filename):
with open(log_filename, 'r') as f:
text = f.read()
pattern = re.compile('Test.*? cost=([0-9]+\.[0-9]+).*?pass-([0-9]+)',
re.S)
results = re.findall(pattern, text)
sorted_results = sorted(results, key=lambda result: float(result[0]))
return sorted_results[0]
log_filename = sys.argv[1]
log = get_best_pass(log_filename)
predict_error = math.sqrt(float(log[0])) / 2
print 'Best pass is %s, error is %s, which means predict get error as %f' % (
log[1], log[0], predict_error)
evaluate_pass = "output/pass-%s" % log[1]
print "evaluating from pass %s" % evaluate_pass
demo/traffic_prediction/README 0 → 100644
浏览文件 @ b23d99dd
run by:
cd ./data
sh get_data.sh
cd ..
sh train.sh
sh predict.sh
demo/traffic_prediction/data/get_data.sh 0 → 100755
浏览文件 @ b23d99dd
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
set -x
DIR="$( cd "$(dirname "$0")" ; pwd -P )"
cd $DIR
#download the dataset
echo "Downloading traffic data..."
wget http://paddlepaddle.cdn.bcebos.com/demo/traffic/traffic_data.tar.gz
#extract package
echo "Unzipping..."
tar -zxvf traffic_data.tar.gz
echo "data/speeds.csv" > train.list
echo "data/speeds.csv" > test.list
echo "data/speeds.csv" > pred.list
echo "Done."
demo/traffic_prediction/dataprovider.py 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from paddle.trainer.PyDataProvider2 import *
import sys
import numpy as np
TERM_NUM = 24
FORECASTING_NUM = 24
LABEL_VALUE_NUM = 4
def initHook(settings, file_list, **kwargs):
"""
Init hook is invoked before process data. It will set obj.slots and store data meta.
:param settings: global object. It will passed to process routine.
:type obj: object
:param file_list: the meta file object, which passed from trainer_config.py,but unused in this function.
:param kwargs: unused other arguments.
"""
del kwargs #unused
settings.pool_size = sys.maxint
#Use a time seires of the past as feature.
#Dense_vector's expression form is [float,float,...,float]
settings.input_types = [dense_vector(TERM_NUM)]
#There are next FORECASTING_NUM fragments you need predict.
#Every predicted condition at time point has four states.
for i in range(FORECASTING_NUM):
settings.input_types.append(integer_value(LABEL_VALUE_NUM))
@provider(
init_hook=initHook, cache=CacheType.CACHE_PASS_IN_MEM, should_shuffle=True)
def process(settings, file_name):
with open(file_name) as f:
#abandon fields name
f.next()
for row_num, line in enumerate(f):
speeds = map(int, line.rstrip('\r\n').split(",")[1:])
# Get the max index.
end_time = len(speeds)
# Scanning and generating samples
for i in range(TERM_NUM, end_time - FORECASTING_NUM):
# For dense slot
pre_spd = map(float, speeds[i - TERM_NUM:i])
# Integer value need predicting, values start from 0, so every one minus 1.
fol_spd = [j - 1 for j in speeds[i:i + FORECASTING_NUM]]
# Predicting label is missing, abandon the sample.
if -1 in fol_spd:
continue
yield [pre_spd] + fol_spd
def predict_initHook(settings, file_list, **kwargs):
settings.pool_size = sys.maxint
settings.input_types = [dense_vector(TERM_NUM)]
@provider(init_hook=predict_initHook, should_shuffle=False)
def process_predict(settings, file_name):
with open(file_name) as f:
#abandon fields name
f.next()
for row_num, line in enumerate(f):
speeds = map(int, line.rstrip('\r\n').split(","))
end_time = len(speeds)
pre_spd = map(float, speeds[end_time - TERM_NUM:end_time])
yield pre_spd
demo/traffic_prediction/gen_result.py 0 → 100644
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
res = []
with open('./rank-00000') as f:
for line in f:
pred = map(int, line.strip('\r\n;').split(";"))
#raw prediction range from 0 to 3
res.append([i + 1 for i in pred])
file_name = open('./data/pred.list').read().strip('\r\n')
FORECASTING_NUM = 24
header = [
'id',
'201604200805',
'201604200810',
'201604200815',
'201604200820',
'201604200825',
'201604200830',
'201604200835',
'201604200840',
'201604200845',
'201604200850',
'201604200855',
'201604200900',
'201604200905',
'201604200910',
'201604200915',
'201604200920',
'201604200925',
'201604200930',
'201604200935',
'201604200940',
'201604200945',
'201604200950',
'201604200955',
'201604201000',
]
###################
## To CSV format ##
###################
with open(file_name) as f:
f.next()
print ','.join(header)
for row_num, line in enumerate(f):
fields = line.rstrip('\r\n').split(',')
linkid = fields[0]
print linkid + ',' + ','.join(map(str, res[row_num]))
demo/traffic_prediction/predict.sh 0 → 100755
浏览文件 @ b23d99dd
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
cfg=trainer_config.py
# pass choice
model="output/pass-00000"
paddle train \
--config=$cfg \
--use_gpu=false \
--job=test \
--init_model_path=$model \
--config_args=is_predict=1 \
--predict_output_dir=.
python gen_result.py > result.txt
rm -rf rank-00000
demo/traffic_prediction/train.sh 0 → 100755
浏览文件 @ b23d99dd
#!/bin/bash
# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
set -e
cfg=trainer_config.py
paddle train \
--config=$cfg \
--save_dir=./output \
--trainer_count=4 \
--log_period=1000 \
--dot_period=10 \
--num_passes=10 \
--use_gpu=false \
--show_parameter_stats_period=3000 \
2>&1 | tee 'train.log'
demo/traffic_prediction/trainer_config.py 0 → 100755
浏览文件 @ b23d99dd
# Copyright (c) 2016 PaddlePaddle Authors, Inc. All Rights Reserved
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from paddle.trainer_config_helpers import *
################################### DATA Configuration #############################################
is_predict = get_config_arg('is_predict', bool, False)
trn = './data/train.list' if not is_predict else None
tst = './data/test.list' if not is_predict else './data/pred.list'
process = 'process' if not is_predict else 'process_predict'
define_py_data_sources2(
train_list=trn, test_list=tst, module="dataprovider", obj=process)
################################### Parameter Configuaration #######################################
TERM_NUM = 24
FORECASTING_NUM = 24
emb_size = 16
batch_size = 128 if not is_predict else 1
settings(
batch_size=batch_size,
learning_rate=1e-3,
learning_method=RMSPropOptimizer())
################################### Algorithm Configuration ########################################
output_label = []
link_encode = data_layer(name='link_encode', size=TERM_NUM)
for i in xrange(FORECASTING_NUM):
# Each task share same weight.
link_param = ParamAttr(
name='_link_vec.w', initial_max=1.0, initial_min=-1.0)
link_vec = fc_layer(input=link_encode, size=emb_size, param_attr=link_param)
score = fc_layer(input=link_vec, size=4, act=SoftmaxActivation())
if is_predict:
maxid = maxid_layer(score)
output_label.append(maxid)
else:
# Multi-task training.
label = data_layer(name='label_%dmin' % ((i + 1) * 5), size=4)
cls = classification_cost(
input=score, name="cost_%dmin" % ((i + 1) * 5), label=label)
output_label.append(cls)
outputs(output_label)
doc/faq/index_cn.rst
浏览文件 @ b23d99dd
......@@ -72,7 +72,7 @@ PaddlePaddle支持非常多的优化算法(Optimizer)，不同的优化算法需
减少数据载入的耗时
++++++++++++++++++
使用 :code:`pydataprovider`时，可以减少缓存池的大小，同时设置内存缓存功能，即可以极大的加速数据载入流程。
使用\ :code:`pydataprovider`\ 时，可以减少缓存池的大小，同时设置内存缓存功能，即可以极大的加速数据载入流程。
:code:`DataProvider` 缓存池的减小，和之前减小通过减小缓存池来减小内存占用的原理一致。
.. literalinclude:: src/reduce_min_pool_size.py
......
doc/howto/deep_model/rnn/index_cn.rst
浏览文件 @ b23d99dd
......@@ -4,6 +4,7 @@ RNN相关模型
.. toctree::
:maxdepth: 1
rnn_config_cn.rst
recurrent_group_cn.md
hierarchical_layer_cn.rst
hrnn_rnn_api_compare_cn.rst
doc/howto/deep_model/rnn/rnn_cn.md 已删除 100644 → 0
浏览文件 @ 343d9973
RNN 配置
=================
本教程将指导你如何在 PaddlePaddle 中配置循环神经网络（RNN）。PaddlePaddle 高度支持灵活和高效的循环神经网络配置。 在本教程中，您将了解如何：
- 准备用来学习循环神经网络的序列数据。
- 配置循环神经网络架构。
- 使用学习完成的循环神经网络模型生成序列。
我们将使用 vanilla 循环神经网络和 sequence to sequence 模型来指导你完成这些步骤。sequence to sequence 模型的代码可以在`demo / seqToseq`找到。
准备序列数据
---------------------
PaddlePaddle 不需要对序列数据进行任何预处理，例如填充。唯一需要做的是将相应类型设置为输入。例如，以下代码段定义了三个输入。 它们都是序列，它们的大小是`src_dict`，`trg_dict`和`trg_dict`：
``` sourceCode
settings.input_types = [
integer_value_sequence(len(settings.src_dict)),
integer_value_sequence(len(settings.trg_dict)),
integer_value_sequence(len(settings.trg_dict))]
```
在`process`函数中，每个`yield`函数将返回三个整数列表。每个整数列表被视为一个整数序列：
``` sourceCode
yield src_ids, trg_ids, trg_ids_next
```
有关如何编写数据提供程序的更多细节描述，请参考 [PyDataProvider2](../../ui/data_provider/index.html)。完整的数据提供文件在 `demo/seqToseq/dataprovider.py`。
配置循环神经网络架构
-----------------------------------------------
### 简单门控循环神经网络(Gated Recurrent Neural Network)
循环神经网络在每个时间步骤顺序地处理序列。下面列出了 LSTM 的架构的示例。
![image](../../../tutorials/sentiment_analysis/bi_lstm.jpg)
一般来说，循环网络从 *t* = 1 到 *t* = *T* 或者反向地从 *t* = *T* 到 *t* = 1 执行以下操作。
*x*<sub>*t* + 1</sub> = *f*<sub>*x*</sub>(*x*<sub>*t*</sub>),*y*<sub>*t*</sub> = *f*<sub>*y*</sub>(*x*<sub>*t*</sub>)
其中 *f*<sub>*x*</sub>(.) 称为**单步函数**（即单时间步执行的函数，step function），而 *f*<sub>*y*</sub>(.) 称为**输出函数**。在 vanilla 循环神经网络中，单步函数和输出函数都非常简单。然而，PaddlePaddle 可以通过修改这两个函数来实现复杂的网络配置。我们将使用 sequence to sequence 模型演示如何配置复杂的循环神经网络模型。在本节中，我们将使用简单的 vanilla 循环神经网络作为使用`recurrent_group`配置简单循环神经网络的例子。 注意，如果你只需要使用简单的RNN，GRU或LSTM，那么推荐使用`grumemory`和`lstmemory`，因为它们的计算效率比`recurrent_group`更高。
对于 vanilla RNN，在每个时间步长，**单步函数**为：
*x*<sub>*t* + 1</sub> = *W*<sub>*x*</sub>*x*<sub>*t*</sub> + *W*<sub>*i*</sub>*I*<sub>*t*</sub> + *b*
其中 *x*<sub>*t*</sub> 是RNN状态，并且 *I*<sub>*t*</sub> 是输入，*W*<sub>*x*</sub> 和 *W*<sub>*i*</sub> 分别是RNN状态和输入的变换矩阵。*b* 是偏差。它的**输出函数**只需要*x*<sub>*t*</sub>作为输出。
`recurrent_group`是构建循环神经网络的最重要的工具。 它定义了**单步函数**，**输出函数**和循环神经网络的输入。注意，这个函数的`step`参数需要实现`step function`（单步函数）和`output function`（输出函数）：
``` sourceCode
def simple_rnn(input,
size=None,
name=None,
reverse=False,
rnn_bias_attr=None,
act=None,
rnn_layer_attr=None):
def __rnn_step__(ipt):
out_mem = memory(name=name, size=size)
rnn_out = mixed_layer(input = [full_matrix_projection(ipt),
full_matrix_projection(out_mem)],
name = name,
bias_attr = rnn_bias_attr,
act = act,
layer_attr = rnn_layer_attr,
size = size)
return rnn_out
return recurrent_group(name='%s_recurrent_group' % name,
step=__rnn_step__,
reverse=reverse,
input=input)
```
PaddlePaddle 使用“Memory”（记忆模块）实现单步函数。**Memory**是在PaddlePaddle中构造循环神经网络时最重要的概念。 Memory是在单步函数中循环使用的状态，例如*x*<sub>*t* + 1</sub> = *f*<sub>*x*</sub>(*x*<sub>*t*</sub>)。 一个Memory包含**输出**和**输入**。当前时间步处的Memory的输出作为下一时间步Memory的输入。Memory也可以具有**boot layer(引导层)**，其输出被用作Memory的初始值。 在我们的例子中，门控循环单元的输出被用作输出Memory。请注意，`rnn_out`层的名称与`out_mem`的名称相同。这意味着`rnn_out` (*x*<sub>*t* + 1</sub>)的输出被用作`out_mem`Memory的**输出**。
Memory也可以是序列。在这种情况下，在每个时间步中，我们有一个序列作为循环神经网络的状态。这在构造非常复杂的循环神经网络时是有用的。 其他高级功能包括定义多个Memory，以及使用子序列来定义分级循环神经网络架构。
我们在函数的结尾返回`rnn_out`。 这意味着 `rnn_out` 层的输出被用作门控循环神经网络的**输出**函数。
### Sequence to Sequence Model with Attention
我们将使用 sequence to sequence model with attention 作为例子演示如何配置复杂的循环神经网络模型。该模型的说明如下图所示。
![image](../../../tutorials/text_generation/encoder-decoder-attention-model.png)
在这个模型中，源序列 *S* = {*s*<sub>1</sub>, …, *s*<sub>*T*</sub>} 用双向门控循环神经网络编码。双向门控循环神经网络的隐藏状态 *H*<sub>*S*</sub> = {*H*<sub>1</sub>, …, *H*<sub>*T*</sub>} 被称为 *编码向量*。解码器是门控循环神经网络。当解读每一个*y*<sub>*t*</sub>时, 这个门控循环神经网络生成一系列权重 *W*<sub>*S*</sub><sup>*t*</sup> = {*W*<sub>1</sub><sup>*t*</sup>, …, *W*<sub>*T*</sub><sup>*t*</sup>}, 用于计算编码向量的加权和。加权和用来生成*y*<sub>*t*</sub>。
模型的编码器部分如下所示。它叫做`grumemory`来表示门控循环神经网络。如果网络架构简单，那么推荐使用循环神经网络的方法，因为它比 `recurrent_group` 更快。我们已经实现了大多数常用的循环神经网络架构，可以参考 [Layers](../../ui/api/trainer_config_helpers/layers_index.html) 了解更多细节。
我们还将编码向量投射到 `decoder_size` 维空间。这通过获得反向循环网络的第一个实例，并将其投射到 `decoder_size` 维空间完成：
``` sourceCode
# 定义源语句的数据层
src_word_id = data_layer(name='source_language_word', size=source_dict_dim)
# 计算每个词的词向量
src_embedding = embedding_layer(
input=src_word_id,
size=word_vector_dim,
param_attr=ParamAttr(name='_source_language_embedding'))
# 应用前向循环神经网络
src_forward = grumemory(input=src_embedding, size=encoder_size)
# 应用反向递归神经网络（reverse=True表示反向循环神经网络）
src_backward = grumemory(input=src_embedding,
size=encoder_size,
reverse=True)
# 将循环神经网络的前向和反向部分混合在一起
encoded_vector = concat_layer(input=[src_forward, src_backward])
# 投射编码向量到 decoder_size
encoder_proj = mixed_layer(input = [full_matrix_projection(encoded_vector)],
size = decoder_size)
# 计算反向RNN的第一个实例
backward_first = first_seq(input=src_backward)
# 投射反向RNN的第一个实例到 decoder size
decoder_boot = mixed_layer(input=[full_matrix_projection(backward_first)], size=decoder_size, act=TanhActivation())
```
解码器使用 `recurrent_group` 来定义循环神经网络。单步函数和输出函数在 `gru_decoder_with_attention` 中定义：
``` sourceCode
group_inputs=[StaticInput(input=encoded_vector,is_seq=True),
StaticInput(input=encoded_proj,is_seq=True)]
trg_embedding = embedding_layer(
input=data_layer(name='target_language_word',
size=target_dict_dim),
size=word_vector_dim,
param_attr=ParamAttr(name='_target_language_embedding'))
group_inputs.append(trg_embedding)
# 对于配备有注意力机制的解码器，在训练中，
# 目标向量（groudtruth）是数据输入，
# 而源序列的编码向量可以被无边界的memory访问
# StaticInput 意味着不同时间步的输入都是相同的值，
# 否则它以一个序列输入，不同时间步的输入是不同的。
# 所有输入序列应该有相同的长度。
decoder = recurrent_group(name=decoder_group_name,
step=gru_decoder_with_attention,
input=group_inputs)
```
单步函数的实现如下所示。首先，它定义解码网络的**Memory**。然后定义 attention，门控循环单元单步函数和输出函数：
``` sourceCode
def gru_decoder_with_attention(enc_vec, enc_proj, current_word):
# 定义解码器的Memory
# Memory的输出定义在 gru_step 内
# 注意 gru_step 应该与它的Memory名字相同
decoder_mem = memory(name='gru_decoder',
size=decoder_size,
boot_layer=decoder_boot)
# 计算 attention 加权编码向量
context = simple_attention(encoded_sequence=enc_vec,
encoded_proj=enc_proj,
decoder_state=decoder_mem)
# 混合当前词向量和attention加权编码向量
decoder_inputs = mixed_layer(inputs = [full_matrix_projection(context),
full_matrix_projection(current_word)],
size = decoder_size * 3)
# 定义门控循环单元循环神经网络单步函数
gru_step = gru_step_layer(name='gru_decoder',
input=decoder_inputs,
output_mem=decoder_mem,
size=decoder_size)
# 定义输出函数
out = mixed_layer(input=[full_matrix_projection(input=gru_step)],
size=target_dict_dim,
bias_attr=True,
act=SoftmaxActivation())
return out
```
生成序列
-----------------
训练模型后，我们可以使用它来生成序列。通常的做法是使用**beam search** 生成序列。以下代码片段定义 beam search 算法。注意，`beam_search` 函数假设 `step` 的输出函数返回的是下一个时刻输出词的 softmax 归一化概率向量。我们对模型进行了以下更改。
- 使用 `GeneratedInput` 来表示 trg\_embedding。 `GeneratedInput` 将上一时间步所生成的词的向量来作为当前时间步的输入。
- 使用 `beam_search` 函数。这个函数需要设置：
- `bos_id`: 开始标记。每个句子都以开始标记开头。
- `eos_id`: 结束标记。每个句子都以结束标记结尾。
- `beam_size`: beam search 算法中的beam大小。
- `max_length`: 生成序列的最大长度。
- 使用 `seqtext_printer_evaluator` 根据索引矩阵和字典打印文本。这个函数需要设置：
- `id_input`: 数据的整数ID，用于标识生成的文件中的相应输出。
- `dict_file`: 用于将词ID转换为词的字典文件。
- `result_file`: 生成结果文件的路径。
代码如下：
``` sourceCode
group_inputs=[StaticInput(input=encoded_vector,is_seq=True),
StaticInput(input=encoded_proj,is_seq=True)]
# 在生成时，解码器基于编码源序列和最后生成的目标词预测下一目标词。
# 编码源序列（编码器输出）必须由只读Memory的 StaticInput 指定。
# 这里， GeneratedInputs 自动获取上一个生成的词，并在最开始初始化为起始词，如 <s>。
trg_embedding = GeneratedInput(
size=target_dict_dim,
embedding_name='_target_language_embedding',
embedding_size=word_vector_dim)
group_inputs.append(trg_embedding)
beam_gen = beam_search(name=decoder_group_name,
step=gru_decoder_with_attention,
input=group_inputs,
bos_id=0, # Beginnning token.
eos_id=1, # End of sentence token.
beam_size=beam_size,
max_length=max_length)
seqtext_printer_evaluator(input=beam_gen,
id_input=data_layer(name="sent_id", size=1),
dict_file=trg_dict_path,
result_file=gen_trans_file)
outputs(beam_gen)
```
注意，这种生成技术只用于类似解码器的生成过程。如果你正在处理序列标记任务，请参阅 [Semantic Role Labeling Demo](../../demo/semantic_role_labeling/index.html) 了解更多详细信息。
完整的配置文件在`demo/seqToseq/seqToseq_net.py`。
doc/howto/deep_model/rnn_config_cn.rst → doc/howto/deep_model/rnn/rnn_config_cn.rst
浏览文件 @ b23d99dd
RNN 配置
RNN配置
========
本教程将指导你如何在 PaddlePaddle
......@@ -20,7 +20,7 @@ PaddlePaddle
不需要对序列数据进行任何预处理，例如填充。唯一需要做的是将相应类型设置为输入。例如，以下代码段定义了三个输入。
它们都是序列，它们的大小是\ ``src_dict``\ ，\ ``trg_dict``\ 和\ ``trg_dict``\ ：
.. code:: sourcecode
.. code:: python
settings.input_types = [
integer_value_sequence(len(settings.src_dict)),
......@@ -29,12 +29,11 @@ PaddlePaddle
在\ ``process``\ 函数中，每个\ ``yield``\ 函数将返回三个整数列表。每个整数列表被视为一个整数序列：
.. code:: sourcecode
.. code:: python
yield src_ids, trg_ids, trg_ids_next
有关如何编写数据提供程序的更多细节描述，请参考
`PyDataProvider2 <../../ui/data_provider/index.html>`__\ 。完整的数据提供文件在
有关如何编写数据提供程序的更多细节描述，请参考 :ref:`api_pydataprovider2` 。完整的数据提供文件在
``demo/seqToseq/dataprovider.py``\ 。
配置循环神经网络架构
......@@ -45,18 +44,17 @@ PaddlePaddle
循环神经网络在每个时间步骤顺序地处理序列。下面列出了 LSTM 的架构的示例。
.. figure:: ../../../tutorials/sentiment_analysis/bi_lstm.jpg
:alt: image
.. image:: ../../../tutorials/sentiment_analysis/bi_lstm.jpg
:align: center
image
一般来说，循环网络从 :math:`t=1` 到 :math:`t=T` 或者反向地从 :math:`t=T` 到 :math:`t=1` 执行以下操作。
一般来说，循环网络从 *t* = 1 到 *t* = *T* 或者反向地从 *t* = *T* 到 *t*
= 1 执行以下操作。
.. math::
*x*\ \ *t* + 1 = *f*\ \ *x*\ (*x*\ \ *t*\ ),\ *y*\ \ *t*\  = *f*\ \ *y*\ (*x*\ \ *t*\ )
x_{t+1} = f_x(x_t), y_t = f_y(x_t)
其中 *f*\ \ *x*\ (.) 称为\ **单步函数**\ （即单时间步执行的函数，step
function），而 *f*\ \ *y*\ (.) 称为\ **输出函数**\ 。在 vanilla
其中 :math:`f_x(.)` 称为\ **单步函数**\ （即单时间步执行的函数，step
function），而 :math:`f_y(.)` 称为\ **输出函数**\ 。在 vanilla
循环神经网络中，单步函数和输出函数都非常简单。然而，PaddlePaddle
可以通过修改这两个函数来实现复杂的网络配置。我们将使用 sequence to
sequence
......@@ -67,16 +65,17 @@ vanilla
对于 vanilla RNN，在每个时间步长，\ **单步函数**\ 为：
*x*\ \ *t* + 1 = *W*\ \ *x*\ \ *x*\ \ *t*\  + *W*\ \ *i*\ \ *I*\ \ *t*\  + *b*
.. math::
其中 *x*\ \ *t*\ 是RNN状态，并且 *I*\ \ *t*\ 是输入，\ *W*\ \ *x*\ 和
*W*\ \ *i*\ 分别是RNN状态和输入的变换矩阵。\ *b*
是偏差。它的\ **输出函数**\ 只需要\ *x*\ \ *t*\ 作为输出。
x_{t+1} = W_x x_t + W_i I_t + b
其中 :math:`x_t` 是RNN状态，并且 :math:`I_t` 是输入，:math:`W_x` 和
:math:`W_i` 分别是RNN状态和输入的变换矩阵。:math:`b` 是偏差。它的\ **输出函数**\ 只需要 :math:`x_t` 作为输出。
``recurrent_group``\ 是构建循环神经网络的最重要的工具。
它定义了\ **单步函数**\ ，\ **输出函数**\ 和循环神经网络的输入。注意，这个函数的\ ``step``\ 参数需要实现\ ``step function``\ （单步函数）和\ ``output function``\ （输出函数）：
.. code:: sourcecode
.. code:: python
def simple_rnn(input,
size=None,
......@@ -102,7 +101,7 @@ vanilla
PaddlePaddle
使用“Memory”（记忆模块）实现单步函数。\ **Memory**\ 是在PaddlePaddle中构造循环神经网络时最重要的概念。
Memory是在单步函数中循环使用的状态，例如\ *x*\ \ *t* + 1 = *f*\ \ *x*\ (*x*\ \ *t*\ )。
Memory是在单步函数中循环使用的状态，例如 :math:`x_{t+1} = f_x(x_t)` 。
一个Memory包含\ **输出**\ 和\ **输入**\ 。当前时间步处的Memory的输出作为下一时间步Memory的输入。Memory也可以具有\ **boot
layer(引导层)**\ ，其输出被用作Memory的初始值。
在我们的例子中，门控循环单元的输出被用作输出Memory。请注意，\ ``rnn_out``\ 层的名称与\ ``out_mem``\ 的名称相同。这意味着\ ``rnn_out``
......@@ -120,30 +119,25 @@ Sequence to Sequence Model with Attention
我们将使用 sequence to sequence model with attention
作为例子演示如何配置复杂的循环神经网络模型。该模型的说明如下图所示。
.. figure:: ../../../tutorials/text_generation/encoder-decoder-attention-model.png
:alt: image
image
.. image:: ../../../tutorials/text_generation/encoder-decoder-attention-model.png
:align: center
在这个模型中，源序列 *S* = {*s*\ 1, …, \ *s*\ \ *T*\ }
在这个模型中，源序列 :math:`S = \{s_1, \dots, s_T\}`
用双向门控循环神经网络编码。双向门控循环神经网络的隐藏状态
*H*\ \ *S*\  = {*H*\ 1, …, \ *H*\ \ *T*\ } 被称为
*编码向量*\ 。解码器是门控循环神经网络。当解读每一个\ *y*\ \ *t*\ 时,
这个门控循环神经网络生成一系列权重
*W*\ \ *S*\ \ *t*\  = {*W*\ 1\ *t*\ , …, \ *W*\ \ *T*\ \ *t*\ },
用于计算编码向量的加权和。加权和用来生成\ *y*\ \ *t*\ 。
:math:`H_S = \{H_1, \dots, H_T\}` 被称为
*编码向量*\ 。解码器是门控循环神经网络。当解读每一个 :math:`y_t` 时,
这个门控循环神经网络生成一系列权重 :math:`W_S^t = \{W_1^t, \dots, W_T^t\}` ,
用于计算编码向量的加权和。加权和用来生成 :math:`y_t` 。
模型的编码器部分如下所示。它叫做\ ``grumemory``\ 来表示门控循环神经网络。如果网络架构简单，那么推荐使用循环神经网络的方法，因为它比
``recurrent_group``
更快。我们已经实现了大多数常用的循环神经网络架构，可以参考
`Layers <../../ui/api/trainer_config_helpers/layers_index.html>`__
了解更多细节。
更快。我们已经实现了大多数常用的循环神经网络架构，可以参考 :ref:`api_trainer_config_helpers_layers` 了解更多细节。
我们还将编码向量投射到 ``decoder_size``
维空间。这通过获得反向循环网络的第一个实例，并将其投射到
``decoder_size`` 维空间完成：
.. code:: sourcecode
.. code:: python
# 定义源语句的数据层
src_word_id = data_layer(name='source_language_word', size=source_dict_dim)
......@@ -174,7 +168,7 @@ Sequence to Sequence Model with Attention
解码器使用 ``recurrent_group`` 来定义循环神经网络。单步函数和输出函数在
``gru_decoder_with_attention`` 中定义：
.. code:: sourcecode
.. code:: python
group_inputs=[StaticInput(input=encoded_vector,is_seq=True),
StaticInput(input=encoded_proj,is_seq=True)]
......@@ -198,7 +192,7 @@ Sequence to Sequence Model with Attention
单步函数的实现如下所示。首先，它定义解码网络的\ **Memory**\ 。然后定义
attention，门控循环单元单步函数和输出函数：
.. code:: sourcecode
.. code:: python
def gru_decoder_with_attention(enc_vec, enc_proj, current_word):
# 定义解码器的Memory
......@@ -253,7 +247,7 @@ attention，门控循环单元单步函数和输出函数：
代码如下：
.. code:: sourcecode
.. code:: python
group_inputs=[StaticInput(input=encoded_vector,is_seq=True),
StaticInput(input=encoded_proj,is_seq=True)]
......@@ -279,9 +273,6 @@ attention，门控循环单元单步函数和输出函数：
result_file=gen_trans_file)
outputs(beam_gen)
注意，这种生成技术只用于类似解码器的生成过程。如果你正在处理序列标记任务，请参阅
`Semantic Role Labeling
Demo <../../demo/semantic_role_labeling/index.html>`__
了解更多详细信息。
注意，这种生成技术只用于类似解码器的生成过程。如果你正在处理序列标记任务，请参阅 :ref:`semantic_role_labeling` 了解更多详细信息。
完整的配置文件在\ ``demo/seqToseq/seqToseq_net.py``\ 。
doc/howto/dev/new_layer_cn.rst 0 → 100644
浏览文件 @ b23d99dd
================
实现新的网络层
================
这份教程展示了如何在PaddlePaddle中实现一个自定义的网络层。在这里我们使用全连接层作为例子来展示实现新网络层所需要的四个步骤。
1. 推导该层前向和后向传递的方程。
2. 实现该层的C++类。
3. 增加梯度检测的单元测试，以保证梯度的正确计算。
4. 封装该层的Python接口。
推导方程
================
首先我们需要推导该网络层的*前向传播*和*后向传播*的方程。前向传播给定输入，计算输出。后向传播给定输出的梯度，计算输入和参数的梯度。
下图是一个全连接层的示意图。在全连接层中，每个输出节点都连接到所有的输入节点上。
.. image:: FullyConnected.jpg
:align: center
:scale: 60 %
一个网络层的前向传播部分把输入转化为相应的输出。
全连接层以一个维度为 :math:`D_i` 的稠密向量作为输入，使用一个尺度为 :math:`D_i \times D_o` 的变换矩阵 :math:`W` 把 :math:`x` 映射到一个维度为 :math:`D_o` 的向量，并在乘积结果上再加上维度为 :math:`D_o` 的偏置向量 :math:`b` 。
.. math::
y = f(W^T x + b)
其中 :math:`f(.)` 是一个非线性的*激活方程*，例如sigmoid， tanh，以及Relu。
变换矩阵 :math:`W` 和偏置向量 :math:`b` 是该网络层的*参数*。一个网络层的参数是在*反向传播*时被训练的。反向传播根据输出的梯度，分别计算每个参数的梯度，以及输入的梯度。优化器则用链式法则来对每个参数计算损失函数的梯度。
假设损失函数是 :math:`c(y)` ，那么
.. math::
\frac{\partial c(y)}{\partial x} = \frac{\partial c(y)}{\partial y} \frac{\partial y}{\partial x}
假设 :math:`z = f(W^T x + b)` ，那么
.. math::
\frac{\partial y}{\partial z} = \frac{\partial f(z)}{\partial z}
PaddlePaddle的base layer类可以自动计算上面的导数。
因此，对全连接层来说，我们需要计算：
.. math::
\frac{\partial z}{\partial x} = W, \frac{\partial z_j}{\partial W_{ij}} = x_i, \frac{\partial z}{\partial b} = \mathbf 1
其中 :math:`\mathbf 1` 是一个全1的向量， :math:`W_{ij}` 是矩阵 :math:`W` 第i行第j列的数值， :math:`z_j` 是向量 :math:`z` 的第j个值， :math:`x_i` 是向量 :math:`x` 的第i个值。
最后我们使用链式法则计算 :math:`\frac{\partial z}{\partial x}` 以及 :math:`\frac{\partial z}{\partial W}` 。计算的细节将在下面的小节给出。
实现C++类
===================
一个网络层的C++类需要实现初始化，前向和后向。全连接层的实现位于:code:`paddle/gserver/layers/FullyConnectedLayer.h`及:code:`paddle/gserver/layers/FullyConnectedLayer.cpp`。这里我们展示一份简化过的代码。
这个类需要继承 :code:`paddle::Layer` 这个基类，并且需要重写基类中的以下几个虚函数：
- 类的构造函数和析构函数。
- :code:`init` 函数。用于初始化参数和设置。
- :code:`forward` 。实现网络层的前向传播。
- :code:`backward` 。实现网络层的后向传播。
- :code:`prefetch` 。用来从参数服务器预取参数矩阵相应的行。如果网络层不需要远程稀疏更新，则不需要重写该函数。（大多数网络层不需要支持远程稀疏更新）
头文件如下：
.. code-block:: c++
namespace paddle {
/**
* 全连接层的每个输出都连接到上一层的所有的神经元上。
* 它的输入与经过学习的参数做内积并加上偏置（可选）。
*
* 配置文件接口是fc_layer。
*/
class FullyConnectedLayer : public Layer {
protected:
WeightList weights_;
std::unique_ptr<Weight> biases_;
public:
explicit FullyConnectedLayer(const LayerConfig& config)
: Layer(config) {}
~FullyConnectedLayer() {}
bool init(const LayerMap& layerMap, const ParameterMap& parameterMap);
Weight& getWeight(int idx) { return *weights_[idx]; }
void prefetch();
void forward(PassType passType);
void backward(const UpdateCallback& callback = nullptr);
};
} // namespace paddle
头文件中把参数定义为类的成员变量。我们使用 :code:`Weight` 类作为参数的抽象，它支持多线程更新。该类的实现细节在“实现细节”中详细介绍。
- :code:`weights_` 是存有一系列变换矩阵的权重。在当前的实现方式下，网络层可以有多个输入。因此，它可能有不止一个权重。每个权重对应一个输入。
- :code:`biases_` 是存有偏置向量的权重。
全连接层没有网络层配置的超参数。如果一个网络层需要配置的话，通常的做法是将配置存于 :code:`LayerConfig& config` 中，并在类构建函数中把它放入一个类成员变量里。
下面的代码片段实现了 :code:`init` 函数。
- 首先，所有的 :code:`init` 函数必须先调用基类中的函数 :code:`Layer::init(layerMap, parameterMap);` 。该语句会为每个层初始化其所需要的变量和连接。
- 之后初始化所有的权重矩阵 :math:`W` 。当前的实现方式下，网络层可以有多个输入。因此，它可能有不止一个权重。
- 最后，初始化偏置向量。
.. code-block:: c++
bool FullyConnectedLayer::init(const LayerMap& layerMap,
const ParameterMap& parameterMap) {
/* 初始化父类 */
Layer::init(layerMap, parameterMap);
/* 初始化权重表 */
CHECK(inputLayers_.size() == parameters_.size());
for (size_t i = 0; i < inputLayers_.size(); i++) {
// 获得参数尺寸
size_t height = inputLayers_[i]->getSize();
size_t width = getSize();
// 新建一个权重
if (parameters_[i]->isSparse()) {
CHECK_LE(parameters_[i]->getSize(), width * height);
} else {
CHECK_EQ(parameters_[i]->getSize(), width * height);
}
Weight* w = new Weight(height, width, parameters_[i]);
// 将新建的权重加入权重表
weights_.emplace_back(w);
}
/* 初始化biases_ */
if (biasParameter_.get() != NULL) {
biases_ = std::unique_ptr<Weight>(new Weight(1, getSize(), biasParameter_));
}
return true;
}
实现前向传播的部分有下面几个步骤。
- 每个层在其 :code:`forward` 函数的开头必须调用 :code:`Layer::forward(passType);` 。
- 之后使用 :code:`reserveOutput(batchSize, size);` 为输出分配内存。由于我们支持训练数据有不同的批次大小，所以这一步是必要的。 :code:`reserveOutput` 会相应地改变输出的尺寸。为了保证效率，如果需要扩大矩阵，我们会重新分配内存；如果需要缩减矩阵，我们会继续使用现有的内存块。
- 之后使用矩阵运算函数来计算 :math:`\sum_i W_i x + b`。:code:`getInput(i).value` 返回第i个输入矩阵。每个输入都是一个 :math:`batchSize \times dim` 的矩阵，每行表示一个批次中的单个输入。对于我们支持的全部矩阵操作，请参考 :code:`paddle/math/Matrix.h`和:code:`paddle/math/BaseMatrix.h` 。
- 最终，使用 :code:`forwardActivation();` 进行激活操作。这会自动进行网络配置中声明的激活操作。
.. code-block:: c++
void FullyConnectedLayer::forward(PassType passType) {
Layer::forward(passType);
/* 若有必要，为output_申请内存 */
int batchSize = getInput(0).getBatchSize();
int size = getSize();
{
// 设置输出的尺寸
reserveOutput(batchSize, size);
}
MatrixPtr outV = getOutputValue();
// 对每个输入乘上变换矩阵
for (size_t i = 0; i != inputLayers_.size(); ++i) {
auto input = getInput(i);
CHECK(input.value) << "The input of 'fc' layer must be matrix";
i == 0 ? outV->mul(input.value, weights_[i]->getW(), 1, 0)
: outV->mul(input.value, weights_[i]->getW(), 1, 1);
}
/* 加上偏置向量 */
if (biases_.get() != NULL) {
outV->addBias(*(biases_->getW()), 1);
}
/* 激活 */ {
forwardActivation();
}
}
实现后向传播的部分有下面几个步骤。
- :code:`backwardActivation()` 计算激活函数的梯度。通过 :code:`getOutputGrad()` 来获得输出的梯度，调用该函数后，梯度会就地（不使用额外空间）乘上输出的梯度。
- 计算偏置的梯度。注意，我们使用 :code:`biases_->getWGrad()` 来得到某个特定参数的梯度矩阵。在一个参数的梯度被更新后，**必须**要调用 :code:`getParameterPtr()->incUpdate(callback);` 。这用于在多线程和多机上更新参数。
- 最后，计算转换矩阵和输入的梯度，并对相应的参数调用 :code:`incUpdate` 。PaddlePaddle可以通过该机制判断是否已经收集齐所有的梯度，从而可以做一些与计算重叠的工作（例如，网络通信）。
.. code-block:: c++
void FullyConnectedLayer::backward(const UpdateCallback& callback) {
/* 对激活求导 */ {
backwardActivation();
}
if (biases_ && biases_->getWGrad()) {
biases_->getWGrad()->collectBias(*getOutputGrad(), 1);
biases_->getParameterPtr()->incUpdate(callback);
}
bool syncFlag = hl_get_sync_flag();
for (size_t i = 0; i != inputLayers_.size(); ++i) {
/* 计算当前层权重的梯度 */
if (weights_[i]->getWGrad()) {
MatrixPtr input_T = getInputValue(i)->getTranspose();
MatrixPtr oGrad = getOutputGrad();
{
weights_[i]->getWGrad()->mul(input_T, oGrad, 1, 1);
}
}
/* 计算输入层的偏差 */
MatrixPtr preGrad = getInputGrad(i);
if (NULL != preGrad) {
MatrixPtr weights_T = weights_[i]->getW()->getTranspose();
preGrad->mul(getOutputGrad(), weights_T, 1, 1);
}
{
weights_[i]->getParameterPtr()->incUpdate(callback);
}
}
}
:code:`prefetch` 函数指出了在训练时需要从参数服务器取出的行。仅在远程稀疏训练时有效。使用远程稀疏方式训练时，完整的参数矩阵被分布在不同的参数服务器上。当网络层用一个批次做训练时，该批次的输入中仅有一个子集是非零的。因此，该层仅需要这些非零样本位置所对应的变换矩阵的那些行。 :code:`prefetch` 表明了这些行的标号。
大多数层不需要远程稀疏训练函数。这种情况下不需要重写该函数。
.. code-block:: c++
void FullyConnectedLayer::prefetch() {
for (size_t i = 0; i != inputLayers_.size(); ++i) {
auto* sparseParam =
dynamic_cast<SparsePrefetchRowCpuMatrix*>(weights_[i]->getW().get());
if (sparseParam) {
MatrixPtr input = getInputValue(i);
sparseParam->addRows(input);
}
}
}
最后，使用 :code:`REGISTER_LAYER(fc, FullyConnectedLayer);` 来注册该层。 :code:`fc` 是该层的标识符， :code:`FullyConnectedLayer` 是该层的类名。
.. code-block:: c++
namespace paddle {
REGISTER_LAYER(fc, FullyConnectedLayer);
}
若 :code:`cpp` 被放在 :code:`paddle/gserver/layers` 目录下，其会自动被加入编译列表。
写梯度检查单元测试
===============================
写梯度检查单元测试是一个验证新实现的层是否正确的相对简单的办法。梯度检查单元测试通过有限差分法来验证一个层的梯度。首先对输入做一个小的扰动 :math:`\Delta x` ，然后观察到输出的变化为 :math:`\Delta y` ，那么，梯度就可以通过这个方程计算得到 :math:`\frac{\Delta y}{\Delta x }` 。之后，再用这个梯度去和 :code:`backward` 函数得到的梯度去对比，以保证梯度计算的正确性。需要注意的是梯度检查仅仅验证了梯度的计算，并不保证 :code:`forward` 和 :code:`backward` 函数的实现是正确的。你需要一些更复杂的单元测试来保证你实现的网络层是正确的。
所有网络层的梯度检查单测都位于 :code:`paddle/gserver/tests/test_LayerGrad.cpp` 。我们建议你在写新网络层时把测试代码放入新的文件中。下面列出了全连接层的梯度检查单元测试。它包含以下几步：
+ 生成网络层配置。网络层配置包含以下几项：
- 偏置参数的大小。（例子中是4096）
- 层的类型。（例子中是fc）
- 层的大小。（例子中是4096）
- 激活的类型。（例子中是softmax）
- dropout的比例。（例子中是0.1）
+ 配置网络层的输入。在这个例子里，我们仅有一个输入。
- 输入的类型（ :code:`INPUT_DATA` ），可以是以下几种：
- :code:`INPUT_DATA` ：稠密向量。
- :code:`INPUT_LABEL` ：整数。
- :code:`INPUT_DATA_TARGET` ：稠密向量，但不用于计算梯度。
- :code:`INPUT_SEQUENCE_DATA` ：含有序列信息的稠密向量。
- :code:`INPUT_HASSUB_SEQUENCE_DATA` ：含有序列信息和子序列信息的稠密向量。
- :code:`INPUT_SEQUENCE_LABEL` ：含有序列信息的整数。
- :code:`INPUT_SPARSE_NON_VALUE_DATA` ：0-1稀疏数据。
- :code:`INPUT_SPARSE_FLOAT_VALUE_DATA` ：浮点稀疏数据。
- 输入的名字。（例子中是 :code:`layer_0` ）
- 输入的大小。（例子中是8192）
- 非零数字的个数，仅对稀疏数据有效。
- 稀疏数据的格式，仅对稀疏数据有效。
+ 对每个输入，都需要调用一次 :code:`config.layerConfig.add_inputs();` 。
+ 调用 :code:`testLayerGrad` 来做梯度检查。它包含以下参数。
- 层和输入的配置。（例子中是 :code:`config` ）
- 网络层的类型。（例子中是 :code:`fc` ）
- 梯度检查的输入数据的批次大小。（例子中是100）
- 输入是否是转置的。大多数层需要设置为 :code:`false` 。（例子中是 :code:`false` ）
- 是否使用权重。有些层或者激活需要做归一化以保证它们的输出的和是一个常数。例如，softmax激活的输出的和总是1。在这种情况下，我们不能通过常规的梯度检查的方式来计算梯度。因此我们采用输出的加权和（非常数）来计算梯度。（例子中是 :code:`true` ，因为全连接层的激活可以是softmax）
.. code-block:: c++
void testFcLayer(string format, size_t nnz) {
// Create layer configuration.
TestConfig config;
config.biasSize = 4096;
config.layerConfig.set_type("fc");
config.layerConfig.set_size(4096);
config.layerConfig.set_active_type("softmax");
config.layerConfig.set_drop_rate(0.1);
// Setup inputs.
config.inputDefs.push_back(
{INPUT_DATA, "layer_0", 8192, nnz, ParaSparse(format)});
config.layerConfig.add_inputs();
LOG(INFO) << config.inputDefs[0].sparse.sparse << " "
<< config.inputDefs[0].sparse.format;
for (auto useGpu : {false, true}) {
testLayerGrad(config, "fc", 100, /* trans */ false, useGpu,
/* weight */ true);
}
}
如果你要为了测试而增加新的文件，例如 :code:`paddle/gserver/tests/testFCGrad.cpp` ，你需要把该文件加入 :code:`paddle/gserver/tests/CMakeLists.txt` 中。下面给出了一个例子。当你执行命令 :code:`make tests` 时，所有的单测都会被执行一次。注意，有些层可能需要高精度来保证梯度检查单测正确执行。你需要在配置cmake时将 :code:`WITH_DOUBLE` 设置为 `ON` 。
.. code-block:: bash
add_unittest_without_exec(test_FCGrad
test_FCGrad.cpp
LayerGradUtil.cpp
TestUtil.cpp)
add_test(NAME test_FCGrad
COMMAND test_FCGrad)
实现python封装
========================
python封装的实现使得我们可以在配置文件中使用新实现的网络层。所有的python封装都在 :code:`python/paddle/trainer/config_parser.py` 中。全连接层python封装的例子中包含下面几步：
- 所有的Python封装都使用 :code:`@config_layer('fc')` 这样的装饰器。网络层的标识符为 :code:`fc` 。
- 实现构造函数 :code:`__init__` 。
- 它首先调用基构造函数 :code:`super(FCLayer, self).__init__(name, 'fc', size, inputs=inputs, **xargs)` 。 :code:`FCLayer` 是Python封装的类名。 :code:`fc` 是网络层的标识符。为了封装能够正确工作，这些名字必须要写对。
- 之后，计算变换矩阵的大小和格式（是否稀疏）。
.. code-block:: python
@config_layer('fc')
class FCLayer(LayerBase):
def __init__(
self,
name,
size,
inputs,
bias=True,
**xargs):
super(FCLayer, self).__init__(name, 'fc', size, inputs=inputs, **xargs)
for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index)
psize = self.config.size * input_layer.size
dims = [input_layer.size, self.config.size]
format = self.inputs[input_index].format
sparse = format == "csr" or format == "csc"
if sparse:
psize = self.inputs[input_index].nnz
self.create_input_parameter(input_index, psize, dims, sparse, format)
self.create_bias_parameter(bias, self.config.size)
在网络配置中，网络层的细节可以通过下面这些代码片段来指定。这个类的参数包括：
- :code:`name` 是网络层实例的名字标识符。
- :code:`type` 是网络层的类型，通过网络层的标识符来指定。
- :code:`size` 是网络层输出的大小。
- :code:`bias` 表明这个层的一个实例是否需要偏置。
- :code:`inputs` 说明这个层的输入，输入是由一个list中的网络层实例的名字组成的。
.. code-block:: python
Layer(
name = "fc1",
type = "fc",
size = 64,
bias = True,
inputs = [Input("pool3")]
)
我们建议你为你的Python封装实现一个“助手”，使得搭模型时更方便。具体可以参考 :code:`python/paddle/trainer_config_helpers/layers.py` 。
doc/howto/dev/new_layer_en.rst
浏览文件 @ b23d99dd
......@@ -209,7 +209,6 @@ The implementation of the backward part has the following steps.
if (biases_ && biases_->getWGrad()) {
biases_->getWGrad()->collectBias(*getOutputGrad(), 1);
/* Increasing the number of gradient */
biases_->getParameterPtr()->incUpdate(callback);
}
......@@ -297,7 +296,7 @@ All the gradient check unit tests are located in :code:`paddle/gserver/tests/tes
+ each inputs needs to call :code:`config.layerConfig.add_inputs();` once.
+ call :code:`testLayerGrad` to perform gradient checks. It has the following arguments.
- layer and input configurations. (:code:`config` in our example)
- type of the input. (:code:`fc` in our example)
- type of the layer. (:code:`fc` in our example)
- batch size of the gradient check. (100 in our example)
- whether the input is transpose. Most layers need to set it to :code:`false`. (:code:`false` in our example)
- whether to use weights. Some layers or activations perform normalization so that the sum of their output is a constant. For example, the sum of output of a softmax activation is one. In this case, we cannot correctly compute the gradients using regular gradient check techniques. A weighted sum of the output, which is not a constant, is utilized to compute the gradients. (:code:`true` in our example, because the activation of a fully connected layer can be softmax)
......@@ -310,7 +309,7 @@ All the gradient check unit tests are located in :code:`paddle/gserver/tests/tes
config.biasSize = 4096;
config.layerConfig.set_type("fc");
config.layerConfig.set_size(4096);
config.layerConfig.set_active_type("sigmoid");
config.layerConfig.set_active_type("softmax");
config.layerConfig.set_drop_rate(0.1);
// Setup inputs.
config.inputDefs.push_back(
......
doc/howto/index_cn.rst
浏览文件 @ b23d99dd
......@@ -7,10 +7,11 @@
.. toctree::
:maxdepth: 1
usage/cmd_parameter/index_cn.rst
usage/concepts/use_concepts_cn.rst
usage/cluster/cluster_train_cn.md
usage/cluster/k8s/k8s_cn.md
usage/cluster/k8s/k8s_distributed_cn.md
usage/k8s/k8s_cn.md
usage/k8s/k8s_distributed_cn.md
开发标准
--------
......
doc/howto/index_en.rst
浏览文件 @ b23d99dd
......@@ -7,8 +7,10 @@ Usage
.. toctree::
:maxdepth: 1
usage/cmd_parameter/index_en.md
usage/cmd_parameter/index_en.rst
usage/cluster/cluster_train_en.md
usage/k8s/k8s_en.md
usage/k8s/k8s_aws_en.md
Development
------------
......
doc/howto/usage/cmd_parameter/arguments_cn.md 0 → 100644
浏览文件 @ b23d99dd
# 参数概述
虽然Paddle看起来包含了众多参数，但是大部分参数是为开发者提供的，或者已经在集群提交环境中自动设置，因此用户并不需要关心它们。在此，根据这些参数的使用场合，我们将它们划分为不同的类别。例如，`通用`类别中的参数可用于所有场合。某些参数只可用于特定的层中，而有些参数需要在集群多机训练中使用等。
<html>
<table border="2" frame="border">
<thead>
<tr>
<th scope="col" class="left"></th>
<th scope="col" class="left">参数</th>
<th scope="col" class="left">本地训练</th>
<th scope="col" class="left">集群训练</th>
<th scope="col" class="left">本地测试</th>
<th scope="col" class="left">集群测试</th>
</tr>
</thead>
<tbody>
<tr>
<td class="left" rowspan="9">通用</td>
<td class="left">job</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">use_gpu</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">local</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">config</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">config_args</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">num_passes</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">trainer_count</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">version</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">show_layer_stat</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left" rowspan="15">训练</td><td class="left">dot_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">test_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">saving_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">show_parameter_stats_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">init_model_path</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left"></td>
</tr>
<tr>
<td class="left">load_missing_parameter_strategy</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">saving_period_by_batches</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">use_old_updater</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">enable_grad_share</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">grad_share_block_num</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">log_error_clipping</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">log_clipping</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">save_only_one</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">allow_inefficient_sparse_update</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">start_pass</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">训练/测试</td><td class="left">save_dir</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left" rowspan = "2">训练过程中测试</td><td class="left">test_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">average_test_period</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left" rowspan = "5">测试</td><td class="left">model_list</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">test_wait</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">test_pass</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">predict_output_dir</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">distribute_test</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">Auc/正负对验证(PnpairValidation)</td><td class="left">predict_file</td>
<td class="left"></td><td class="left"></td><td class="left"></td>√<td class="left">√</td>
</tr>
<tr>
<td class="left" rowspan = "6">GPU</td><td class="left">gpu_id</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">parallel_nn</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">allow_only_one_model_on_one_gpu</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">cudnn_dir</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">cuda_dir</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">cudnn_conv_workspace_limit_in_mb</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left" rowspan = "4">递归神经网络(RNN)</td>
<td class="left">beam_size</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">rnn_use_batch</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">prev_batch_state</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">diy_beam_search_prob_so</td>
<td class="left"></td><td class="left"></td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left" rowspan = "2">度量学习(metric learning)</td><td class="left">external</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
<tr>
<td class="left">data_server_port</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left" rowspan = "16">参数服务器(PServer)</td><td class="left">start_pserver</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">pservers</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">port</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">port_num</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">ports_num_for_sparse</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">nics</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">rdma_tcp</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">small_messages</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">loadsave_parameters_in_pserver</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left">√</td>
</tr>
<tr>
<td class="left">log_period_server</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">pserver_num_threads</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">sock_send_buf_size</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">sock_recv_buf_size</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">num_gradient_servers</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">parameter_block_size</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">parameter_block_size_for_sparse</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left" rowspan = "3">异步随机梯度下降(Async SGD)</td><td class="left">async_count</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">async_lagged_ratio_min</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">async_lagged_ratio_default</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left" rowspan = "8">性能调优(Performance Tuning)</td><td class="left">log_barrier_abstract</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">log_barrier_lowest_nodes</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">log_barrier_show_log</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">check_sparse_distribution_batches</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">check_sparse_distribution_ratio</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">check_sparse_distribution_unbalance_degree</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">check_sparse_distribution_in_pserver</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">show_check_sparse_distribution_log</td>
<td class="left"></td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">数据提供器(Data Provider)</td><td class="left">memory_threshold_on_load_data</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left" rowspan = "2">随机数</td><td class="left">seed</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">thread_local_rand_use_global_seed</td>
<td class="left">√</td><td class="left">√</td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">单元测试</td><td class="left">checkgrad_eps</td>
<td class="left"></td><td class="left"></td><td class="left"></td><td class="left"></td>
</tr>
<tr>
<td class="left">矩阵/向量</td><td class="left">enable_parallel_vector</td>
<td class="left">√</td><td class="left">√</td><td class="left">√</td><td class="left">√</td>
</tr>
</tbody>
</table>
</html>
doc/howto/usage/cmd_parameter/detail_introduction_cn.md 0 → 100644
浏览文件 @ b23d99dd
# 细节描述
## 通用
* `--job`
- 工作模式，包括: **train, test, checkgrad**，其中checkgrad主要为开发者使用，使用者不需要关心。
- 类型: string (默认: train)
* `--config`
- 用于指定网络配置文件。
- 类型: string (默认: null).
* `--use_gpu`
- 训练过程是否使用GPU，设置为true使用GPU模式，否则使用CPU模式。
- 类型: bool (默认: 1).
* `--local`
 - 训练过程是否为本地模式，设置为true使用本地训练或者使用集群上的一个节点，否则使用多机训练。
- 类型: bool (默认: 1).
* `--trainer_count`
- 指定一台机器上使用的线程数。例如，trainer_count = 4, 意思是在GPU模式下使用4个GPU，或者在CPU模式下使用4个线程。每个线程（或GPU）分配到当前数据块样本数的四分之一。也就是说，如果在训练配置中设置batch_size为512，每个线程分配到128个样本用于训练。
- 类型: int32 (默认: 1).
* `--num_passes`
- 当模式为`--job=train`时, 该参数的意思是训练num_passes轮。每轮会将数据集中的所有训练样本使用一次。当模式为`--job=test`时，意思是使用第test_pass个模型到第 num_passes-1 个模型测试数据。
- 类型: int32 (默认: 100).
* `--config_args`
- 传递给配置文件的参数。格式: key1=value1,key2=value2.
- 类型: string (默认: null).
* `--version`
- 是否打印版本信息。
- 类型: bool (默认: 0).
* `--show_layer_stat`
- 是否显示**每个批次数据**中每层的数值统计.
- 类型: bool (默认: 0).
## 训练
* `--log_period`
- 每log_period个批次打印日志进度.
- 类型: int32 (默认: 100).
* `--dot_period`
- 每dot_period个批次输出符号'.'.
- 类型: int32 (默认: 1).
* `--saving_period`
- 每saving_period轮保存训练参数.
- 类型: int32 (默认: 1).
* `--save_dir`
- 保存模型参数的目录，需要明确指定，但不需要提前创建。
- 类型: string (默认: null).
* `--start_pass`
- 从start_pass轮开始训练，会加载上一轮的参数。
- 类型: int32 (默认: 0).
* `--show_parameter_stats_period`
- 在训练过程中每show_parameter_stats_period个批次输出参数统计。默认不显示。
- 类型: int32 (默认: 0).
* `--save_only_one`
- 只保存最后一轮的参数，而之前的参数将会被删除。
- 类型: bool (默认: 0).
* `--load_missing_parameter_strategy`
- 当模型参数不存在时，指定加载的方式。目前支持fail/rand/zero三种操作.
- `fail`: 程序直接退出.
- `rand`: 根据网络配置中的**initial\_strategy**采用均匀分布或者高斯分布初始化。均匀分布的范围是: **[mean - std, mean + std]**, 其中mean和std是训练配置中的参数.
- `zero`: 所有参数置为零.
- 类型: string (默认: fail).
* `--init_model_path`
- 初始化模型的路径。如果设置该参数，start\_pass将不起作用。同样也可以在测试模式中指定模型路径。
- 类型: string (默认: null).
* `--saving_period_by_batches`
- 在一轮中每saving_period_by_batches个批次保存一次参数。
- 类型: int32 (默认: 0).
* `--log_error_clipping`
- 当在网络层配置中设置**error_clipping_threshold**时，该参数指示是否打印错误截断日志。如果为true，**每批次**的反向传播将会打印日志信息。该截断会影响**输出的梯度**.
- 类型: bool (默认: 0).
* `--log_clipping`
- 当在训练配置中设置**gradient_clipping_threshold**时，该参数指示是否打印日志截断信息。该截断会影响**权重更新的梯度**.
- 类型: bool (默认: 0).
* `--use_old_updater`
- 是否使用旧的RemoteParameterUpdater。 默认使用ConcurrentRemoteParameterUpdater，主要为开发者使用，使用者通常无需关心.
- 类型: bool (默认: 0).
* `--enable_grad_share`
- 启用梯度参数的阈值，在多CPU训练时共享该参数.
- 类型: int32 (默认: 100 \* 1024 \* 1024).
* `--grad_share_block_num`
- 梯度参数的分块数目，在多CPU训练时共享该参数.
- 类型: int32 (默认: 64).
## 测试
* `--test_pass`
- 加载test_pass轮的模型用于测试.
- 类型: int32 (默认: -1).
* `--test_period`
- 如果为0，每轮结束时对所有测试数据进行测试；如果不为0，每test_period个批次对所有测试数据进行测试.
- 类型: int32 (默认: 0).
* `--test_wait`
- 指示当指定轮的测试模型不存在时，是否需要等待该轮模型参数。如果在训练期间同时发起另外一个进程进行测试，可以使用该参数.
- 类型: bool (默认: 0).
* `--model_list`
- 测试时指定的存储模型列表的文件.
- 类型: string (默认: "", null).
* `--predict_output_dir`
- 保存网络层输出结果的目录。该参数在网络配置的Outputs()中指定，默认为null，意思是不保存结果。在测试阶段，如果你想要保存某些层的特征图，请指定该目录。需要注意的是，网络层的输出是经过激活函数之后的值.
- 类型: string (默认: "", null).
* `--average_test_period`
- 使用`average_test_period`个批次的参数平均值进行测试。该参数必须能被FLAGS_log_period整除，默认为0，意思是不使用平均参数执行测试.
- 类型: int32 (默认: 0).
* `--distribute_test`
- 在分布式环境中测试，将多台机器的测试结果合并.
- 类型: bool (默认: 0).
* `--predict_file`
- 保存预测结果的文件名。该参数默认为null，意思是不保存结果。目前该参数仅用于AucValidationLayer和PnpairValidationLayer层，每轮都会保存预测结果.
- 类型: string (默认: "", null).
## GPU
* `--gpu_id`
- 指示使用哪个GPU核.
- 类型: int32 (默认: 0).
* `--allow_only_one_model_on_one_gpu`
- 如果为true，一个GPU设备上不允许配置多个模型.
- 类型: bool (默认: 1).
* `--parallel_nn`
- 指示是否使用多线程来计算一个神经网络。如果为false，设置gpu_id指定使用哪个GPU核（训练配置中的设备属性将会无效）。如果为true，GPU核在训练配置中指定（gpu_id无效）.
- 类型: bool (默认: 0).
* `--cudnn_dir`
- 选择路径来动态加载NVIDIA CuDNN库，例如，/usr/local/cuda/lib64. [默认]: LD_LIBRARY_PATH
- 类型: string (默认: "", null)
* `--cuda_dir`
- 选择路径来动态加载NVIDIA CUDA库，例如，/usr/local/cuda/lib64. [默认]: LD_LIBRARY_PATH
- 类型: string (默认: "", null)
* `--cudnn_conv_workspace_limit_in_mb`
- 指定cuDNN的最大工作空间容限，单位是MB，默认为4096MB=4GB.
- 类型: int32 (默认: 4096MB=4GB)
## 自然语言处理(NLP): RNN/LSTM/GRU
* `--rnn_use_batch`
- 指示在简单的RecurrentLayer层的计算中是否使用批处理方法.
- 类型: bool (默认: 0).
* `--prev_batch_state`
- 标识是否为连续的batch计算.
- 类型: bool (默认: 0).
* `--beam_size`
- 集束搜索使用广度优先搜索的方式构建查找树。在树的每一层上，都会产生当前层状态的所有继承结果，按启发式损失的大小递增排序。然而，每层上只能保存固定数目个最好的状态，该数目是提前定义好的，称之为集束大小.
- 类型: int32 (默认: 1).
* `--diy_beam_search_prob_so`
 - 用户可以自定义beam search的方法，编译成动态库，供PaddlePaddle加载。 该参数用于指定动态库路径.
- 类型: string (默认: "", null).
## 度量学习(Metric Learning)
* `--external`
- 指示是否使用外部机器进行度量学习.
- 类型: bool (默认: 0).
* `--data_server_port`
- 数据服务器(data server)的监听端口，主要用在度量学习中.
- 类型: int32 (默认: 21134).
## 数据支持(DataProvider)
* `--memory_threshold_on_load_data`
- 内存容限阈值，当超过该阈值时，停止加载数据.
- 类型: double (默认: 1.0).
## 单元测试
* `--checkgrad_eps`
- 使用checkgrad模式时的参数变化大小.
- 类型: double (默认: 1e-05).
## 参数服务器和分布式通信
* `--start_pserver`
- 指示是否开启参数服务器(parameter server).
- 类型: bool (默认: 0).
* `--pservers`
- 参数服务器的IP地址，以逗号间隔.
- 类型: string (默认: "127.0.0.1").
* `--port`
- 参数服务器的监听端口.
- 类型: int32 (默认: 20134).
* `--ports_num`
- 发送参数的端口号，根据默认端口号递增.
- 类型: int32 (默认: 1).
* `--trainer_id`
 - 在分布式训练中，每个训练节点必须指定一个唯一的id号，从0到num_trainers-1。0号训练节点是主训练节点。使用者无需关心这个参数.
- 类型: int32 (默认: 0).
* `--num_gradient_servers`
- 梯度服务器的数量，该参数在集群提交环境中自动设置.
- 类型: int32 (默认: 1).
* `--small_messages`
- 如果消息数据太小，建议将该参数设为true，启动快速应答，无延迟.
- 类型: bool (默认: 0).
* `--sock_send_buf_size`
- 限制套接字发送缓冲区的大小。如果仔细设置的话，可以有效减小网络的阻塞.
- 类型: int32 (默认: 1024 \* 1024 \* 40).
* `--sock_recv_buf_size`
- 限制套接字接收缓冲区的大小.
- 类型: int32 (默认: 1024 \* 1024 \* 40).
* `--parameter_block_size`
- 参数服务器的参数分块大小。如果未设置，将会自动计算出一个合适的值.
- 类型: int32 (默认: 0).
* `--parameter_block_size_for_sparse`
- 参数服务器稀疏更新的参数分块大小。如果未设置，将会自动计算出一个合适的值.
- 类型: int32 (默认: 0).
* `--log_period_server`
- 在参数服务器终端每log_period_server个批次打印日志进度.
- 类型: int32 (默认: 500).
* `--loadsave_parameters_in_pserver`
- 在参数服务器上加载和保存参数，只有当设置了sparse_remote_update参数时才有效.
- 类型: bool (默认: 0).
* `--pserver_num_threads`
- 同步执行操作的线程数.
- 类型: bool (默认: 1).
* `--ports_num_for_sparse`
- 发送参数的端口号，根据默认值递增(port + ports_num)，用于稀疏训练中.
- 类型: int32 (默认: 0).
* `--nics`
- 参数服务器的网络设备名称，已经在集群提交环境中完成设置.
- 类型: string (默认: "xgbe0,xgbe1").
* `--rdma_tcp`
- 使用rdma还是tcp传输协议，该参数已经在集群提交环境中完成设置.
- 类型: string (默认: "tcp").
## 异步随机梯度下降(Async SGD)
* `--async_count`
- 定义异步训练的长度，如果为0，则使用同步训练.
- 类型: int32 (默认: 0).
* `--async_lagged_ratio_min`
- 控制`config_.async_lagged_grad_discard_ratio()`的最小值.
- 类型: double (默认: 1.0).
* `--async_lagged_ratio_default`
- 如果在网络配置中未设置async_lagged_grad_discard_ratio，则使用该参数作为默认值.
- 类型: double (默认: 1.5).
## 性能调优(Performance Tuning)
* `--log_barrier_abstract`
- 如果为true，则显示阻隔性能的摘要信息.
- 类型: bool (默认: 1).
* `--log_barrier_show_log`
- 如果为true，则总会显示阻隔摘要信息，即使间隔很小.
- 类型: bool (默认: 0).
* `--log_barrier_lowest_nodes`
- 最少显示多少个节点.
- 类型: int32 (默认: 5).
* `--check_sparse_distribution_in_pserver`
- 指示是否检查所有参数服务器上的稀疏参数的分布是均匀的.
- 类型: bool (默认: 0).
* `--show_check_sparse_distribution_log`
- 指示是否显示参数服务器上的稀疏参数分布的日志细节.
- 类型: bool (默认: 0).
* `--allow_inefficient_sparse_update`
- 指示是否允许低效率的稀疏更新.
- 类型: bool (默认: 0).
* `--check_sparse_distribution_batches`
- 每运行多少个批次执行一次稀疏参数分布的检查.
- 类型: int32 (默认: 100).
* `--check_sparse_distribution_ratio`
- 如果检查到分配在不同参数服务器上的参数的分布不均匀次数大于check_sparse_distribution_ratio * check_sparse_distribution_batches次，程序停止.
- 类型: double (默认: 0.6).
* `--check_sparse_distribution_unbalance_degree`
- 不同参数服务器上数据大小的最大值与最小值的比率.
- 类型: double (默认: 2).
## 矩阵/向量/随机数
* `--enable_parallel_vector`
- 启动并行向量的阈值.
- 类型: int32 (默认: 0).
* `--seed`
- 随机数的种子。srand(time)的为0.
- 类型: int32 (默认: 1)
* `--thread_local_rand_use_global_seed`
- 是否将全局种子应用于本地线程的随机数.
- 类型: bool (默认: 0).
doc/howto/usage/cmd_parameter/detail_introduction_en.md
浏览文件 @ b23d99dd
......@@ -73,7 +73,7 @@
- type: bool (default: 0).
* `--load_missing_parameter_strategy`
- Specify the loading operation when model file is missing. Now support fail/rand/zere three operations.
- Specify the loading operation when model file is missing. Now support fail/rand/zero three operations.
- `fail`: program will exit.
- `rand`: uniform or normal distribution according to **initial\_strategy** in network config. Uniform range is: **[mean - std, mean + std]**, where mean and std are configures in trainer config.
- `zero`: all parameters are zero.
......@@ -118,11 +118,11 @@
- type: int32 (default: 0).
* `--test_wait`
- Whether to wait for parameter per pass if not exist. If set test_data_path in submitting environment of cluster, it will launch one process to perfom testing, so we need to set test_wait=1. Note that in the cluster submitting environment, this argument has been set True by default.
 - Whether to wait for parameter per pass if not exist. It can be used when user launch another process to perfom testing during the training process.
- type: bool (default: 0).
* `--model_list`
- File that saves the model list when testing. It was set automatically when using cluster submitting environment after setting model_path.
- File that saves the model list when testing.
- type: string (default: "", null).
* `--predict_output_dir`
......@@ -212,7 +212,7 @@
- type: bool (default: 0).
* `--pservers`
- Comma separated IP addresses of pservers. It is set automatically in cluster submitting environment.
- Comma separated IP addresses of pservers.
- type: string (default: "127.0.0.1").
* `--port`
......
doc/howto/usage/cmd_parameter/index_cn.rst 0 → 100644
浏览文件 @ b23d99dd
.. _cmd_line_index:
设置命令行参数
===============
.. toctree::
:maxdepth: 1
use_case_cn.md
arguments_cn.md
detail_introduction_cn.md
doc/howto/usage/cmd_parameter/index_en.md 已删除 100644 → 0
浏览文件 @ 343d9973
```eval_rst
.. _cmd_line_index:
```
# Set Command-line Parameters
* [Use Case](use_case_en.md)
* [Arguments](arguments_en.md)
* [Detailed Descriptions](detail_introduction_en.md)
doc/howto/usage/cmd_parameter/index_en.rst 0 → 100644
浏览文件 @ b23d99dd
.. _cmd_line_index:
Set Command-line Parameters
===========================
.. toctree::
:maxdepth: 1
use_case_en.md
arguments_en.md
detail_introduction_en.md
doc/howto/usage/cmd_parameter/use_case_cn.md 0 → 100644
浏览文件 @ b23d99dd
# 使用案例
## 本地训练
本地训练的实验，诸如图像分类，自然语言处理等，通常都会使用下面这些命令行参数。
```
paddle train \
--use_gpu=1/0 \ #1:GPU,0:CPU(默认为1)
--config=network_config \
--save_dir=output \
--trainer_count=COUNT \ #(默认为1)
--test_period=M \ #(默认为0)
--num_passes=N \ #(默认为100)
--log_period=K \ #(默认为100)
--dot_period=1000 \ #(默认为1)
#[--show_parameter_stats_period=100] \ #(默认为0)
#[--saving_period_by_batches=200] \ #(默认为0)
```
根据你的任务，可以选择是否使用参数`show_parameter_stats_period`和`saving_period_by_batches`。
### 1) 将命令参数传给网络配置
`config_args`是一个很有用的参数，用于将参数传递给网络配置。
```
--config_args=generating=1,beam_size=5,layer_num=10 \
```
`get_config_arg`可用于在网络配置中解析这些参数，如下所示：
```
generating = get_config_arg('generating', bool, False)
beam_size = get_config_arg('beam_size', int, 3)
layer_num = get_config_arg('layer_num', int, 8)
```
`get_config_arg`:
```
get_config_arg(name, type, default_value)
```
- name: `--config_args`中指定的名字
- type: 值类型，包括bool, int, str, float等
- default_value: 默认值
### 2) 使用模型初始化网络
增加如下参数：
```
--init_model_path=model_path
--load_missing_parameter_strategy=rand
```
## 本地测试
方法一：
```
paddle train --job=test \
--use_gpu=1/0 \
--config=network_config \
--trainer_count=COUNT \
--init_model_path=model_path \
```
- 使用init\_model\_path指定测试的模型
- 只能测试单个模型
方法二：
```
paddle train --job=test \
--use_gpu=1/0 \
--config=network_config \
--trainer_count=COUNT \
--model_list=model.list \
```
- 使用model_list指定测试的模型列表
- 可以测试多个模型，文件model.list如下所示：
```
./alexnet_pass1
./alexnet_pass2
```
方法三：
```
paddle train --job=test \
--use_gpu=1/0 \
--config=network_config \
--trainer_count=COUNT \
--save_dir=model \
--test_pass=M \
--num_passes=N \
```
这种方式必须使用Paddle存储的模型路径格式，如：`model/pass-%5d`。测试的模型包括从第M轮到第N-1轮存储的所有模型。例如，M=12，N=14这种写法将会测试模型`model/pass-00012`和`model/pass-00013`。
## 稀疏训练
当输入是维度很高的稀疏数据时，通常使用稀疏训练来加速计算过程。例如，输入数据的字典维数是1百万，但是每个样本仅包含几个词。在Paddle中，稀疏矩阵的乘积应用于前向传播过程，而稀疏更新在反向传播之后的权重更新时进行。
### 1) 本地训练
用户需要在网络配置中指定**sparse\_update=True**。请参照网络配置的文档了解更详细的信息。
### 2) 集群训练
在集群上训练一个稀疏模型需要加上下面的参数。同时用户需要在网络配置中指定**sparse\_remote\_update=True**。请参照网络配置的文档了解更详细的信息。
```
--ports_num_for_sparse=1 #(默认为0)
```
## parallel_nn
用户可以设置`parallel_nn`来混合使用GPU和CPU计算网络层的参数。也就是说，你可以将网络配置成某些层使用GPU计算，而其他层使用CPU计算。另一种方式是将网络层划分到不同的GPU上去计算，这样可以减小GPU内存，或者采用并行计算来加速某些层的更新。
如果你想使用这些特性，你需要在网络配置中指定设备的ID号(表示为deviceId)，并且加上下面的命令行参数:
```
--parallel_nn=true
```
### 案例一：GPU和CPU混合使用
请看下面的例子：
```
#command line:
paddle train --use_gpu=true --parallel_nn=true trainer_count=COUNT
default_device(0)
fc1=fc_layer(...)
fc2=fc_layer(...)
fc3=fc_layer(...,layer_attr=ExtraAttr(device=-1))
```
- default_device(0): 设置默认设备号为0。这意味着除了指定device=-1的层之外，其他所有层都会使用GPU计算，每层使用的GPU号依赖于参数trainer\_count和gpu\_id(默认为0)。在此，fc1和fc2层在GPU上计算。
- device=-1: fc3层使用CPU计算。
- trainer_count:
- trainer_count=1: 如果未设置gpu\_id，那么fc1和fc2层将会使用第1个GPU来计算。否则使用gpu\_id指定的GPU。
- trainer_count>1: 在trainer\_count个GPU上使用数据并行来计算某一层。例如，trainer\_count=2意味着0号和1号GPU将会使用数据并行来计算fc1和fc2层。
### 案例二：在不同设备上指定层
```
#command line:
paddle train --use_gpu=true --parallel_nn=true --trainer_count=COUNT
#network:
fc2=fc_layer(input=l1, layer_attr=ExtraAttr(device=0), ...)
fc3=fc_layer(input=l1, layer_attr=ExtraAttr(device=1), ...)
fc4=fc_layer(input=fc2, layer_attr=ExtraAttr(device=-1), ...)
```
在本例中，我们假设一台机器上有4个GPU。
- trainer_count=1:
- 使用0号GPU计算fc2层。
- 使用1号GPU计算fc3层。
- 使用CPU计算fc4层。
- trainer_count=2:
- 使用0号和1号GPU计算fc2层。
- 使用2号和3号GPU计算fc3层。
- 使用CPU两线程计算fc4层。
- trainer_count=4:
- 运行失败（注意到我们已经假设机器上有4个GPU），因为参数`allow_only_one_model_on_one_gpu`默认设置为真。
**当`device!=-1`时设备ID号的分配：**
```
(deviceId + gpu_id + threadId * numLogicalDevices_) % numDevices_
deviceId: 在层中指定
gpu_id: 默认为0
threadId: 线程ID号，范围: 0,1,..., trainer_count-1
numDevices_: 机器的设备(GPU)数目
numLogicalDevices_: min(max(deviceId + 1), numDevices_)
```
doc/howto/usage/cmd_parameter/use_case_en.md
浏览文件 @ b23d99dd
......@@ -134,14 +134,14 @@ fc2=fc_layer(...)
fc3=fc_layer(...,layer_attr=ExtraAttr(device=-1))
```
- default_device(0): set default device ID to 0. This means that except the layers with device=-1, all layers will use a GPU, and the specific GPU used for each layer depends on trainer\_count and gpu\_id (0 by default). Here, layer l1 and l2 are computed on the GPU.
- default_device(0): set default device ID to 0. This means that except the layers with device=-1, all layers will use a GPU, and the specific GPU used for each layer depends on trainer\_count and gpu\_id (0 by default). Here, layer fc1 and fc2 are computed on the GPU.
- device=-1: use the CPU for layer l3.
- device=-1: use the CPU for layer fc3.
- trainer_count:
- trainer_count=1: if gpu\_id is not set, then use the first GPU to compute layers l1 and l2. Otherwise use the GPU with gpu\_id.
- trainer_count=1: if gpu\_id is not set, then use the first GPU to compute layers fc1 and fc2. Otherwise use the GPU with gpu\_id.
- trainer_count>1: use trainer\_count GPUs to compute one layer using data parallelism. For example, trainer\_count=2 means that GPUs 0 and 1 will use data parallelism to compute layer l1 and l2.
- trainer_count>1: use trainer\_count GPUs to compute one layer using data parallelism. For example, trainer\_count=2 means that GPUs 0 and 1 will use data parallelism to compute layer fc1 and fc2.
### Case 2: Specify Layers in Different Devices
......@@ -157,14 +157,14 @@ fc4=fc_layer(input=fc2, layer_attr=ExtraAttr(device=-1), ...)
In this case, we assume that there are 4 GPUs in one machine.
- trainer_count=1:
- Use GPU 0 to compute layer l2.
- Use GPU 1 to compute layer l3.
- Use CPU to compute layer l4.
- Use GPU 0 to compute layer fc2.
- Use GPU 1 to compute layer fc3.
- Use CPU to compute layer fc4.
- trainer_count=2:
- Use GPU 0 and 1 to compute layer l2.
- Use GPU 2 and 3 to compute layer l3.
- Use CPU to compute l4 in two threads.
- Use GPU 0 and 1 to compute layer fc2.
- Use GPU 2 and 3 to compute layer fc3.
- Use CPU to compute fc4 in two threads.
- trainer_count=4:
- It will fail (note, we have assumed that there are 4 GPUs in machine), because argument `allow_only_one_model_on_one_gpu` is true by default.
......
doc/howto/usage/k8s/k8s_aws_en.md 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
doc/howto/usage/cluster/k8s/k8s_cn.md → doc/howto/usage/k8s/k8s_cn.md
浏览文件 @ b23d99dd
# Kubernetes 单机训练
# Kubernetes单机训练
在这篇文档里，我们介绍如何在 Kubernetes 集群上启动一个单机使用CPU的Paddle训练作业。在下一篇中，我们将介绍如何启动分布式训练作业。
......
doc/howto/usage/cluster/k8s/k8s_distributed_cn.md → doc/howto/usage/k8s/k8s_distributed_cn.md
浏览文件 @ b23d99dd
# Kubernetes 分布式训练
# Kubernetes分布式训练
前一篇文章介绍了如何在Kubernetes集群上启动一个单机PaddlePaddle训练作业 (Job)。在这篇文章里，我们介绍如何在Kubernetes集群上进行分布式PaddlePaddle训练作业。关于PaddlePaddle的分布式训练，文章 [Cluster Training](https://github.com/baidu/Paddle/blob/develop/doc/cluster/opensource/cluster_train.md)介绍了一种通过SSH远程分发任务，进行分布式训练的方法，与此不同的是，本文将介绍在Kubernetes容器管理平台上快速构建PaddlePaddle容器集群，进行分布式训练的方案。
......@@ -22,7 +22,7 @@
首先，我们需要拥有一个Kubernetes集群，在这个集群中所有node与pod都可以互相通信。关于Kubernetes集群搭建，可以参考[官方文档](http://kubernetes.io/docs/getting-started-guides/kubeadm/)，在以后的文章中我们也会介绍AWS上搭建的方案。本文假设大家能找到几台物理机，并且可以按照官方文档在上面部署Kubernetes。在本文的环境中，Kubernetes集群中所有node都挂载了一个[MFS](http://moosefs.org/)（Moose filesystem，一种分布式文件系统）共享目录，我们通过这个目录来存放训练文件与最终输出的模型。关于MFS的安装部署，可以参考[MooseFS documentation](https://moosefs.com/documentation.html)。在训练之前，用户将配置与训练数据切分好放在MFS目录中，训练时，程序从此目录拷贝文件到容器内进行训练，将结果保存到此目录里。整体的结构图如下：
![paddle on kubernetes结构图](k8s-paddle-arch.png)
![paddle on kubernetes结构图](src/k8s-paddle-arch.png)
上图描述了一个3节点的分布式训练场景，Kubernetes集群的每个node上都挂载了一个MFS目录，这个目录可以通过volume的形式挂载到容器中。Kubernetes为这次训练创建了3个pod并且调度到了3个node上运行，每个pod包含一个PaddlePaddle容器。在容器创建后，会启动pserver与trainer进程，读取volume中的数据进行这次分布式训练。
......
doc/howto/usage/k8s/k8s_en.md 0 → 100644
浏览文件 @ b23d99dd
# Paddle On Kubernetes
>In this article, we will introduce how to run Paddle training job on single CPU machine using Kubernetes. In next article, we will introduce how to run Paddle training job on distributed cluster.
## Build Docker Image
In distributed Kubernetes cluster, we will use Ceph or other shared storage system for storing training related data so that all processes in Paddle training can retrieve data from Ceph. In this example, we will only demo training job on single machine. In order to simplify the requirement of the environment, we will directly put training data into Paddle's Docker Image, so we need to create a Paddle Docker image that already includes the training data.
Paddle's [Quick Start Tutorial](http://www.paddlepaddle.org/doc/demo/quick_start/index_en.html) introduces how to download and train data by using script from Paddle's source code.
And `paddledev/paddle:cpu-demo-latest` image has the Paddle source code and demo. (Caution: Default Paddle image `paddledev/paddle:cpu-latest` doesn't include the source code, Paddle's different versions of image can be referred here: [Docker installation guide](http://www.paddlepaddle.org/doc/build/docker_install.html)), so we run this container and download the training data, and then commit the whole container to be a new Docker image.
### Run Docker Container
```
$ docker run --name quick_start_data -it paddledev/paddle:cpu-demo-latest
```
### Download Training Data
Getting into `/root/paddle/demo/quick_start/data` Directory，using `get_data.sh` to download training data.
Then getting into `/root/paddle/demo/quick_start` Directory, using `preprocess.sh` to pre-process training data.
```
$ root@fbd1f2bb71f4:~/paddle/demo/quick_start/data# ./get_data.sh
Downloading Amazon Electronics reviews data...
--2016-10-31 01:33:43-- http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz
Resolving snap.stanford.edu (snap.stanford.edu)... 171.64.75.80
Connecting to snap.stanford.edu (snap.stanford.edu)|171.64.75.80|:80... connected.
HTTP request sent, awaiting response... 200 OK
Length: 495854086 (473M) [application/x-gzip]
Saving to: 'reviews_Electronics_5.json.gz'
10% [=======> ] 874,279 64.7KB/s eta 2h 13m
```
### Modify Startup Script
After downloading the data，modify `/root/paddle/demo/quick_start/train.sh` file contents are as follows (one more cd cmd):
```
set -e
cd /root/paddle/demo/quick_start
cfg=trainer_config.lr.py
#cfg=trainer_config.emb.py
#cfg=trainer_config.cnn.py
#cfg=trainer_config.lstm.py
#cfg=trainer_config.bidi-lstm.py
#cfg=trainer_config.db-lstm.py
paddle train \
--config=$cfg \
--save_dir=./output \
--trainer_count=4 \
--log_period=20 \
--num_passes=15 \
--use_gpu=false \
--show_parameter_stats_period=100 \
--test_all_data_in_one_period=1 \
2>&1 | tee 'train.log'
```
### Commit Docker Image
```
$ docker commit quick_start_data mypaddle/paddle:quickstart
```
## Use Kubernetes For Training
>We will use Kubernetes job for training process, following steps shows how to do the training with Kubernetes.
### Create Yaml Files
The output result in container will be demolished when job finished (container stopped running), so we need to mount the volume out to the local disk when creating the container to store the training result. Using our previously created image, we can create a [Kubernetes Job](http://kubernetes.io/docs/user-guide/jobs/#what-is-a-job), the yaml contents are as follows:
```
apiVersion: batch/v1
kind: Job
metadata:
name: quickstart
spec:
parallelism: 1
completions: 1
template:
metadata:
name: quickstart
spec:
volumes:
- name: output
hostPath:
path: /home/work/paddle_output
containers:
- name: pi
image: mypaddle/paddle:quickstart
command: ["bin/bash", "-c", "/root/paddle/demo/quick_start/train.sh"]
volumeMounts:
- name: output
mountPath: /root/paddle/demo/quick_start/output
restartPolicy: Never
```
### Start Paddle Job
Using the above yaml file to start the Kubernetes job.
```
$ kubectl create -f paddle.yaml
```
Get the detailed status of the job:
```
$ kubectl get job
NAME DESIRED SUCCESSFUL AGE
quickstart 1 0 58s
$ kubectl describe job quickstart
Name: quickstart
Namespace: default
Image(s): registry.baidu.com/public/paddle:cpu-demo-latest
Selector: controller-uid=f120da72-9f18-11e6-b363-448a5b355b84
Parallelism: 1
Completions: 1
Start Time: Mon, 31 Oct 2016 11:20:16 +0800
Labels: controller-uid=f120da72-9f18-11e6-b363-448a5b355b84,job-name=quickstart
Pods Statuses: 0 Running / 1 Succeeded / 0 Failed
Volumes:
output:
Type: HostPath (bare host directory volume)
Path: /home/work/paddle_output
Events:
FirstSeen LastSeen Count From SubobjectPath Type Reason Message
--------- -------- ----- ---- ------------- -------- ------ -------
1m 1m 1 {job-controller } Normal SuccessfulCreate Created pod: quickstart-fa0wx
```
### Get Training Result
We can use kubectl command to take a look at the status of related pod.
```
$ kubectl describe pod quickstart-fa0wx
Name: quickstart-fa0wx
Namespace: default
Node: paddle-demo-let02/10.206.202.44
Start Time: Mon, 31 Oct 2016 11:20:17 +0800
Labels: controller-uid=f120da72-9f18-11e6-b363-448a5b355b84,job-name=quickstart
Status: Succeeded
IP: 10.0.0.9
Controllers: Job/quickstart
Containers:
quickstart:
Container ID: docker://b8561f5c79193550d64fa47418a9e67ebdd71546186e840f88de5026b8097465
Image: registry.baidu.com/public/paddle:cpu-demo-latest
Image ID: docker://18e457ce3d362ff5f3febf8e7f85ffec852f70f3b629add10aed84f930a68750
Port:
Command:
bin/bash
-c
/root/paddle/demo/quick_start/train.sh
QoS Tier:
cpu: BestEffort
memory: BestEffort
State: Terminated
Reason: Completed
Exit Code: 0
Started: Mon, 31 Oct 2016 11:20:20 +0800
Finished: Mon, 31 Oct 2016 11:21:46 +0800
Ready: False
Restart Count: 0
Environment Variables:
Conditions:
Type Status
Ready False
Volumes:
output:
Type: HostPath (bare host directory volume)
Path: /home/work/paddle_output
```
We can also ssh to Kubernetes node to take a look at the training result.
```
[root@paddle-demo-let02 paddle_output]# ll
total 60
drwxr-xr-x 2 root root 4096 Oct 31 11:20 pass-00000
drwxr-xr-x 2 root root 4096 Oct 31 11:20 pass-00001
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00002
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00003
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00004
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00005
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00006
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00007
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00008
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00009
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00010
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00011
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00012
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00013
drwxr-xr-x 2 root root 4096 Oct 31 11:21 pass-00014
```
doc/tutorials/embedding_model/index_cn.md 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
doc/tutorials/gan/gan.png
查看替换文件 @ 343d9973 浏览文件 @ b23d99dd
doc/tutorials/gan/gan.png

32.5 KB | W: | H:

doc/tutorials/gan/gan.png

17.4 KB | W: | H:

doc/tutorials/gan/gan.png
doc/tutorials/gan/gan.png
doc/tutorials/gan/gan.png
doc/tutorials/gan/gan.png
  • 2-up
  • Swipe
  • Onion skin
doc/tutorials/gan/index_en.md
浏览文件 @ b23d99dd
......@@ -4,9 +4,7 @@ This demo implements GAN training described in the original [GAN paper](https://
The high-level structure of GAN is shown in Figure. 1 below. It is composed of two major parts: a generator and a discriminator, both of which are based on neural networks. The generator takes in some kind of noise with a known distribution and transforms it into an image. The discriminator takes in an image and determines whether it is artificially generated by the generator or a real image. So the generator and the discriminator are in a competitive game in which generator is trying to generate image to look as real as possible to fool the discriminator, while the discriminator is trying to distinguish between real and fake images.
<p align="center">
<img src="./gan.png" width="500" height="300">
</p>
<center>![](./gan.png)</center>
<p align="center">
Figure 1. GAN-Model-Structure
<a href="https://ishmaelbelghazi.github.io/ALI/">figure credit</a>
......@@ -111,9 +109,7 @@ $python gan_trainer.py -d uniform --useGpu 1
```
The generated samples can be found in ./uniform_samples/ and one example is shown below as Figure 2. One can see that it roughly recovers the 2D uniform distribution.
<p align="center">
<img src="./uniform_sample.png" width="300" height="300">
</p>
<center>![](./uniform_sample.png)</center>
<p align="center">
Figure 2. Uniform Sample
</p>
......@@ -135,9 +131,7 @@ To train the GAN model on mnist data, one can use the following command:
$python gan_trainer.py -d mnist --useGpu 1
```
The generated sample images can be found at ./mnist_samples/ and one example is shown below as Figure 3.
<p align="center">
<img src="./mnist_sample.png" width="300" height="300">
</p>
<center>![](./mnist_sample.png)</center>
<p align="center">
Figure 3. MNIST Sample
</p>
doc/tutorials/gan/uniform_sample.png
查看替换文件 @ 343d9973 浏览文件 @ b23d99dd
doc/tutorials/gan/uniform_sample.png

20.1 KB | W: | H:

doc/tutorials/gan/uniform_sample.png

24.3 KB | W: | H:

doc/tutorials/gan/uniform_sample.png
doc/tutorials/gan/uniform_sample.png
doc/tutorials/gan/uniform_sample.png
doc/tutorials/gan/uniform_sample.png
  • 2-up
  • Swipe
  • Onion skin
doc/tutorials/index_cn.md
浏览文件 @ b23d99dd
......@@ -2,6 +2,7 @@
* [快速入门](quick_start/index_cn.rst)
* [个性化推荐](rec/ml_regression_cn.rst)
* [图像分类](image_classification/index_cn.md)
* [情感分析](sentiment_analysis/index_cn.md)
* [语义角色标注](semantic_role_labeling/index_cn.md)
* [机器翻译](text_generation/index_cn.md)
......@@ -9,3 +10,4 @@
## 常用模型
* [ResNet模型](imagenet_model/resnet_model_cn.md)
* [词向量模型](embedding_model/index_cn.md)
doc/tutorials/index_en.md
浏览文件 @ b23d99dd
......@@ -7,6 +7,7 @@ There are several examples and demos here.
* [Sentiment Analysis](sentiment_analysis/index_en.md)
* [Semantic Role Labeling](semantic_role_labeling/index_en.md)
* [Text Generation](text_generation/index_en.md)
* [Image Auto-Generation](gan/index_en.md)
## Model Zoo
* [ImageNet: ResNet](imagenet_model/resnet_model_en.md)
......
paddle/CMakeLists.txt
浏览文件 @ b23d99dd
add_subdirectory(cuda)
add_subdirectory(function)
add_subdirectory(utils)
add_subdirectory(testing)
add_subdirectory(math)
add_subdirectory(parameter)
add_subdirectory(gserver)
......
paddle/api/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/DisableCopy.h → paddle/api/Evaluator.cpp
浏览文件 @ b23d99dd
......@@ -11,13 +11,19 @@ distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT 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 <sstream>
#include "PaddleAPI.h"
#include "PaddleAPIPrivate.h"
#pragma once
Evaluator::Evaluator() : m(new EvaluatorPrivate()) {}
Evaluator::~Evaluator() { delete m; }
/**
* Disable copy macro.
*/
#define DISABLE_COPY(CLASS_NAME) \
CLASS_NAME(CLASS_NAME &&) = delete; \
CLASS_NAME(const CLASS_NAME &other) = delete; \
CLASS_NAME &operator=(const CLASS_NAME &other) = delete
void Evaluator::start() { m->rawPtr->start(); }
void Evaluator::finish() { m->rawPtr->finish(); }
std::string Evaluator::toString() {
std::ostringstream sout;
m->rawPtr->printStats(sout);
return sout.str();
}
paddle/api/GradientMachine.cpp
浏览文件 @ b23d99dd
......@@ -64,6 +64,10 @@ GradientMachine* GradientMachine::createByModelConfig(
return GradientMachine::createFromPaddleModelPtr(confPtr, mode, types);
}
void GradientMachine::start() { m->machine->start(); }
void GradientMachine::finish() { m->machine->finish(); }
void GradientMachine::onPassEnd() { m->machine->onPassEnd(); }
void GradientMachine::prefetch(const Arguments& inArgs) {
......@@ -166,3 +170,13 @@ SequenceGenerator* GradientMachine::asSequenceGenerator(
r->setBeamSize(beam_size);
return r;
}
Evaluator* GradientMachine::makeEvaluator() {
auto ev = new Evaluator();
ev->m->rawPtr = m->machine->makeEvaluator();
return ev;
}
void GradientMachine::eval(Evaluator* evaluator) {
m->machine->eval(evaluator->m->rawPtr);
}
paddle/api/Paddle.swig
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/PaddleAPI.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/PaddleAPIPrivate.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/Parameter.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/ParameterUpdater.cpp 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/Util.cpp
浏览文件 @ b23d99dd
......@@ -15,12 +15,11 @@ limitations under the License. */
#include "PaddleAPI.h"
#include "paddle/parameter/Parameter.h"
#include "paddle/utils/Excepts.h"
#include "paddle/utils/Common.h"
#include "paddle/utils/Flags.h"
#include "paddle/utils/PythonUtil.h"
#include "paddle/utils/Util.h"
#include <fenv.h>
#include <algorithm>
#include <iostream>
#include <iterator>
......
paddle/api/Vector.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/paddle_api_config.py.in
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/paddle_ld_flags.py
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/test/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/api/test/run_tests.sh
浏览文件 @ b23d99dd
此差异已折叠。
paddle/cuda/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/cuda/include/hl_base.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/cuda/include/hl_sequence.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/cuda/src/hl_cuda_sequence.cu
浏览文件 @ b23d99dd
此差异已折叠。
paddle/cuda/src/hl_warpctc_wrap.cc
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/ContextProjectionOp.cpp 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/ContextProjectionOp.h 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/ContextProjectionOpGpu.cu 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/ContextProjectionOpTest.cpp 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/Function.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/Function.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/function/FunctionTest.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/gserver/layers/GruCompute.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/gserver/layers/LstmCompute.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/gserver/layers/Projection.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/gserver/tests/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/BaseMatrix.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/Matrix.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/Matrix.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/TensorExpression.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/Vector.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/tests/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/tests/test_Matrix.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/math/tests/test_Tensor.cu
浏览文件 @ b23d99dd
此差异已折叠。
paddle/parameter/Parameter.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/pserver/BaseClient.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/pserver/ParameterClient2.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/pserver/ParameterServer2.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/scripts/docker/Dockerfile
浏览文件 @ b23d99dd
此差异已折叠。
paddle/scripts/docker/build.sh
浏览文件 @ b23d99dd
此差异已折叠。
paddle/scripts/docker/entrypoint 0 → 100755
浏览文件 @ b23d99dd
此差异已折叠。
paddle/scripts/travis/before_install.linux.sh 已删除 100755 → 0
浏览文件 @ 343d9973
此差异已折叠。
paddle/scripts/travis/build_submodules.sh 已删除 100755 → 0
浏览文件 @ 343d9973
此差异已折叠。
paddle/setup.py.in
浏览文件 @ b23d99dd
此差异已折叠。
paddle/testing/CMakeLists.txt 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
paddle/trainer/Trainer.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/trainer/TrainerMain.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/trainer/tests/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/.gitignore
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/CpuId.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/Excepts.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/Locks.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/ThreadLocal.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/Util.cpp
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/Util.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/Version.h
浏览文件 @ b23d99dd
此差异已折叠。
paddle/utils/tests/test_Thread.cpp
浏览文件 @ b23d99dd
此差异已折叠。
proto/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
python/CMakeLists.txt
浏览文件 @ b23d99dd
此差异已折叠。
python/paddle/trainer_config_helpers/config_parser.py 0 → 100644
浏览文件 @ b23d99dd
此差异已折叠。
python/setup.py.in
浏览文件 @ b23d99dd
此差异已折叠。
warp-ctc @ bd535c8d
比较 bd535c8d...bd535c8d
此差异已折叠。
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