fix dash and space bug,

ISSUE=4586495

git-svn-id: https://svn.baidu.com/idl/trunk/paddle@1408 1ad973e4-5ce8-4261-8a94-b56d1f490c56

.clang-format

+# This file is used by clang-format to autoformat paddle source code
+#
+# The clang-format is part of llvm toolchain.
+# It need to install llvm and clang to format source code style.
+#
+# The basic usage is,
+#   clang-format -i -style=file PATH/TO/SOURCE/CODE
+#
+# The -style=file implicit use ".clang-format" file located in one of 
+# parent directory. 
+# The -i means inplace change.
+#
+# The document of clang-format is 
+#   http://clang.llvm.org/docs/ClangFormat.html
+#   http://clang.llvm.org/docs/ClangFormatStyleOptions.html
+#
+# TODO(yuyang18): Add python and other language code style
+---
+Language:        Cpp
+BasedOnStyle:  Google
+IndentWidth:     2
+TabWidth:        2
+ContinuationIndentWidth: 4
+AccessModifierOffset: -2  # The private/protected/public has no indent in class
+PointerAlignment: Left    # int* p/int& p, not int *p/int &p
+Standard:  Cpp11 
+AllowAllParametersOfDeclarationOnNextLine: true
+...
+

.gitignore

+*.DS_Store
+build/

CMakeLists.txt

+cmake_minimum_required(VERSION 2.8)
+
+project(paddle CXX C)
+set(PADDLE_MAJOR_VERSION 0)
+set(PADDLE_MINOR_VERSION 8)
+set(PADDLE_PATCH_VERSION 0b)
+set(PADDLE_VERSION ${PADDLE_MAJOR_VERSION}.${PADDLE_MINOR_VERSION}.${PADDLE_PATCH_VERSION})
+
+set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/cmake")
+set(PROJ_ROOT ${CMAKE_SOURCE_DIR})
+include(package)
+include(swig)
+find_package(CUDA QUIET)
+find_package(Protobuf REQUIRED)
+find_package(PythonLibs 2.7 REQUIRED)
+find_package(PythonInterp 2.7 REQUIRED)
+find_package(NumPy)
+find_package(Threads REQUIRED)
+find_package(Glog)
+find_package(Gflags QUIET)
+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 instructs" ON) # TODO(yuyang18): Check AVX is supported or not as default value
+option(WITH_PYTHON "Compile PaddlePaddle with python interpretor" ON)
+option(WITH_STYLE_CHECK "Style Check for PaddlePaddle" ${PYTHONINTERP_FOUND})
+option(WITH_RDMA "Compile PaddlePaddle with rdma support" OFF)
+option(WITH_GLOG "Compile PaddlePaddle use glog, otherwise use a log implement internally" ${LIBGLOG_FOUND})
+option(WITH_GFLAGS "Compile PaddlePaddle use gflags, otherwise use a flag implement internally" ${GFLAGS_FOUND})
+option(WITH_TIMER "Compile PaddlePaddle use timer" OFF)
+option(WITH_TESTING "Compile and run unittest for PaddlePaddle" ${GTEST_FOUND})
+option(WITH_DOC "Compile PaddlePaddle with documentation" OFF)
+option(WITH_DOC_CN "Compile PaddlePaddle with Chinese documentation" OFF)
+option(WITH_SWIG_PY "Compile PaddlePaddle with py PaddlePaddle predict api" ${SWIG_FOUND})
+if(NOT CMAKE_BUILD_TYPE)
+    set(CMAKE_BUILD_TYPE "RelWithDebInfo" CACHE STRING 
+        "Choose the type of build, options are: Debug Release RelWithDebInfo MinSizeRel"
+        FORCE)
+endif()
+
+include(enableCXX11)
+include(cpplint)
+include(ccache)
+include(util)
+include(flags)
+include(cudnn)
+include(FindPythonModule)
+include(check_packages)
+
+# add PaddlePaddle version
+if(DEFINED ENV{PADDLE_VERSION})
+    add_definitions(-DPADDLE_VERSION=\"$ENV{PADDLE_VERSION}\")
+else()
+    if(EXISTS ${PROJ_ROOT}/.svn/)
+        find_package(Subversion REQUIRED)
+        if(SUBVERSION_FOUND)
+            Subversion_WC_INFO(${PROJ_ROOT} Project)
+            add_definitions(-DPADDLE_VERSION=${Project_WC_REVISION})
+        endif()
+    endif()
+endif()
+
+
+if(NOT WITH_GPU)
+    add_definitions(-DPADDLE_ONLY_CPU)
+    add_definitions(-DHPPL_STUB_FUNC)
+    list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
+else()
+    # TODO(yuyang18): Change it to remove std=c++11 in cuda compile.
+    set(CUDA_PROPAGATE_HOST_FLAGS OFF)
+    if(NOT CUDNN_FOUND)
+        message(FATAL_ERROR "Paddle need cudnn to compile")
+    endif()
+
+    if(WITH_DSO)
+        set(CUDA_LIBRARIES "")
+        add_definitions(-DPADDLE_USE_DSO)
+    endif(WITH_DSO)
+
+    # Include cuda and cudnn
+    include_directories(${CUDNN_INCLUDE_DIR})
+    include_directories(${CUDA_TOOLKIT_INCLUDE})
+endif(NOT WITH_GPU)
+
+if(WITH_DOUBLE)
+    add_definitions(-DPADDLE_TYPE_DOUBLE -DHPPL_TYPE_DOUBLE)
+    set(ACCURACY double)
+else(WITH_DOUBLE)
+    set(ACCURACY float)
+endif(WITH_DOUBLE)
+
+if(NOT WITH_TIMER)
+    add_definitions(-DPADDLE_DISABLE_TIMER)
+endif(NOT WITH_TIMER)
+
+if(WITH_AVX)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mavx")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx")
+else(WITH_AVX)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -msse3")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3")
+endif(WITH_AVX)
+
+if(WITH_PYTHON)
+    include_directories(${PYTHON_INCLUDE_DIR})
+    include_directories(${PYTHON_NUMPY_INCLUDE_DIR})
+else(WITH_PYTHON)
+    add_definitions(-DPADDLE_NO_PYTHON)
+endif(WITH_PYTHON)
+
+if(NOT WITH_RDMA)
+    add_definitions(-DPADDLE_DISABLE_RDMA)
+endif()
+
+if(WITH_GLOG)
+    add_definitions(-DPADDLE_USE_GLOG)
+endif()
+
+if(WITH_GFLAGS)
+    add_definitions(-DPADDLE_USE_GFLAGS)
+    add_definitions(-DGFLAGS_NS=${GFLAGS_NAMESPACE})
+    include_directories(${GFLAGS_INCLUDE_DIRS})
+endif()
+
+if(WITH_TESTING)
+    enable_testing()
+    include_directories(${GTEST_INCLUDE_DIRS})
+endif()
+
+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()
+add_subdirectory(proto)
+add_subdirectory(paddle)
+add_subdirectory(python)
+if(WITH_DOC)
+    add_subdirectory(doc)
+endif()
+if(WITH_DOC_CN)
+    add_subdirectory(doc_cn)
+endif()

LICENSE

+Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright (c) 2016 Baidu, Inc. All Rights Reserve.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

+# PaddlePaddle
+
+[![Documentation Status](https://readthedocs.org/projects/ctcspeechrecognition/badge/?version=latest)](http://ctcspeechrecognition.readthedocs.io/en/latest/?badge=latest)
+
+PaddlePaddle (PArallel Distributed Deep LEarning) is an easy-to-use,
+efficient, flexible and scalable deep learning platform, which is originally
+developed by Baidu scientists and engineers for the purpose of applying deep
+learning to many products at Baidu.
+
+## Features
+
+- **Flexibility**
+
+   PaddlePaddle supports a wide range of neural network architectures and
+   optimization algorithms. It is easy to configure complex models such as
+   neural machine translation model with attention mechanism or complex memory
+   connection.
+
+-  **Efficiency**
+  
+  In order to unleash the power of heterogeneous computing resource,
+  optimization occurs at different levels of PaddlePaddle, including
+  computing, memory, architecture and communication. The following are some
+  examples:
+  1. Optimized math operations through SSE/AVX intrinsics, BLAS libraries
+  (e.g. MKL, ATLAS, cuBLAS) or customized CPU/GPU kernels. 
+  2. Highly optimized recurrent networks which can handle **variable-length** 
+  sequence without padding.
+  3. Optimized local and distributed training for models with high dimensional
+  sparse data.
+
+- **Scalability**
+
+  With PaddlePaddle, it is easy to use many CPUs/GPUs and machines to speed
+  up your training. PaddlePaddle can achieve high throughput and performance
+  via optimized communication.
+
+- **Connected to Products**
+
+  In addition, PaddlePaddle is also designed to be easily deployable. At Baidu,
+  PaddlePaddle has been deployed into products or service with a vast number
+  of users, including ad click-through rate (CTR) prediction, large-scale image
+  classification, optical character recognition(OCR), search ranking, computer
+  virus detection, recommendation, etc. It is widely utilized in products at
+  Baidu and it has achieved a significant impact. We hope you can also exploit
+  the capability of PaddlePaddle to make a huge impact for your product.
+
+## Installation
+See [installation guide]() to build and install from the source code or install
+the Docker Image.
+
+## Documentation
+- [Quick Start]() <br>
+   You can follow the quick start tutorial to learn how use PaddlePaddle
+   step-by-step.
+    
+- [Example and Demo]() <br>
+   We provide five demos, including: image classification, sentiment analysis,
+   sequence to sequence model, recommendation, semantic role labelling. 
+   
+- [Distributed Training]() <br>
+  This system supports training deep learning models on multiple machines
+  with data parallelism.
+   
+- [Python API]() <br>
+   PaddlePaddle supports using either Python interface or C++ to build your
+   system. We also use SWIG to wrap C++ source code to create a user friendly
+   interface for Python. You can also use SWIG to create interface for your
+   favorite programming language.
+ 
+- [How to Contribute]() <br>
+   We sincerely appreciate your interest and contributions. If you’d like to
+   contribute, please read the contribution guide.   
+
+- [Source Code Documents]() <br>
+      
+## Ask Questions
+
+If you want to ask questions and discuss about methods and models, welcome
+to send email to paddle-dev@baidu.com. Framework development discussions and
+bug reports are collected on [Issues](https://github.com/paddle/paddle/issues).
+
+## Copyright and License
+PaddlePaddle is provided under the [Apache-2.0 license](LICENSE).

authors

+Cao, Ying
+Cheng, Yujuan
+Dang, Qingqing
+Dong, Tengfei
+Du, Dalong
+Feng, Shouqiang
+Gao, Haoyuan
+Han, Baochang
+Han, Jinchen
+Hao, Nanyu
+He, Daoyuan
+He, Zhengyan
+Hou, Jue
+Huang, Chang
+Huang, Zhiheng
+Hu, Na
+Kong, Qi
+Liao, Gang
+Li, Bo
+Li, Jiajie
+Li, Jing
+Li, Lei
+Li, Peng
+Liu, Sheng
+Liu, Yuan
+Li, Yuze
+Luo, Heng
+Luo, Tao
+Lyu, Qin
+Mao, Hongyue
+Qian, Xiaojun
+Qi, Jun
+Qin, Duohao
+Shen, Guolong
+Shi, Guangchuan
+Song, Xiang
+Wang, Jiang
+Wang, Yanfei
+Wang, Yong
+Weng, Renliang
+Xu, Tianbing
+Xu, Wei
+Xu, Xingyu
+Yan, Chong
+Yan, Chunwei
+Yang, Yi
+Yu, Yang
+Yu, Yinan
+Zhang, Jian
+Zhang, Ruiqing
+Zhang, Weide
+Zhao, Liang
+Zhou, Jie

cmake/FindGlog.cmake

+#
+# Find libglog
+#
+#  LIBGLOG_INCLUDE_DIR - where to find glog/logging.h, etc.
+#  LIBGLOG_LIBRARY     - List of libraries when using libglog.
+#  LIBGLOG_FOUND       - True if libglog found.
+#
+# from https://github.com/facebook/hhvm/blob/master/CMake/FindGlog.cmake
+
+IF (LIBGLOG_INCLUDE_DIR)
+  # Already in cache, be silent
+  SET(LIBGLOG_FIND_QUIETLY TRUE)
+ENDIF ()
+
+FIND_PATH(LIBGLOG_INCLUDE_DIR glog/logging.h)
+
+FIND_LIBRARY(LIBGLOG_LIBRARY glog)
+
+# handle the QUIETLY and REQUIRED arguments and set LIBGLOG_FOUND to TRUE if
+# all listed variables are TRUE
+INCLUDE(FindPackageHandleStandardArgs)
+FIND_PACKAGE_HANDLE_STANDARD_ARGS(LIBGLOG DEFAULT_MSG LIBGLOG_LIBRARY LIBGLOG_INCLUDE_DIR)
+
+MARK_AS_ADVANCED(LIBGLOG_LIBRARY LIBGLOG_INCLUDE_DIR)
\ No newline at end of file

cmake/FindNumPy.cmake

+# Find the Python NumPy package
+# PYTHON_NUMPY_INCLUDE_DIR
+# NUMPY_FOUND
+# will be set by this script
+
+cmake_minimum_required(VERSION 2.6)
+
+if(NOT PYTHON_EXECUTABLE)
+  if(NumPy_FIND_QUIETLY)
+    find_package(PythonInterp QUIET)
+  else()
+    find_package(PythonInterp)
+    set(_numpy_out 1)
+  endif()
+endif()
+
+if (PYTHON_EXECUTABLE)
+  # write a python script that finds the numpy path
+  file(WRITE ${PROJECT_BINARY_DIR}/FindNumpyPath.py
+      "try: import numpy; print(numpy.get_include())\nexcept:pass\n")
+
+  # execute the find script
+  exec_program("${PYTHON_EXECUTABLE}" ${PROJECT_BINARY_DIR}
+    ARGS "FindNumpyPath.py"
+    OUTPUT_VARIABLE NUMPY_PATH)
+elseif(_numpy_out)
+  message(STATUS "Python executable not found.")
+endif(PYTHON_EXECUTABLE)
+
+find_path(PYTHON_NUMPY_INCLUDE_DIR numpy/arrayobject.h
+  HINTS "${NUMPY_PATH}" "${PYTHON_INCLUDE_PATH}")
+
+if(PYTHON_NUMPY_INCLUDE_DIR)
+  set(PYTHON_NUMPY_FOUND 1 CACHE INTERNAL "Python numpy found")
+endif(PYTHON_NUMPY_INCLUDE_DIR)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(NumPy DEFAULT_MSG PYTHON_NUMPY_INCLUDE_DIR)

cmake/FindPythonModule.cmake

+# Find if a Python module is installed
+# Found at http://www.cmake.org/pipermail/cmake/2011-January/041666.html
+# To use do: find_python_module(PyQt4 REQUIRED)
+function(find_python_module module)
+    string(TOUPPER ${module} module_upper)
+    if(NOT PY_${module_upper})
+        if(ARGC GREATER 1 AND ARGV1 STREQUAL "REQUIRED")
+            set(${module}_FIND_REQUIRED TRUE)
+        else()
+            set(${module}_FIND_REQUIRED FALSE)
+        endif()
+        # A module's location is usually a directory, but for binary modules
+        # it's a .so file.
+        execute_process(COMMAND "${PYTHON_EXECUTABLE}" "-c"
+            "import re, ${module}; print(re.compile('/__init__.py.*').sub('',${module}.__file__))"
+            RESULT_VARIABLE _${module}_status
+            OUTPUT_VARIABLE _${module}_location
+            ERROR_QUIET
+            OUTPUT_STRIP_TRAILING_WHITESPACE)
+        if(NOT _${module}_status)
+            set(PY_${module_upper} ${_${module}_location} CACHE STRING
+                "Location of Python module ${module}")
+        endif(NOT _${module}_status)
+    endif(NOT PY_${module_upper})
+    find_package_handle_standard_args(PY_${module} DEFAULT_MSG PY_${module_upper})
+    if(NOT PY_${module_upper}_FOUND AND ${module}_FIND_REQUIRED)
+        message(FATAL_ERROR "python module ${module} is not found")
+    endif()
+    set(PY_${module_upper}_FOUND ${PY_${module_upper}_FOUND} PARENT_SCOPE)
+endfunction(find_python_module)

cmake/FindSphinx.cmake

+# - This module looks for Sphinx
+# Find the Sphinx documentation generator
+#
+# This modules defines
+#  SPHINX_EXECUTABLE
+#  SPHINX_FOUND
+
+find_program(SPHINX_EXECUTABLE
+  NAMES sphinx-build
+  PATHS
+    /usr/bin
+    /usr/local/bin
+    /opt/local/bin
+  DOC "Sphinx documentation generator"
+)
+
+if( NOT SPHINX_EXECUTABLE )
+  set(_Python_VERSIONS
+    2.7 2.6 2.5 2.4 2.3 2.2 2.1 2.0 1.6 1.5
+  )
+
+  foreach( _version ${_Python_VERSIONS} )
+    set( _sphinx_NAMES sphinx-build-${_version} )
+
+    find_program( SPHINX_EXECUTABLE
+      NAMES ${_sphinx_NAMES}
+      PATHS
+        /usr/bin
+        /usr/local/bin
+        /opt/loca/bin
+      DOC "Sphinx documentation generator"
+    )
+  endforeach()
+endif()
+
+include(FindPackageHandleStandardArgs)
+
+find_package_handle_standard_args(Sphinx DEFAULT_MSG
+  SPHINX_EXECUTABLE
+)
+
+
+option( SPHINX_HTML_OUTPUT "Build a single HTML with the whole content." ON )
+option( SPHINX_DIRHTML_OUTPUT "Build HTML pages, but with a single directory per document." OFF )
+option( SPHINX_HTMLHELP_OUTPUT "Build HTML pages with additional information for building a documentation collection in htmlhelp." OFF )
+option( SPHINX_QTHELP_OUTPUT "Build HTML pages with additional information for building a documentation collection in qthelp." OFF )
+option( SPHINX_DEVHELP_OUTPUT "Build HTML pages with additional information for building a documentation collection in devhelp." OFF )
+option( SPHINX_EPUB_OUTPUT "Build HTML pages with additional information for building a documentation collection in epub." OFF )
+option( SPHINX_LATEX_OUTPUT "Build LaTeX sources that can be compiled to a PDF document using pdflatex." OFF )
+option( SPHINX_MAN_OUTPUT "Build manual pages in groff format for UNIX systems." OFF )
+option( SPHINX_TEXT_OUTPUT "Build plain text files." OFF )
+
+
+mark_as_advanced(
+  SPHINX_EXECUTABLE
+  SPHINX_HTML_OUTPUT
+  SPHINX_DIRHTML_OUTPUT
+  SPHINX_HTMLHELP_OUTPUT
+  SPHINX_QTHELP_OUTPUT
+  SPHINX_DEVHELP_OUTPUT
+  SPHINX_EPUB_OUTPUT
+  SPHINX_LATEX_OUTPUT
+  SPHINX_MAN_OUTPUT
+  SPHINX_TEXT_OUTPUT
+)
+
+function( Sphinx_add_target target_name builder conf cache source destination )
+  add_custom_target( ${target_name} ALL
+    COMMAND ${SPHINX_EXECUTABLE} -b ${builder}
+    -d ${cache}
+    -c ${conf}
+    ${source}
+    ${destination}
+    COMMENT "Generating sphinx documentation: ${builder}"
+    )
+
+  set_property(
+    DIRECTORY APPEND PROPERTY
+    ADDITIONAL_MAKE_CLEAN_FILES
+    ${destination}
+    )
+endfunction()
+
+# Target dependencies can be optionally listed at the end.
+function( Sphinx_add_targets target_base_name conf source base_destination )
+
+  set( _dependencies )
+
+  foreach( arg IN LISTS ARGN )
+    set( _dependencies ${_dependencies} ${arg} )
+  endforeach()
+
+  if( ${SPHINX_HTML_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_html html ${conf} ${source} ${base_destination}/html )
+
+    add_dependencies( ${target_base_name}_html ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_DIRHTML_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_dirhtml dirhtml ${conf} ${source} ${base_destination}/dirhtml )
+
+    add_dependencies( ${target_base_name}_dirhtml ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_QTHELP_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_qthelp qthelp ${conf} ${source} ${base_destination}/qthelp )
+
+    add_dependencies( ${target_base_name}_qthelp ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_DEVHELP_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_devhelp devhelp ${conf} ${source} ${base_destination}/devhelp )
+
+    add_dependencies( ${target_base_name}_devhelp ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_EPUB_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_epub epub ${conf} ${source} ${base_destination}/epub )
+
+    add_dependencies( ${target_base_name}_epub ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_LATEX_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_latex latex ${conf} ${source} ${base_destination}/latex )
+
+    add_dependencies( ${target_base_name}_latex ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_MAN_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_man man ${conf} ${source} ${base_destination}/man )
+
+    add_dependencies( ${target_base_name}_man ${_dependencies} )
+  endif()
+
+  if( ${SPHINX_TEXT_OUTPUT} )
+    Sphinx_add_target( ${target_base_name}_text text ${conf} ${source} ${base_destination}/text )
+
+    add_dependencies( ${target_base_name}_text ${_dependencies} )
+  endif()
+
+  if( ${BUILD_TESTING} )
+    sphinx_add_target( ${target_base_name}_linkcheck linkcheck ${conf} ${source} ${base_destination}/linkcheck )
+
+    add_dependencies( ${target_base_name}_linkcheck ${_dependencies} )
+  endif()
+endfunction()
\ No newline at end of file

cmake/cblas.cmake

+# Find the CBlas libraries
+#
+# It will search MKL, atlas, OpenBlas, reference-cblas in order.
+#
+# If any cblas implementation found, the following variable will be set.
+#    CBLAS_PROVIDER  # one of MKL, ATLAS, OPENBLAS, REFERENCE
+#    CBLAS_INC_DIR   # the include directory for cblas.
+#    CBLAS_LIBS      # a list of libraries should be linked by paddle. 
+#                    # Each library should be full path to object file.
+#
+# User should set one of MKL_ROOT, ATLAS_ROOT, OPENBLAS_ROOT, REFERENCE_CBLAS_ROOT
+# during cmake. If none of them set, it will try to find cblas implementation in
+# system paths.
+#
+
+
+## Find MKL First.
+set(MKL_ROOT $ENV{MKL_ROOT} CACHE PATH "Folder contains MKL")
+
+find_path(MKL_INCLUDE_DIR mkl.h PATHS ${MKL_ROOT}/include)
+find_library(MKL_CORE_LIB NAMES mkl_core PATHS ${MKL_ROOT}/lib)
+find_library(MKL_SEQUENTIAL_LIB NAMES mkl_sequential PATHS ${MKL_ROOT}/lib)
+find_library(MKL_INTEL_LP64 NAMES mkl_intel_lp64 PATHS ${MKL_ROOT}/lib)
+
+
+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}
+          ${MKL_SEQUENTIAL_LIB}
+          ${MKL_CORE_LIB})
+  add_definitions(-DPADDLE_USE_MKL)
+  return() # return file.
+endif()
+
+## Then find atlas.
+set(ATLAS_ROOT $ENV{ATLAS_ROOT} CACHE PATH "Folder contains Atlas")
+set(ATLAS_INCLUDE_SEARCH_PATHS
+        ${ATLAS_ROOT}/include
+        /usr/include
+        /usr/include/atlas)
+set(ATLAS_LIB_SEARCH_PATHS
+        ${ATLAS_ROOT}/lib
+        /usr/lib
+        /usr/lib/blas/atlas
+        /usr/lib/atlas
+        /usr/lib/atlas-base   # special for ubuntu 14.04.
+    )
+find_path(ATLAS_INC_DIR NAMES cblas.h 
+  PATHS ${ATLAS_INCLUDE_SEARCH_PATHS})
+find_library(ATLAS_CBLAS_LIB NAMES cblas libcblas.so.3 
+  PATHS ${ATLAS_LIB_SEARCH_PATHS})
+find_library(ATLAS_LIB NAMES atlas libatlas.so.3
+  PATHS ${ATLAS_LIB_SEARCH_PATHS})
+
+if(ATLAS_INC_DIR AND ATLAS_CBLAS_LIB AND ATLAS_LIB)
+  set(CBLAS_PROVIDER ATLAS)
+  set(CBLAS_INC_DIR ${ATLAS_INC_DIR})
+  set(CBLAS_LIBS ${ATLAS_LIB} ${ATLAS_CBLAS_LIB})
+  return()
+endif()
+
+## Then find openblas.
+set(OPENBLAS_ROOT $ENV{OPENBLAS_ROOT} CACHE PATH "Folder contains Openblas")
+set(OPENBLAS_INCLUDE_SEARCH_PATHS
+        ${OPENBLAS_ROOT}/include
+        /usr/include
+        /usr/include/openblas)
+set(OPENBLAS_LIB_SEARCH_PATHS
+        ${OPENBLAS_ROOT}/lib
+        /usr/lib
+        /usr/lib/blas/openblas
+        /usr/lib/openblas)
+
+find_path(OPENBLAS_INC_DIR NAMES cblas.h
+  PATHS ${OPENBLAS_INCLUDE_SEARCH_PATHS})
+find_library(OPENBLAS_LIB NAMES openblas
+  PATHS ${OPENBLAS_LIB_SEARCH_PATHS})
+
+if(OPENBLAS_INC_DIR AND OPENBLAS_LIB)
+  set(CBLAS_PROVIDER OPENBLAS)
+  set(CBLAS_INC_DIR ${OPENBLAS_INC_DIR})
+  set(CBLAS_LIBS ${OPENBLAS_LIB})
+  return()
+endif()
+
+
+## Then find the reference-cblas.  www.netlib.org/blas/
+
+
+set(REFERENCE_CBLAS_ROOT $ENV{REFERENCE_CBLAS_ROOT} CACHE PATH 
+  "Folder contains reference-cblas")
+set(REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS
+  ${REFERENCE_CBLAS_ROOT}/include
+  /usr/include
+  /usr/include/cblas
+)
+
+set(REFERENCE_CBLAS_LIB_SEARCH_PATHS
+  ${REFERENCE_CBLAS_ROOT}/lib
+  /usr/lib
+  /usr/lib/blas/reference/
+  /usr/lib/reference/
+)
+
+find_path(REFERENCE_CBLAS_INCLUDE_DIR NAMES cblas.h PATHS
+        ${REFERENCE_CBLAS_INCLUDE_SEARCH_PATHS})
+find_library(REFERENCE_CBLAS_LIBRARY NAMES cblas PATHS
+        ${REFERENCE_CBLAS_LIB_SEARCH_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()
+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/ccache.cmake

+# Use ccache if found ccache program
+
+find_program(CCACHE_FOUND ccache)
+
+if(CCACHE_FOUND)
+    message(STATUS "Ccache is founded, use ccache to speed up compile.")
+    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE ccache)
+    set_property(GLOBAL PROPERTY RULE_LAUNCH_LINK ccache)
+endif(CCACHE_FOUND)
\ No newline at end of file

cmake/check_packages.cmake

+# 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()
+
+if(WITH_GLOG)
+  find_package(Glog REQUIRED)
+endif()
+
+if(WITH_GFLAGS)
+  find_package(Gflags REQUIRED)
+endif()
+
+if(WITH_TESTING)
+  find_package(GTest REQUIRED)
+endif()
+
+if(WITH_DOC)
+  find_package(Sphinx REQUIRED)
+  find_package(Doxygen REQUIRED)
+  find_python_module(recommonmark REQUIRED)
+  find_python_module(breathe 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/cpplint.cmake

+# util to check C++ file style
+# * it basically use google cpplint.py.
+# * It provide "add_style_check_target" for cmake.
+#   Usage see add_style_check_target's document
+#
+# TODO(yuyang18): Add python style check.
+
+set(STYLE_FILTER)
+
+# diable unwanted filters
+
+# paddle do not indent public/potected/private in class
+set(STYLE_FILTER "${STYLE_FILTER}-whitespace/indent,")
+# paddle use mutable reference. BUT IT IS NOT RECOMMANDED
+set(STYLE_FILTER "${STYLE_FILTER}-runtime/references,")
+# paddle use relative path for include.
+set(STYLE_FILTER "${STYLE_FILTER}-build/include,")
+# paddle use <thread>, <mutex>, etc.
+set(STYLE_FILTER "${STYLE_FILTER}-build/c++11,")
+# paddle use c style casting. BUT IT IS NOT RECOMMANDED
+set(STYLE_FILTER "${STYLE_FILTER}-readability/casting")
+
+
+# IGNORE SOME FILES
+set(IGNORE_PATTERN
+    .*ImportanceSampler.*
+    .*cblas\\.h.*
+    .*LtrDataProvider.*
+    .*MultiDataProvider.*)
+
+# add_style_check_target
+#
+# attach check code style step for target.
+#
+# first argument: target name to attach
+# rest arguments: source list to check code style.
+# 
+# NOTE: If WITH_STYLE_CHECK is OFF, then this macro just do nothing.
+macro(add_style_check_target TARGET_NAME)
+    if(WITH_STYLE_CHECK)
+        set(SOURCES_LIST ${ARGN})
+        list(REMOVE_DUPLICATES SOURCES_LIST)
+        list(SORT SOURCES_LIST)
+
+        foreach(filename ${SOURCES_LIST})
+            set(LINT ON)
+            foreach(pattern ${IGNORE_PATTERN})
+                if(filename MATCHES ${pattern})
+                    message(STATUS "DROP LINT ${filename}")
+                    set(LINT OFF)
+                endif() 
+            endforeach()
+            if(LINT MATCHES ON)
+                add_custom_command(TARGET ${TARGET_NAME}
+                    PRE_BUILD
+                    COMMAND "${PYTHON_EXECUTABLE}" "${PROJ_ROOT}/paddle/scripts/cpplint.py"
+                                "--filter=${STYLE_FILTER}" ${filename}
+                    WORKING_DIRECTORY ${CMAKE_CURRENT_LIST_DIR})
+            endif()
+        endforeach()
+    endif()
+endmacro()

cmake/cudnn.cmake

+set(CUDNN_ROOT "" CACHE PATH "CUDNN ROOT")
+find_path(CUDNN_INCLUDE_DIR cudnn.h
+    PATHS ${CUDNN_ROOT} ${CUDNN_ROOT}/include
+    $ENV{CUDNN_ROOT} $ENV{CUDNN_ROOT}/include ${CUDA_TOOLKIT_INCLUDE}
+    NO_DEFAULT_PATH
+)
+
+get_filename_component(__libpath_hist ${CUDA_CUDART_LIBRARY} PATH)
+
+list(APPEND CUDNN_CHECK_LIBRARY_DIRS
+    ${CUDNN_ROOT}
+    ${CUDNN_ROOT}/lib64
+    ${CUDNN_ROOT}/lib
+    $ENV{CUDNN_ROOT}
+    $ENV{CUDNN_ROOT}/lib64
+    $ENV{CUDNN_ROOT}/lib
+    /usr/lib)
+find_library(CUDNN_LIBRARY NAMES libcudnn.so # libcudnn_static.a
+    PATHS ${CUDNN_CHECK_LIBRARY_DIRS} ${CUDNN_INCLUDE_DIR} ${__libpath_hist}
+          NO_DEFAULT_PATH
+    DOC "Path to cuDNN library.")
+
+
+if(CUDNN_INCLUDE_DIR AND CUDNN_LIBRARY)
+    set(CUDNN_FOUND ON)
+else()
+    set(CUDNN_FOUND OFF)
+endif()
+
+if(CUDNN_FOUND)
+    file(READ ${CUDNN_INCLUDE_DIR}/cudnn.h CUDNN_VERSION_FILE_CONTENTS)
+
+    get_filename_component(CUDNN_LIB_PATH ${CUDNN_LIBRARY} DIRECTORY)
+
+    string(REGEX MATCH "define CUDNN_VERSION +([0-9]+)"
+        CUDNN_VERSION "${CUDNN_VERSION_FILE_CONTENTS}")
+    string(REGEX REPLACE "define CUDNN_VERSION +([0-9]+)" "\\1"
+        CUDNN_VERSION "${CUDNN_VERSION}")
+
+    if("${CUDNN_VERSION}" STREQUAL "2000")
+        message(STATUS "Current cuDNN version is v2. ")
+    else()
+        string(REGEX MATCH "define CUDNN_MAJOR +([0-9]+)" CUDNN_MAJOR_VERSION
+            "${CUDNN_VERSION_FILE_CONTENTS}")
+        string(REGEX REPLACE "define CUDNN_MAJOR +([0-9]+)" "\\1"
+            CUDNN_MAJOR_VERSION "${CUDNN_MAJOR_VERSION}")
+        string(REGEX MATCH "define CUDNN_MINOR +([0-9]+)" CUDNN_MINOR_VERSION
+            "${CUDNN_VERSION_FILE_CONTENTS}")
+        string(REGEX REPLACE "define CUDNN_MINOR +([0-9]+)" "\\1"
+            CUDNN_MINOR_VERSION "${CUDNN_MINOR_VERSION}")
+        string(REGEX MATCH "define CUDNN_PATCHLEVEL +([0-9]+)"
+            CUDNN_PATCHLEVEL_VERSION "${CUDNN_VERSION_FILE_CONTENTS}")
+        string(REGEX REPLACE "define CUDNN_PATCHLEVEL +([0-9]+)" "\\1"
+            CUDNN_PATCHLEVEL_VERSION "${CUDNN_PATCHLEVEL_VERSION}")
+
+        if(NOT CUDNN_MAJOR_VERSION)
+            set(CUDNN_VERSION "???")
+        else()
+            math(EXPR CUDNN_VERSION
+                "${CUDNN_MAJOR_VERSION} * 1000 +
+                 ${CUDNN_MINOR_VERSION} * 100 + ${CUDNN_PATCHLEVEL_VERSION}")
+        endif()
+
+        message(STATUS "Current cuDNN header is ${CUDNN_INCLUDE_DIR}/cudnn.h. "
+            "Current cuDNN version is v${CUDNN_MAJOR_VERSION}. ")
+
+    endif()
+endif()

cmake/enableCXX11.cmake

+# Enable C++ 11 for GCC.
+# NOTE: It's only tested for gcc.
+include(CheckCXXCompilerFlag)
+CHECK_CXX_COMPILER_FLAG("-std=c++11" COMPILER_SUPPORT_CXX11)
+CHECK_CXX_COMPILER_FLAG("-std=c++0x" COMPILER_SUPPORT_CXX0X)
+
+if(COMPILER_SUPPORT_CXX11)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+elseif(COMPILER_SUPPORT_CXX0X)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++0x")
+else()
+    message(FATAL_ERROR "Your compiler must support c++11")
+endif()
\ No newline at end of file

cmake/flags.cmake

+# Setting Paddle Compile Flags
+include(CheckCXXCompilerFlag)
+include(CheckCCompilerFlag)
+include(CheckCXXSymbolExists)
+# safe_set_flag
+#
+# Set a compile flag only if compiler is support
+# is_c: is C flag or C++ flag, bool type.
+# src_list: The list name which the flag name will be append to.
+# flag_name: the flag name for compiler, such as '-Werror' '-Wall' etc
+# rest arguments: not used. 
+function(safe_set_flag is_c src_list flag_name)
+    string(REPLACE "-" "_" safe_name ${flag_name})
+    string(REPLACE "=" "_" safe_name ${safe_name})
+    if(is_c)
+        CHECK_C_COMPILER_FLAG(${flag_name} C_COMPILER_SUPPORT_FLAG_${safe_name})
+        set(safe_name C_COMPILER_SUPPORT_FLAG_${safe_name})
+    else()
+        CHECK_CXX_COMPILER_FLAG(${flag_name} CXX_COMPILER_SUPPORT_FLAG_${safe_name})
+        set(safe_name CXX_COMPILER_SUPPORT_FLAG_${safe_name})
+    endif()
+    if(${safe_name})
+        set(${src_list} "${${src_list}} ${flag_name}" PARENT_SCOPE)
+        if(is_c)
+          set(CUDA_NVCC_FLAGS
+              --compiler-options;${flag_name}
+              ${CUDA_NVCC_FLAGS}
+              PARENT_SCOPE)
+        endif()
+    endif()
+endfunction()
+
+# helper macro to set cflag
+macro(safe_set_cflag src_list flag_name)
+    safe_set_flag(ON ${src_list} ${flag_name})
+endmacro()
+
+# helper macro to set cxxflag
+macro(safe_set_cxxflag src_list flag_name)
+    safe_set_flag(OFF ${src_list} ${flag_name})
+endmacro()
+
+CHECK_CXX_SYMBOL_EXISTS(UINT64_MAX "stdint.h" UINT64_MAX_EXISTS)
+if(NOT UINT64_MAX_EXISTS)
+  set(CMAKE_REQUIRED_DEFINITIONS -D__STDC_LIMIT_MACROS)
+  CHECK_CXX_SYMBOL_EXISTS(UINT64_MAX "stdint.h" UINT64_MAX_EXISTS_HERE)
+  if(UINT64_MAX_EXISTS_HERE) 
+    set(CMAKE_REQUIRED_DEFINITIONS)
+    add_definitions(-D__STDC_LIMIT_MACROS)
+  else()
+    message(FATAL_ERROR "Cannot find symbol UINT64_MAX")
+  endif()
+endif()
+
+# Common flags. the compiler flag used for C/C++ sources whenever release or debug
+# Do not care if this flag is support for gcc.
+set(COMMON_FLAGS
+    -fPIC
+    -fno-omit-frame-pointer
+    -Wall
+    -Wextra
+    -Werror
+    -Wnon-virtual-dtor
+    -Wdelete-non-virtual-dtor
+    -Wno-unused-parameter
+    -Wno-error=literal-suffix
+    -Wno-error=unused-local-typedefs)
+
+foreach(flag ${COMMON_FLAGS})
+    safe_set_cflag(CMAKE_C_FLAGS ${flag})
+    safe_set_cxxflag(CMAKE_CXX_FLAGS ${flag})
+endforeach()
+
+# Release/Debug flags set by cmake. Such as -O3 -g -DNDEBUG etc.
+# So, don't set these flags here.
+
+foreach(capability 30 35 50)
+    list(APPEND __arch_flags "-gencode arch=compute_${capability},code=sm_${capability}")
+endforeach()
+
+if (CUDA_VERSION VERSION_GREATER "7.0")
+    list(APPEND __arch_flags "-gencode arch=compute_52,code=sm_52")
+endif()
+
+set(CUDA_NVCC_FLAGS ${__arch_flags} ${CUDA_NVCC_FLAGS})
+

cmake/package.cmake

+set(CPACK_PACKAGE_NAME paddle)
+set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "")
+set(CPACK_PACKAGE_VERSION_MAJOR ${PADDLE_MAJOR_VERSION})
+set(CPACK_PACKAGE_VERSION_MINOR ${PADDLE_MINOR_VERSION})
+set(CPACK_PACKAGE_VERSION_PATCH ${PADDLE_PATCH_VERSION})
+set(CPACK_PACKAGE_VERSION ${PADDLE_VERSION})
+## DEB Settings
+set(CPACK_DEBIAN_PACKAGE_NAME paddle)
+set(CPACK_DEBIAN_PACKAGE_ARCHITECTURE amd64)
+set(CPACK_DEBIAN_PACKAGE_MAINTAINER PaddlePaddle Dev <paddle-dev@baidu.com>)
+set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "Paddle")
+set(CPACK_PACKAGE_DESCRIPTION "")
+set(CPACK_DEBIAN_PACKAGE_DEPENDS "libatlas3-base, libgflags2, libgoogle-glog0, libprotobuf8, libpython2.7, libstdc++6, python-numpy, python-pip, python-pip-whl, python-protobuf")
+set(CPACK_DEBIAN_PACKAGE_SECTION Devel)
+set(CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${PROJ_ROOT}/paddle/scripts/deb/postinst")
+#set(CPACK_GENERATOR "DEB")
+# Start cpack
+include (CMakePackageConfigHelpers)
+include (CPack)
+
+

cmake/swig.cmake

+find_program(
+    SWIG_BINARY_PATH
+    swig)
+
+if(${SWIG_BINARY_PATH} STREQUAL "SWIG_BINARY_PATH-NOTFOUND")
+    set(SWIG_FOUND OFF)
+else()
+    set(SWIG_FOUND ON)
+endif()
+
+set(MIN_SWIG_VERSION 2)
+if(SWIG_FOUND)
+    execute_process(COMMAND sh -c "${SWIG_BINARY_PATH} -version | grep Version | cut -f3 -d' '"
+        OUTPUT_VARIABLE _SWIG_VERSION
+        OUTPUT_STRIP_TRAILING_WHITESPACE)
+    if(${_SWIG_VERSION} VERSION_LESS ${MIN_SWIG_VERSION})
+        message("swig version ${MIN_SWIG_VERSION} or greater is needed for generating python api. "
+                 "Only version ${_SWIG_VERSION} is found. Set SWIG_FOUND to FALSE")
+        set(SWIG_FOUND FALSE)
+    endif(${_SWIG_VERSION} VERSION_LESS ${MIN_SWIG_VERSION})
+endif(SWIG_FOUND)
+
+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
+        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/util.cmake

+# Some common routine for paddle compile.
+
+
+# target_circle_link_libraries
+# Link libraries to target which has circle dependencies.
+#
+# First Argument: target name want to be linked with libraries
+# Rest Arguments: libraries which link together.
+function(target_circle_link_libraries TARGET_NAME)
+    target_link_libraries(${TARGET_NAME}
+        -Wl,--start-group
+        ${ARGN}
+        -Wl,--end-group)
+endfunction()
+
+# compile_cu_as_cpp
+# Make a cu file compiled as C++
+# Arguments: Source files
+macro(compile_cu_as_cpp)
+    foreach(s ${ARGN})
+        set_source_files_properties(${s} PROPERTIES LANGUAGE CXX)
+        set_source_files_properties(${s} PROPERTIES COMPILE_FLAGS "-x c++")
+    endforeach()
+endmacro()
+
+# link_paddle_exe
+# add paddle library for a paddle executable, such as trainer, pserver.
+#
+# It will handle WITH_PYTHON/WITH_GLOG etc.
+function(link_paddle_exe TARGET_NAME)
+    if(WITH_METRIC)
+        if(WITH_GPU)
+            set(METRIC_LIBS paddle_metric_learning paddle_dserver_lib metric metric_cpu)
+        else()
+            set(METRIC_LIBS paddle_metric_learning paddle_dserver_lib metric_cpu)
+        endif()
+    else()
+        set(METRIC_LIBS "")
+    endif()
+
+    if(PADDLE_WITH_INTERNAL)
+        set(INTERAL_LIBS paddle_internal_gserver paddle_internal_parameter)
+        target_circle_link_libraries(${TARGET_NAME}
+            -Wl,--whole-archive
+            paddle_internal_gserver
+            paddle_internal_owlqn
+            -Wl,--no-whole-archive
+            paddle_internal_parameter)
+    else()
+        set(INTERAL_LIBS "")
+    endif()
+
+    target_circle_link_libraries(${TARGET_NAME}
+        -Wl,--whole-archive
+        paddle_gserver
+        ${METRIC_LIBS}
+        -Wl,--no-whole-archive
+        paddle_pserver
+        paddle_trainer_lib
+        paddle_network
+        paddle_math
+        paddle_utils
+        paddle_parameter
+        paddle_proto
+        paddle_cuda
+        ${METRIC_LIBS}
+        ${PROTOBUF_LIBRARY}
+        ${CMAKE_THREAD_LIBS_INIT}
+        ${CBLAS_LIBS}
+        ${CMAKE_DL_LIBS}
+        ${INTERAL_LIBS}
+        -lz)
+    
+    if(WITH_PYTHON)
+        target_link_libraries(${TARGET_NAME}
+            ${PYTHON_LIBRARIES})
+    endif()
+
+    if(WITH_GLOG)
+        target_link_libraries(${TARGET_NAME}
+            ${LIBGLOG_LIBRARY})
+    endif()
+
+    if(WITH_GFLAGS)
+        target_link_libraries(${TARGET_NAME}
+            ${GFLAGS_LIBRARIES})
+    endif()
+
+    if(WITH_GPU)
+        if(NOT WITH_DSO OR WITH_METRIC) 
+            target_link_libraries(${TARGET_NAME}
+                ${CUDNN_LIBRARY}
+                ${CUDA_curand_LIBRARY}) 
+            CUDA_ADD_CUBLAS_TO_TARGET(${TARGET_NAME})
+        endif()
+
+        check_library_exists(rt clock_gettime "time.h" HAVE_CLOCK_GETTIME )
+        if(HAVE_CLOCK_GETTIME)
+            target_link_libraries(${TARGET_NAME} rt)
+        endif()
+    endif()
+endfunction()
+
+# link_paddle_test
+# Link a paddle unittest for target
+# TARGET_NAME: the unittest target name
+# Rest Arguemnts: not used.
+function(link_paddle_test TARGET_NAME)
+    link_paddle_exe(${TARGET_NAME})
+    target_link_libraries(${TARGET_NAME} ${GTEST_MAIN_LIBRARIES}
+        ${GTEST_LIBRARIES})
+endfunction()
+
+# add_unittest_without_exec
+#
+# create a paddle unittest. not specifically define how to run this unittest.
+# TARGET_NAME: the unittest target name, same as executable file name
+# Rest Arguments: the source files to compile this unittest.
+macro(add_unittest_without_exec TARGET_NAME)
+    add_executable(${TARGET_NAME} ${ARGN})
+    link_paddle_test(${TARGET_NAME})
+    add_style_check_target(${TARGET_NAME} ${ARGN})
+endmacro()
+
+# add_unittest
+# create a paddle unittest and just to execute this binary to make unittest.
+#
+# TARGET_NAME: the unittest target name, same as executable file name
+# Rest Arguments: the source files to compile this unittest.
+macro(add_unittest TARGET_NAME)
+    add_unittest_without_exec(${TARGET_NAME} ${ARGN})
+    add_test(${TARGET_NAME} ${TARGET_NAME})
+endmacro()
+
+# add_simple_unittest
+# create a paddle unittest with file name. It just compile ${TARGET_NAME}.cpp to
+# ${TARGET_NAME} and then execute it.
+macro(add_simple_unittest TARGET_NAME)
+    add_unittest(${TARGET_NAME} ${TARGET_NAME}.cpp)
+endmacro()
+
+macro(add_paddle_culib TARGET_NAME)
+    set(NVCC_FLAG ${CUDA_NVCC_FLAGS})
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--use_fast_math)
+    cuda_add_library(${TARGET_NAME} STATIC ${ARGN})
+    set(CUDA_NVCC_FLAGS ${NVCC_FLAG})
+endmacro()

demo/image_classification/.gitignore

+data/cifar-10-batches-py
+data/cifar-out
+cifar_vgg_model/*
+plot.png
+train.log
+image_provider_copy_1.py
+*pyc

demo/image_classification/classify.py

+../model_zoo/resnet/classify.py
\ No newline at end of file

demo/image_classification/classify.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+python classify.py \
+     --job=predict \
+     --conf=vgg_16_cifar.py \
+     --model=./cifar_vgg_model/pass-00299 \
+     --multi_crop \
+     --data=./example/test.list

demo/image_classification/data/download_cifar.sh

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+wget https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
+tar zxf cifar-10-python.tar.gz
+rm cifar-10-python.tar.gz
+rm -rf cifar-out/*
+echo Converting CIFAR data to images.....
+python process_cifar.py ./cifar-10-batches-py ./cifar-out

demo/image_classification/data/process_cifar.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+import sys
+import os
+import PIL.Image as Image
+
+"""
+  Usage: python process_cifar input_dir output_dir
+"""
+
+
+def mkdir_not_exist(path):
+    """
+    Make dir if the path does not exist.
+    path: the path to be created.
+    """
+    if not os.path.exists(path):
+        os.mkdir(path)
+
+def create_dir_structure(output_dir):
+    """
+    Create the directory structure for the directory.
+    output_dir: the direcotry structure path.
+    """
+    mkdir_not_exist(os.path.join(output_dir))
+    mkdir_not_exist(os.path.join(output_dir, "train"))
+    mkdir_not_exist(os.path.join(output_dir, "test"))
+
+def convert_batch(batch_path, label_set, label_map,
+                  output_dir, data_split):
+    """
+    Convert CIFAR batch to the structure of Paddle format.
+    batch_path: the batch to be converted.
+    label_set: the set of labels.
+    output_dir: the output path.
+    data_split: whether it is training or testing data.
+    """
+    data = np.load(batch_path)
+    for data, label, filename in zip(data['data'], data['labels'],
+                                     data['filenames']):
+        data = data.reshape((3, 32, 32))
+        data = np.transpose(data, (1, 2, 0))
+        label = label_map[label]
+        output_dir_this = os.path.join(output_dir, data_split, str(label))
+        output_filename = os.path.join(output_dir_this, filename)
+        if not label in label_set:
+            label_set[label] = True
+            mkdir_not_exist(output_dir_this)
+        Image.fromarray(data).save(output_filename)
+
+
+if __name__ == '__main__':
+    input_dir = sys.argv[1]
+    output_dir = sys.argv[2]
+    num_batch = 5
+    create_dir_structure(output_dir)
+    label_map = {0: "airplane", 1: "automobile", 2: "bird", 3: "cat", 4: "deer",
+                 5: "dog", 6: "frog", 7: "horse", 8: "ship", 9: "truck"}
+    labels = {}
+    for i in range(1, num_batch + 1):
+        convert_batch(os.path.join(input_dir, "data_batch_%d" % i), labels,
+                      label_map, output_dir, "train")
+    convert_batch(os.path.join(input_dir, "test_batch"), {},
+                  label_map, output_dir, "test")
\ No newline at end of file

demo/image_classification/image_predictor.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import numpy as np
+from optparse import OptionParser
+
+from py_paddle import swig_paddle, util, DataProviderWrapperConverter
+from paddle.trainer.PyDataProviderWrapper import DenseSlot
+from paddle.trainer.config_parser import parse_config
+
+
+
+"""
+Will merge predictor from Qingqing.
+"""

demo/image_classification/image_provider.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import io
+import random
+
+import paddle.utils.image_util as image_util
+from paddle.trainer.PyDataProvider2 import *
+
+
+#
+# {'img_size': 32,
+# 'settings': <paddle.trainer.PyDataProviderWrapper.Cls instance at 0x7fea27cb6050>,
+# 'color': True,
+# 'mean_img_size': 32,
+# 'meta': './data/cifar-out/batches/batches.meta',
+# 'num_classes': 10,
+# 'file_list': ('./data/cifar-out/batches/train_batch_000',),
+# 'use_jpeg': True}
+def hook(settings, img_size, mean_img_size, num_classes, color, meta, use_jpeg,
+         is_train, **kwargs):
+    settings.mean_img_size = mean_img_size
+    settings.img_size = img_size
+    settings.num_classes = num_classes
+    settings.color = color
+    settings.is_train = is_train
+
+    if settings.color:
+        settings.img_raw_size = settings.img_size * settings.img_size * 3
+    else:
+        settings.img_raw_size = settings.img_size * settings.img_size
+
+    settings.meta_path = meta
+    settings.use_jpeg = use_jpeg
+
+    settings.img_mean = image_util.load_meta(settings.meta_path,
+                                             settings.mean_img_size,
+                                             settings.img_size,
+                                             settings.color)
+
+    settings.logger.info('Image size: %s', settings.img_size)
+    settings.logger.info('Meta path: %s', settings.meta_path)
+    settings.input_types = [
+        dense_vector(settings.img_raw_size),  # image feature
+        integer_value(settings.num_classes)]  # labels
+
+    settings.logger.info('DataProvider Initialization finished')
+
+
+@provider(init_hook=hook)
+def processData(settings, file_name):
+    """
+    The main function for loading data.
+    Load the batch, iterate all the images and labels in this batch.
+    file_name: the batch file name.
+    """
+    data = cPickle.load(io.open(file_name, 'rb'))
+    indexes = list(range(len(data['images'])))
+    if settings.is_train:
+        random.shuffle(indexes)
+    for i in indexes:
+        if settings.use_jpeg == 1:
+            img = image_util.decode_jpeg(data['images'][i])
+        else:
+            img = data['images'][i]
+        img_feat = image_util.preprocess_img(img, settings.img_mean,
+                                             settings.img_size, settings.is_train,
+                                             settings.color)
+        label = data['labels'][i]
+        yield img_feat.tolist(), int(label)

demo/image_classification/image_util.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import numpy as np
+from PIL import Image
+from cStringIO import StringIO
+
+def resize_image(img, target_size):
+    """
+    Resize an image so that the shorter edge has length target_size.
+    img: the input image to be resized.
+    target_size: the target resized image size.
+    """
+    percent = (target_size/float(min(img.size[0], img.size[1])))
+    resized_size = int(round(img.size[0] * percent)), int(round(img.size[1] * percent))
+    img = img.resize(resized_size, Image.ANTIALIAS)
+    return img
+
+def flip(im):
+    """
+    Return the flipped image.
+    Flip an image along the horizontal direction.
+    im: input image, (H x W x K) ndarrays 
+    """
+    if len(im.shape) == 3:
+        return im[:, :, ::-1]
+    else:
+        return im[:, ::-1]
+
+def crop_img(im, inner_size, color=True, test=True):
+    """
+    Return cropped image.
+    The size of the cropped image is inner_size * inner_size.
+    im: (K x H x W) ndarrays
+    inner_size: the cropped image size.
+    color: whether it is color image.
+    test: whether in test mode.
+      If False, does random cropping and flipping.
+      If True, crop the center of images.
+    """
+    if color:
+        height, width = max(inner_size, im.shape[1]), max(inner_size, im.shape[2])
+        padded_im = np.zeros((3, height, width))
+        startY = (height - im.shape[1]) / 2
+        startX = (width - im.shape[2]) / 2
+        endY, endX = startY + im.shape[1], startX + im.shape[2]
+        padded_im[:, startY: endY, startX: endX] = im
+    else:
+        im = im.astype('float32')
+        height, width = max(inner_size, im.shape[0]), max(inner_size, im.shape[1])
+        padded_im = np.zeros((height, width))
+        startY = (height - im.shape[0]) / 2
+        startX = (width - im.shape[1]) / 2
+        endY, endX = startY + im.shape[0], startX + im.shape[1]
+        padded_im[startY: endY, startX: endX] = im
+    if test:
+        startY = (height - inner_size) / 2
+        startX = (width - inner_size) / 2
+    else:
+        startY = np.random.randint(0, height - inner_size + 1)
+        startX = np.random.randint(0, width - inner_size + 1)
+    endY, endX = startY + inner_size, startX + inner_size
+    if color:
+        pic = padded_im[:, startY: endY, startX: endX]
+    else:
+        pic = padded_im[startY: endY, startX: endX]
+    if (not test) and (np.random.randint(2) == 0):
+        pic = flip(pic)
+    return pic
+
+def decode_jpeg(jpeg_string):
+    np_array = np.array(Image.open(StringIO(jpeg_string)))
+    if len(np_array.shape) == 3:
+        np_array = np.transpose(np_array, (2, 0, 1))
+    return np_array
+
+def preprocess_img(im, img_mean, crop_size, is_train, color=True):
+    """
+    Does data augmentation for images.
+    If is_train is false, cropping the center region from the image.
+    If is_train is true, randomly crop a region from the image,
+    and randomy does flipping.
+    im: (K x H x W) ndarrays
+    """
+    im = im.astype('float32')
+    test = not is_train
+    pic = crop_img(im, crop_size, color, test)
+    pic -= img_mean
+    return pic.flatten()
+
+def load_meta(meta_path, mean_img_size, crop_size, color=True):
+    """
+    Return the loaded meta file.
+    Load the meta image, which is the mean of the images in the dataset.
+    The mean image is subtracted from every input image so that the expected mean
+    of each input image is zero.
+    """
+    mean = np.load(meta_path)['data_mean']
+    border = (mean_img_size - crop_size) / 2
+    if color:
+        assert(mean_img_size * mean_img_size * 3 == mean.shape[0])
+        mean = mean.reshape(3, mean_img_size, mean_img_size)
+        mean = mean[:, border: border + crop_size,
+                       border: border + crop_size].astype('float32')
+    else:
+        assert(mean_img_size * mean_img_size == mean.shape[0])
+        mean = mean.reshape(mean_img_size, mean_img_size)
+        mean = mean[border: border + crop_size,
+                    border: border + crop_size].astype('float32')
+    return mean
+
+def load_image(img_path, is_color=True):
+    """
+    Load image and return. 
+    img_path: image path.
+    is_color: is color image or not.
+    """
+    img = Image.open(img_path)
+    img.load()
+    return img
+
+def oversample(img, crop_dims):
+    """
+    image : iterable of (H x W x K) ndarrays
+    crop_dims: (height, width) tuple for the crops.
+    Returned data contains ten crops of input image, namely,
+    four corner patches and the center patch as well as their
+    horizontal reflections.
+    """
+    # Dimensions and center.
+    im_shape = np.array(img[0].shape)
+    crop_dims = np.array(crop_dims)
+    im_center = im_shape[:2] / 2.0
+
+    # Make crop coordinates
+    h_indices = (0, im_shape[0] - crop_dims[0])
+    w_indices = (0, im_shape[1] - crop_dims[1])
+    crops_ix = np.empty((5, 4), dtype=int)
+    curr = 0
+    for i in h_indices:
+        for j in w_indices:
+            crops_ix[curr] = (i, j, i + crop_dims[0], j + crop_dims[1])
+            curr += 1
+    crops_ix[4] = np.tile(im_center, (1, 2)) + np.concatenate([
+        -crop_dims / 2.0,
+         crop_dims / 2.0
+    ])
+    crops_ix = np.tile(crops_ix, (2, 1))
+
+    # Extract crops
+    crops = np.empty((10 * len(img), crop_dims[0], crop_dims[1],
+                      im_shape[-1]), dtype=np.float32)
+    ix = 0
+    for im in img:
+        for crop in crops_ix:
+            crops[ix] = im[crop[0]:crop[2], crop[1]:crop[3], :]
+            ix += 1
+        crops[ix-5:ix] = crops[ix-5:ix, :, ::-1, :]  # flip for mirrors
+    return crops
+
+class ImageTransformer:
+    def __init__(self, transpose = None,
+                 channel_swap = None, mean = None, is_color = True):
+        self.transpose = transpose
+        self.channel_swap = None
+        self.mean = None
+        self.is_color = is_color 
+
+    def set_transpose(self, order): 
+        if self.is_color:
+            assert 3 == len(order) 
+        self.transpose = order
+
+    def set_channel_swap(self, order): 
+        if self.is_color:
+            assert 3 == len(order) 
+        self.channel_swap = order
+
+    def set_mean(self, mean):
+        # mean value, may be one value per channel 
+        if mean.ndim == 1:
+            mean = mean[:, np.newaxis, np.newaxis]       
+        else: 
+            # elementwise mean
+            if self.is_color:
+                assert len(mean.shape) == 3
+        self.mean = mean 
+
+    def transformer(self, data):
+        if self.transpose is not None:
+            data = data.transpose(self.transpose)
+        if self.channel_swap is not None:
+            data = data[self.channel_swap, :, :]
+        if self.mean is not None:
+            data -= self.mean
+        return data

demo/image_classification/preprocess.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.utils.preprocess_img import ImageClassificationDatasetCreater
+from optparse import OptionParser
+
+
+def option_parser():
+    parser = OptionParser(usage="usage: python preprcoess.py "\
+                          "-i data_dir [options]")
+    parser.add_option("-i", "--input", action="store",
+                      dest="input", help="Input data directory.")
+    parser.add_option("-s", "--size", action="store",
+                      dest="size", help="Processed image size.")
+    parser.add_option("-c", "--color", action="store",
+                      dest="color", help="whether to use color images.")
+    return parser.parse_args()
+
+if __name__ == '__main__':
+     options, args = option_parser()
+     data_dir = options.input
+     processed_image_size = int(options.size)
+     color = options.color == "1"
+     data_creator = ImageClassificationDatasetCreater(data_dir,
+                                                      processed_image_size,
+                                                      color)
+     data_creator.num_per_batch = 1000
+     data_creator.overwrite = True
+     data_creator.create_batches()

demo/image_classification/preprocess.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+export PYTHONPATH=$PYTHONPATH:../../
+
+data_dir=./data/cifar-out
+
+python preprocess.py -i $data_dir -s 32 -c 1

demo/image_classification/train.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+config=vgg_16_cifar.py
+output=./cifar_vgg_model
+log=train.log
+
+paddle train \
+--config=$config \
+--dot_period=10 \
+--log_period=100 \
+--test_all_data_in_one_period=1 \
+--use_gpu=1 \
+--trainer_count=1 \
+--num_passes=200 \
+--save_dir=$output \
+2>&1 | tee $log
+
+python -m paddle.utils.plotcurve -i $log > plot.png

demo/image_classification/upload_hadoop.sh

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+hadoop fs -Dhadoop.job.ugi=paddle_demo,paddle_demo -put data/cifar-out/batches/train_batch_* /app/idl/idl-dl/paddle/demo/image_classification/train/
+hadoop fs -Dhadoop.job.ugi=paddle_demo,paddle_demo -put data/cifar-out/batches/test_batch_* /app/idl/idl-dl/paddle/demo/image_classification/test/
+hadoop fs -Dhadoop.job.ugi=paddle_demo,paddle_demo -put data/cifar-out/batches/batches.meta /app/idl/idl-dl/paddle/demo/image_classification/train_meta
+hadoop fs -Dhadoop.job.ugi=paddle_demo,paddle_demo -put data/cifar-out/batches/batches.meta /app/idl/idl-dl/paddle/demo/image_classification/test_meta

demo/image_classification/vgg_16_cifar.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer_config_helpers import *
+
+is_predict = get_config_arg("is_predict", bool, False)
+
+####################Data Configuration ##################
+if not is_predict:
+  data_dir='data/cifar-out/batches/'
+  meta_path=data_dir+'batches.meta'
+
+  args = {'meta':meta_path,'mean_img_size': 32,
+          'img_size': 32,'num_classes': 10,
+          'use_jpeg': 1,'color': "color"}
+
+  define_py_data_sources2(train_list=data_dir+"train.list",
+                          test_list=data_dir+'test.list',
+                          module='image_provider',
+                          obj='processData',
+                          args=args)
+
+######################Algorithm Configuration #############
+settings(
+    batch_size = 128,
+    learning_rate = 0.1 / 128.0,
+    learning_method = MomentumOptimizer(0.9),
+    regularization = L2Regularization(0.0005 * 128)
+)
+
+#######################Network Configuration #############
+data_size=3*32*32
+label_size=10
+img = data_layer(name='image',
+                 size=data_size)
+# small_vgg is predined in trainer_config_helpers.network
+predict = small_vgg(input_image=img,
+                    num_channels=3,
+                    num_classes=label_size)
+
+if not is_predict:
+    lbl = data_layer(name="label", size=label_size)
+    outputs(classification_cost(input=predict, label=lbl))
+else:
+    outputs(predict)

demo/model_zoo/embedding/.gitignore

+baidu.dict
+model_*.emb

demo/model_zoo/embedding/extract_para.py

+#!/bin/env python
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Example:
+    python extract_para.py --preModel PREMODEL --preDict PREDICT \
+                            --usrModel USRMODEL --usrDict USRDICT -d DIM
+
+Options:
+    -h, --help          show this help message and exit
+    --preModel PREMODEL the name of pretrained embedding model
+    --preDict PREDICT   the name of pretrained dictionary
+    --usrModel usrModel the name of output usr embedding model
+    --usrDict usrDict   the name of user specified dictionary
+    -d DIM              dimension of parameter
+"""
+from optparse import OptionParser
+import struct
+
+def get_row_index(preDict, usrDict):
+    """
+    Get the row positions for all words in user dictionary from pre-trained dictionary.
+    return: a list of row positions
+    Example: preDict='a\nb\nc\n', usrDict='a\nc\n', then return [0,2]
+    """
+    pos = []
+    index = dict()
+    with open(preDict, "r") as f:
+        for line_index, line in enumerate(f):
+            word = line.strip().split()[0]
+            index[word] = line_index
+    with open(usrDict, "r") as f:
+        for line in f:
+            word = line.strip().split()[0]
+            pos.append(index[word])
+    return pos
+
+def extract_parameters_by_usrDict(preModel, preDict, usrModel, usrDict, paraDim):
+    """
+    Extract desired parameters from a pretrained embedding model based on user dictionary
+    """
+    if paraDim not in [32, 64, 128, 256]:
+        raise RuntimeError("We only support 32, 64, 128, 256 dimensions now")
+
+    fi = open(preModel, "rb")
+    fo = open(usrModel, "wb")
+
+    # write filehead
+    rowIndex = get_row_index(preDict, usrDict)
+    newHead = struct.pack("iil", 0, 4, len(rowIndex) * paraDim)
+    fo.write(newHead)
+    bytes = 4 * paraDim
+    for i in range(0, len(rowIndex)):
+        # find the absolute position of input file
+        fi.seek(rowIndex[i] * bytes + 16, 0)
+        fo.write(fi.read(bytes))
+
+    print "extract parameters finish, total", len(rowIndex), "lines"
+    fi.close()
+
+def main():
+    """
+    Main entry for running paraconvert.py 
+    """
+    usage = "usage: \n" \
+            "python %prog --preModel PREMODEL --preDict PREDICT" \
+            " --usrModel USRMODEL --usrDict USRDICT -d DIM"
+    parser = OptionParser(usage)
+    parser.add_option("--preModel", action="store", dest="preModel",
+                      help="the name of pretrained embedding model")
+    parser.add_option("--preDict", action="store", dest="preDict",
+                      help="the name of pretrained dictionary")
+    parser.add_option("--usrModel", action="store", dest="usrModel",
+                      help="the name of output usr embedding model")
+    parser.add_option("--usrDict", action="store", dest="usrDict",
+                      help="the name of user specified dictionary")
+    parser.add_option("-d", action="store", dest="dim",
+                      help="dimension of parameter")
+    (options, args) = parser.parse_args()
+    extract_parameters_by_usrDict(options.preModel, options.preDict, 
+                      options.usrModel, options.usrDict, int(options.dim))
+
+if __name__ == '__main__':
+    main()

demo/model_zoo/embedding/paraconvert.py

+#!/bin/env python
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Example:
+    python paraconvert.py --b2t -i INPUT -o OUTPUT -d DIM
+    python paraconvert.py --t2b -i INPUT -o OUTPUT
+
+Options:
+    -h, --help  show this help message and exit
+    --b2t       convert parameter file of embedding model from binary to text
+    --t2b       convert parameter file of embedding model from text to binary
+    -i INPUT    input parameter file name
+    -o OUTPUT   output parameter file name
+    -d DIM      dimension of parameter
+"""
+from optparse import OptionParser
+import struct
+
+def binary2text(input, output, paraDim):
+    """
+    Convert a binary parameter file of embedding model to be a text file.  
+    input: the name of input binary parameter file, the format is:
+           1) the first 16 bytes is filehead:
+                version(4 bytes): version of paddle, default = 0
+                floatSize(4 bytes): sizeof(float) = 4
+                paraCount(8 bytes): total number of parameter
+           2) the next (paraCount * 4) bytes is parameters, each has 4 bytes 
+    output: the name of output text parameter file, for example:
+           0,4,32156096
+           -0.7845433,1.1937413,-0.1704215,...
+           0.0000909,0.0009465,-0.0008813,...
+           ...
+           the format is:
+           1) the first line is filehead: 
+              version=0, floatSize=4, paraCount=32156096
+           2) other lines print the paramters
+              a) each line prints paraDim paramters splitted by ','
+              b) there is paraCount/paraDim lines (embedding words)
+    paraDim: dimension of parameters 
+    """
+    fi = open(input, "rb")
+    fo = open(output, "w")
+    """
+    """
+    version, floatSize, paraCount = struct.unpack("iil", fi.read(16))
+    newHead = ','.join([str(version), str(floatSize), str(paraCount)])
+    print >> fo, newHead
+
+    bytes = 4 * int(paraDim)
+    format = "%df" % int(paraDim)
+    context = fi.read(bytes)
+    line = 0
+
+    while context:
+        numbers = struct.unpack(format, context)
+        lst = []
+        for i in numbers:
+            lst.append('%8.7f' % i)
+        print >> fo, ','.join(lst)
+        context = fi.read(bytes)
+        line += 1
+    fi.close()
+    fo.close()
+    print "binary2text finish, total", line, "lines"
+
+def get_para_count(input):
+    """
+    Compute the total number of embedding parameters in input text file. 
+    input: the name of input text file
+    """
+    numRows = 1 
+    paraDim = 0
+    with open(input) as f:
+        line = f.readline()
+        paraDim = len(line.split(","))
+        for line in f:
+            numRows += 1
+    return numRows * paraDim
+
+def text2binary(input, output, paddle_head=True):
+    """
+    Convert a text parameter file of embedding model to be a binary file.
+    input: the name of input text parameter file, for example:
+           -0.7845433,1.1937413,-0.1704215,...
+           0.0000909,0.0009465,-0.0008813,... 
+           ...
+           the format is:
+           1) it doesn't have filehead
+           2) each line stores the same dimension of parameters, 
+              the separator is commas ','
+    output: the name of output binary parameter file, the format is:
+           1) the first 16 bytes is filehead: 
+             version(4 bytes), floatSize(4 bytes), paraCount(8 bytes)
+           2) the next (paraCount * 4) bytes is parameters, each has 4 bytes
+    """
+    fi = open(input, "r")
+    fo = open(output, "wb")
+
+    newHead = struct.pack("iil", 0, 4, get_para_count(input))
+    fo.write(newHead)
+
+    count = 0
+    for line in fi:
+        line = line.strip().split(",")
+        for i in range(0, len(line)):
+            binary_data = struct.pack("f", float(line[i]))
+            fo.write(binary_data)
+        count += 1
+    fi.close()
+    fo.close()
+    print "text2binary finish, total", count, "lines"
+
+def main():
+    """
+    Main entry for running paraconvert.py 
+    """
+    usage = "usage: \n" \
+            "python %prog --b2t -i INPUT -o OUTPUT -d DIM \n" \
+            "python %prog --t2b -i INPUT -o OUTPUT"
+    parser = OptionParser(usage)
+    parser.add_option("--b2t", action="store_true",
+                      help="convert parameter file of embedding model from binary to text")
+    parser.add_option("--t2b", action="store_true",
+                      help="convert parameter file of embedding model from text to binary")
+    parser.add_option("-i", action="store", dest="input",
+                      help="input parameter file name")
+    parser.add_option("-o", action="store", dest="output",
+                      help="output parameter file name")
+    parser.add_option("-d", action="store", dest="dim",
+                      help="dimension of parameter")
+    (options, args) = parser.parse_args()
+    if options.b2t:
+        binary2text(options.input, options.output, options.dim)
+    if options.t2b:
+        text2binary(options.input, options.output)
+
+if __name__ == '__main__':
+    main()

demo/model_zoo/embedding/pre_DictAndModel.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+set -x
+
+# download the dictionary and pretrained model 
+for file in baidu.dict model_32.emb model_64.emb model_128.emb model_256.emb
+do 
+  # following is the google drive address
+  # you can also directly download from https://pan.baidu.com/s/1o8q577s
+  wget https://www.googledrive.com/host/0B7Q8d52jqeI9ejh6Q1RpMTFQT1k/embedding/$file --no-check-certificate
+done

demo/model_zoo/resnet/.gitignore

+fea_output/
+features/
+model.list
+ResNet_50.dot
+ResNet_50.png

demo/model_zoo/resnet/classify.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import cPickle
+import logging
+from PIL import Image
+import numpy as np
+from optparse import OptionParser
+
+import paddle.utils.image_util as image_util
+
+from py_paddle import swig_paddle, util
+from py_paddle import DataProviderWrapperConverter
+from paddle.trainer.PyDataProviderWrapper import DenseSlot
+from paddle.trainer.config_parser import parse_config
+
+logging.basicConfig(format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
+logging.getLogger().setLevel(logging.INFO)
+
+class ImageClassifier():
+    def __init__(self, train_conf, model_dir=None,
+                 resize_dim=256, crop_dim=224,
+                 mean_file=None,
+                 output_layer=None,
+                 oversample=False, is_color=True):
+        """
+        train_conf: network configure.
+        model_dir: string, directory of model.
+        resize_dim: int, resized image size.
+        crop_dim: int, crop size.
+        mean_file: string, image mean file.
+        oversample: bool, oversample means multiple crops, namely five
+                    patches (the four corner patches and the center
+                    patch) as well as their horizontal reflections,
+                    ten crops in all.
+        """
+        self.train_conf = train_conf
+        self.model_dir = model_dir
+        if model_dir is None:
+            self.model_dir = os.path.dirname(train_conf)
+
+        self.resize_dim = resize_dim
+        self.crop_dims = [crop_dim, crop_dim]
+        self.oversample = oversample
+        self.is_color = is_color
+
+        self.output_layer = output_layer
+        if self.output_layer:
+            assert isinstance(self.output_layer, basestring)
+            self.output_layer = self.output_layer.split(",")
+
+        self.transformer = image_util.ImageTransformer(is_color = is_color)
+        self.transformer.set_transpose((2,0,1))
+        self.transformer.set_channel_swap((2,1,0))
+
+        self.mean_file = mean_file
+        if self.mean_file is not None:
+            mean = np.load(self.mean_file)['data_mean']
+            mean = mean.reshape(3, self.crop_dims[0], self.crop_dims[1])
+            self.transformer.set_mean(mean) # mean pixel
+        else:
+            # if you use three mean value, set like:
+            # this three mean value is calculated from ImageNet.
+            self.transformer.set_mean(np.array([103.939,116.779,123.68]))
+
+        conf_args = "is_test=1,use_gpu=1,is_predict=1"
+        conf = parse_config(train_conf, conf_args)
+        swig_paddle.initPaddle("--use_gpu=1")
+        self.network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
+        assert isinstance(self.network, swig_paddle.GradientMachine)
+        self.network.loadParameters(self.model_dir)
+
+        data_size = 3 * self.crop_dims[0] * self.crop_dims[1]
+        slots = [DenseSlot(data_size)]
+        is_sequence = False
+        self.converter = util.DataProviderWrapperConverter(is_sequence, slots)
+
+    def get_data(self, img_path):
+        """
+        1. load image from img_path.
+        2. resize or oversampling.
+        3. transformer data: transpose, channel swap, sub mean.
+        return K x H x W ndarray.
+
+        img_path: image path.
+        """
+        image = image_util.load_image(img_path, self.is_color)
+        # Another way to extract oversampled features is that
+        # cropping and averaging from large feature map which is
+        # calculated by large size of image.
+        # This way reduces the computation.
+        if self.oversample:
+            # image_util.resize_image: short side is self.resize_dim
+            image = image_util.resize_image(image, self.resize_dim)
+            image = np.array(image)
+            input = np.zeros((1, image.shape[0], image.shape[1], 3),
+                             dtype=np.float32)
+            input[0] = image.astype(np.float32)
+            input = image_util.oversample(input, self.crop_dims)
+        else:
+            image = image.resize(self.crop_dims, Image.ANTIALIAS)
+            input = np.zeros((1, self.crop_dims[0], self.crop_dims[1], 3),
+                             dtype=np.float32)
+            input[0] = np.array(image).astype(np.float32)
+
+        data_in = []
+        for img in input:
+            img = self.transformer.transformer(img).flatten()
+            data_in.append([img.tolist()])
+        # paddle input: [[[]],[[]],...], [[]] is one sample.
+        return data_in
+
+    def forward(self, input_data):
+        """
+        return output arguments which are the Outputs() in network configure.
+
+        input_data: py_paddle input data.
+        call forward.
+        """
+        in_arg = self.converter(input_data)
+        return self.network.forwardTest(in_arg)
+
+    def forward(self, data, output_layer):
+        """
+        return output arguments which are the Outputs() in network configure.
+
+        input_data: py_paddle input data.
+        call forward.
+        """
+        input = self.converter(data)
+        self.network.forwardTest(input)
+        output = self.network.getLayerOutputs(output_layer)
+        res = {}
+        if isinstance(output_layer, basestring):
+            output_layer = [output_layer]
+        for name in output_layer:
+            # For oversampling, average predictions across crops.
+            # If not, the shape of output[name]: (1, class_number),
+            # the mean is also applicable.
+            res[name] = output[name].mean(0)
+
+        return res
+
+    def predict(self, data_file):
+        """
+        call forward and predicting.
+
+        data_file: input image list.
+        """
+        image_files = open(data_file, 'rb').readlines()
+        results = {}
+        if self.output_layer is None:
+            self.output_layer = ["output"]
+        for line in image_files:
+            image = line.split()[0]
+            data = self.get_data(image)
+            prob = self.forward(data, self.output_layer)
+            lab = np.argsort(-prob[self.output_layer[0]])
+            results[image] = lab[0]
+            logging.info("Label of %s is: %d", image, lab[0])
+        return results
+
+    def extract(self, data_file, output_dir, batch_size = 10000):
+        """
+        extract and save features of output layers, which are
+        specify in Outputs() in network configure.
+
+        data_file: file name of input data.
+        output_dir: saved directory of extracted features.
+        batch_size: sample number of one batch file.
+        """
+        if not os.path.exists(output_dir):
+            os.mkdir(output_dir)
+
+        sample_num = 0
+        batch_num = 0
+        image_feature = {}
+        image_files = open(data_file, 'rb').readlines()
+        for idx, line in enumerate(image_files):
+            image = line.split()[0]
+            data = self.get_data(image)
+            feature = self.forward(data, self.output_layer)
+            # save extracted features
+            file_name = image.split("/")[-1]
+            image_feature[file_name] = feature
+            sample_num += 1
+            if sample_num == batch_size:
+                batch_name = os.path.join(output_dir, 'batch_%d' %(batch_num))
+                self.save_file(image_feature, batch_name)
+                logging.info('Finish batch %d', batch_num)
+                batch_num += 1
+                sample_num = 0
+                image_feature = {}
+            if idx % 1000 == 0:
+                logging.info('%d/%d, %s', idx, len(image_files), file_name)
+        if sample_num > 0:
+            batch_name = os.path.join(output_dir, 'batch_%d' %(batch_num))
+            self.save_file(image_feature, batch_name)
+            logging.info('Finish batch %d', batch_num)
+        logging.info('Done: make image feature batch')
+
+    def save_file(self, data, file):
+        of = open(file, 'wb')
+        cPickle.dump(data, of, protocol=cPickle.HIGHEST_PROTOCOL)
+
+def option_parser():
+    """
+    Main entry for predciting
+    """
+    usage = "%prog -c config -i data_list -w model_dir [options]"
+    parser = OptionParser(usage="usage: %s" % usage)
+    parser.add_option("-j", "--job",
+                      action="store", dest="job_type",
+                      help="job type: predict, extract\
+                            predict: predicting,\
+                            extract: extract features")
+    parser.add_option("-c", "--conf",
+                      action="store", dest="train_conf",
+                      help="network config")
+    parser.add_option("-i", "--data",
+                      action="store", dest="data_file",
+                      help="image list")
+    parser.add_option("-w", "--model",
+                      action="store", dest="model_path",
+                      default=None, help="model path")
+    parser.add_option("-o", "--output_dir",
+                      action="store", dest="output_dir",
+                      default="output", help="output path")
+    parser.add_option("-m", "--mean", action="store",
+                      dest="mean", default=None,
+                      help="mean file.")
+    parser.add_option("-p", "--multi_crop", action="store_true",
+                      dest="multi_crop", default=False,
+                      help="Wether to use multiple crops on image.")
+    parser.add_option("-l", "--output_layer", action="store",
+                      dest="output_layer", default=None,
+                      help="--job=extract, specify layers to extract "\
+                           "features, --job=predict, specify layer of "
+                           "classification probability, output in resnet.py.")
+    return parser.parse_args()
+
+def main():
+    """
+    1. parse input arguments.
+    2. predicting or extract features according job type.
+    """
+    options, args = option_parser()
+    obj = ImageClassifier(options.train_conf,
+                        options.model_path,
+                        mean_file=options.mean,
+                        output_layer=options.output_layer,
+                        oversample=options.multi_crop)
+    if options.job_type == "predict":
+        obj.predict(options.data_file)
+
+    elif options.job_type == "extract":
+        obj.extract(options.data_file,
+                    options.output_dir)
+
+if __name__ == '__main__':
+    main()

demo/model_zoo/resnet/example/.gitignore

+*image_list_provider_copy_1.py

demo/model_zoo/resnet/example/__init__.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+

demo/model_zoo/resnet/example/image_list_provider.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.utils.image_util import *
+from paddle.trainer.PyDataProvider2 import *
+
+
+def hook(settings, image_size, crop_size, color, file_list,
+         is_train, **kwargs):
+    """
+    Description: Init with a list of data file
+    file_list is the name list of input files.
+    kwargs["load_data_args"] is the value of 'load_data_args'
+    which can be set in config.
+    Each args is separated by a column.
+    image_size: the crop image size.
+    mean_meta: the path of the meta file to store the mean image.
+    mean_value: can be mean value, not a file.
+                can not set mean_meta and mean_value at the same time.
+    color: 'color' means a color image. Otherwise, it means a gray image.
+    is_train: whether the data provider is used for training.
+              Data argumentation might be different for training and testing.
+    """
+    settings.img_size = image_size
+    settings.crop_size = crop_size
+    settings.mean_img_size = settings.crop_size
+    settings.color = color  # default is color
+    settings.is_train = is_train
+
+    settings.is_swap_channel = kwargs.get('swap_channel', None)
+    if settings.is_swap_channel is not None:
+        settings.swap_channel = settings.is_swap_channel
+        settings.is_swap_channel = True
+
+    if settings.color:
+        settings.img_input_size = settings.crop_size * settings.crop_size * 3
+    else:
+        settings.img_input_size = settings.crop_size * settings.crop_size
+
+    settings.file_list = file_list
+    settings.mean_meta = kwargs.get('mean_meta', None)
+    settings.mean_value = kwargs.get('mean_value', None)
+    # can not specify both mean_meta and mean_value.
+    assert not (settings.mean_meta and settings.mean_value)
+    if not settings.mean_meta:
+        settings.mean_value = kwargs.get('mean_value')
+        sz = settings.crop_size * settings.crop_size
+        settings.img_mean = np.zeros(sz * 3, dtype=np.single)
+        for idx, value in enumerate(settings.mean_value):
+            settings.img_mean[idx * sz: (idx + 1) * sz] = value
+        settings.img_mean = settings.img_mean.reshape(3, settings.crop_size,
+                                                      settings.crop_size)
+
+    else:
+        settings.img_mean = load_meta(settings.mean_meta,
+                                      settings.mean_img_size,
+                                      settings.crop_size, settings.color)
+
+    settings.input_types = [
+        dense_vector(settings.img_input_size),  # image feature
+        integer_value(1)]  # labels
+
+    settings.logger.info('Image short side: %s', settings.img_size)
+    settings.logger.info('Crop size: %s', settings.crop_size)
+    settings.logger.info('Meta path: %s', settings.mean_meta)
+    if settings.is_swap_channel:
+        settings.logger.info('swap channel: %s', settings.swap_channel)
+    settings.logger.info('DataProvider Initialization finished')
+
+
+@provider(init_hook=hook, should_shuffle=False)
+def processData(settings, file_list):
+    """
+    The main function for loading data.
+    Load the batch, iterate all the images and labels in this batch.
+    file_name: the batch file name.
+    """
+    img_path, lab = file_list.strip().split(' ')
+    img = Image.open(img_path)
+    img.load()
+    img = img.resize((settings.img_size, settings.img_size), Image.ANTIALIAS)
+    img = np.array(img).astype(np.float32)
+    if len(img.shape) == 3:
+        img = np.swapaxes(img, 1, 2)
+        img = np.swapaxes(img, 1, 0)
+    # swap channel
+    if settings.is_swap_channel:
+        img = img[settings.swap_channel, :, :]
+    img_feat = preprocess_img(img,
+                              settings.img_mean,
+                              settings.crop_size,
+                              settings.is_train,
+                              settings.color)
+    yield img_feat.tolist(), int(lab.strip())

demo/model_zoo/resnet/example/test.list

+example/dog.jpg 0
+example/cat.jpg 0

demo/model_zoo/resnet/extract_fea_c++.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+#set names of layer which you want to extract feature
+#in Outputs() of resnet.py 
+#like: Outputs("res5_3_branch2c_conv", "res5_3_branch2c_bn")
+layer_num=50
+configure=./resnet.py
+model_path=./model/resnet_$layer_num
+fea_dir=fea_output
+#Output is text file.
+#Each line is one sample's features.
+#If you set N layer names in Outputs()
+#each line contains N features sperated by ";". 
+
+# create model list file.
+model_list=./model.list
+touch $model_list | echo $model_path > $model_list
+
+paddle train \
+  --local=true \
+  --job=test \
+  --config=$configure \
+  --model_list=$model_list \
+  --use_gpu=1 \
+  --predict_output_dir=$fea_dir \
+  --config_args=is_test=1,layer_num=$layer_num

demo/model_zoo/resnet/extract_fea_py.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+python classify.py \
+     --job=extract \
+     --conf=resnet.py \
+     --mean=model/mean_meta_224/mean.meta \
+     --model=model/resnet_50 \
+     --data=./example/test.list \
+     --output_layer="res5_3_branch2c_conv,res5_3_branch2c_bn" \
+     --output_dir=features 

demo/model_zoo/resnet/get_model.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+DIR="$( cd "$(dirname "$0")" ; pwd -P )"
+cd $DIR
+
+mkdir model
+cd model
+
+echo "Downloading ResNet models..."
+
+for file in resnet_50.tar.gz resnet_101.tar.gz resnet_152.tar.gz mean_meta_224.tar.gz 
+do 
+  # following is the google drive address
+  # you can also directly download from https://pan.baidu.com/s/1o8q577s
+  wget https://www.googledrive.com/host/0B7Q8d52jqeI9ejh6Q1RpMTFQT1k/imagenet/$file --no-check-certificate
+  tar -xvf $file 
+  rm $file
+done
+
+echo "Done."

demo/model_zoo/resnet/load_feature.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import sys
+import cPickle
+import logging
+
+logging.basicConfig(format='[%(levelname)s %(asctime)s %(filename)s:%(lineno)s] %(message)s')
+logging.getLogger().setLevel(logging.INFO)
+
+def load_feature_c(file):
+    """
+    Load feature extracted by C++ interface.
+    Return a list.
+    file: feature file.
+    """
+    features = []
+    f = open(file, 'r')
+    for line in f:
+        sample = []
+        for slot in line.strip().split(";"): 
+            fea = [float(val) for val in slot.strip().split()] 
+            if fea:
+                sample.append(fea)
+        features.append(sample)
+    f.close()
+    return features
+
+def load_feature_py(feature_dir):
+    """
+    Load feature extracted by python interface.
+    Return a dictionary.
+    feature_dir: directory of feature file.
+    """
+    file_list = os.listdir(feature_dir)
+    file_list = [os.path.join(feature_dir, f) for f in file_list]
+    features = {}
+    for file_name in file_list:
+        with open(file_name, 'rb') as f:
+            feature = cPickle.load(f)
+            features.update(feature)
+            logging.info('Load feature file %s', file_name)
+    return features
+
+if __name__ == '__main__':
+    print load_feature_py(sys.argv[1]) 
+    #print load_feature_c(sys.argv[1]) 

demo/model_zoo/resnet/net_diagram.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+:'
+Visual deep residual network
+1. Using make_model_diagram.py to generate dot file.
+2. Using graphviz to convert dot file.
+
+Usage:
+./net_diagram.sh
+'
+
+DIR="$( cd "$(dirname "$0")" ; pwd -P )"
+cd $DIR
+
+img_type=png
+img_fileprefix=ResNet_50
+conf_filename=resnet.py
+dot_filename=ResNet_50.dot
+config_str="layer_num=50,data_provider=0"
+
+python -m paddle.utils.make_model_diagram $conf_filename $dot_filename $config_str
+
+# If you have installed graphviz, running like this:
+# dot -Tpng -o ResNet.png ResNet.dot

demo/model_zoo/resnet/predict.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+python classify.py \
+     --job=predict \
+     --conf=resnet.py\
+     --model=model/resnet_50 \
+     --multi_crop \
+     --data=./example/test.list

demo/model_zoo/resnet/resnet.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer_config_helpers import *
+
+"""
+paper: https://arxiv.org/abs/1512.03385
+"""
+is_test = get_config_arg("is_test", bool, False)
+is_predict = get_config_arg("is_predict", bool, False)
+data_provider = get_config_arg("data_provider", bool, True)
+layer_num = get_config_arg("layer_num", int, 50)
+
+if not is_predict and data_provider:
+    train_list = 'train.list' if not is_test else None
+    # mean.meta is mean file of ImageNet dataset.
+    # mean.meta size : 3 x 224 x 224.
+    # If you use three mean value, set like:
+    # "mean_value:103.939,116.779,123.68;"
+    args={
+        'mean_meta': "model/mean_meta_224/mean.meta",
+        'image_size': 224, 'crop_size': 224,
+        'color': True,'swap_channel:': [2, 1, 0]}
+    define_py_data_sources2(train_list,
+                           'example/test.list',
+                           module="example.image_list_provider",
+                           obj="processData",
+                           args=args)
+
+batch_size = 1
+learning_rate = 0.1 / batch_size
+momentum = 0.9
+weight_decay = 0.0001 * batch_size
+default_momentum(momentum)
+default_decay_rate(weight_decay)
+
+Settings(
+    algorithm='sgd',
+    batch_size=batch_size,
+    learning_rate=learning_rate,
+
+    # set the appropriate parameters according your schedule
+    learning_method='momentum',
+    learning_rate_decay_a=0.5,
+    learning_rate_decay_b=1200000 * 10,
+    learning_rate_schedule="discexp",
+)
+
+
+def conv_bn_layer(name, input, filter_size, num_filters,
+                  stride, padding, channels=None,
+                  active_type=ReluActivation()):
+    """
+    A wrapper for conv layer with batch normalization layers.
+    Note:
+    conv layer has no activation.
+    """
+
+    tmp = img_conv_layer(name=name + "_conv",
+                         input=input,
+                         filter_size=filter_size,
+                         num_channels=channels,
+                         num_filters=num_filters,
+                         stride=stride,
+                         padding=padding,
+                         act=LinearActivation(),
+                         bias_attr=False)
+    return batch_norm_layer(name=name + "_bn",
+                            input=tmp,
+                            act=active_type,
+                            use_global_stats=is_test)
+
+
+def bottleneck_block(name, input, num_filters1, num_filters2):
+    """
+    A wrapper for bottlenect building block in ResNet.
+    Last conv_bn_layer has no activation.
+    Addto layer has activation of relu.
+    """
+    last_name = conv_bn_layer(name=name + '_branch2a',
+                              input=input,
+                              filter_size=1,
+                              num_filters=num_filters1,
+                              stride=1,
+                              padding=0)
+    last_name = conv_bn_layer(name=name + '_branch2b',
+                              input=last_name,
+                              filter_size=3,
+                              num_filters=num_filters1,
+                              stride=1,
+                              padding=1)
+    last_name = conv_bn_layer(name=name + '_branch2c',
+                              input=last_name,
+                              filter_size=1,
+                              num_filters=num_filters2,
+                              stride=1,
+                              padding=0,
+                              active_type=LinearActivation())
+
+    return addto_layer(name=name + "_addto",
+                       input=[input, last_name],
+                       act=ReluActivation())
+
+
+def mid_projection(name, input, num_filters1, num_filters2, stride=2):
+    """
+    A wrapper for middile projection in ResNet.
+    projection shortcuts are used for increasing dimensions,
+    and other shortcuts are identity
+    branch1: projection shortcuts are used for increasing
+    dimensions, has no activation.
+    branch2x: bottleneck building block, shortcuts are identity.
+    """
+    # stride = 2
+    branch1 = conv_bn_layer(name=name + '_branch1',
+                            input=input,
+                            filter_size=1,
+                            num_filters=num_filters2,
+                            stride=stride,
+                            padding=0,
+                            active_type=LinearActivation())
+
+    last_name = conv_bn_layer(name=name + '_branch2a',
+                              input=input,
+                              filter_size=1,
+                              num_filters=num_filters1,
+                              stride=stride,
+                              padding=0)
+    last_name = conv_bn_layer(name=name + '_branch2b',
+                              input=last_name,
+                              filter_size=3,
+                              num_filters=num_filters1,
+                              stride=1,
+                              padding=1)
+
+    last_name = conv_bn_layer(name=name + '_branch2c',
+                              input=last_name,
+                              filter_size=1,
+                              num_filters=num_filters2,
+                              stride=1,
+                              padding=0,
+                              active_type=LinearActivation())
+
+    return addto_layer(name=name + "_addto",
+                       input=[branch1, last_name],
+                       act=ReluActivation())
+
+
+def deep_res_net(res2_num=3, res3_num=4, res4_num=6, res5_num=3):
+    """
+    A wrapper for 50,101,152 layers of ResNet.
+    res2_num: number of blocks stacked in conv2_x
+    res3_num: number of blocks stacked in conv3_x
+    res4_num: number of blocks stacked in conv4_x
+    res5_num: number of blocks stacked in conv5_x
+    """
+    # For ImageNet
+    # conv1: 112x112
+    img = data_layer(name='input', size=224 * 224 * 3)
+    tmp = conv_bn_layer("conv1", img,
+                        filter_size=7,
+                        channels=3,
+                        num_filters=64,
+                        stride=2,
+                        padding=3)
+    tmp = img_pool_layer(name="pool1", input=tmp, pool_size=3, stride=2)
+
+    # conv2_x: 56x56
+    tmp = mid_projection(name="res2_1",
+                         input=tmp,
+                         num_filters1=64,
+                         num_filters2=256,
+                         stride=1)
+    for i in xrange(2, res2_num + 1, 1):
+        tmp = bottleneck_block(name="res2_" + str(i),
+                               input=tmp,
+                               num_filters1=64,
+                               num_filters2=256)
+
+    # conv3_x: 28x28
+    tmp = mid_projection(name="res3_1",
+                         input=tmp,
+                         num_filters1=128,
+                         num_filters2=512)
+    for i in xrange(2, res3_num + 1, 1):
+        tmp = bottleneck_block(name="res3_" + str(i),
+                               input=tmp, num_filters1=128,
+                               num_filters2=512)
+
+    # conv4_x: 14x14
+    tmp = mid_projection(name="res4_1", input=tmp,
+                         num_filters1=256, num_filters2=1024)
+    for i in xrange(2, res4_num + 1, 1):
+        tmp = bottleneck_block(name="res4_" + str(i),
+                               input=tmp,
+                               num_filters1=256,
+                               num_filters2=1024)
+
+    # conv5_x: 7x7
+    tmp = mid_projection(name="res5_1", input=tmp,
+                         num_filters1=512, num_filters2=2048)
+    for i in xrange(2, res5_num + 1, 1):
+        tmp = bottleneck_block(name="res5_" + str(i),
+                               input=tmp, num_filters1=512,
+                               num_filters2=2048)
+
+    tmp = img_pool_layer(name='avgpool',
+                         input=tmp,
+                         pool_size=7,
+                         stride=1,
+                         pool_type=AvgPooling())
+
+    output = fc_layer(name='output',
+                      input=tmp,
+                      size=1000,
+                      act=SoftmaxActivation())
+
+    if not is_predict:
+        classification_cost(input=output, label=data_layer(name='label',
+                                                           size=1))
+
+
+def res_net_50():
+    deep_res_net(3, 4, 6, 3)
+
+
+def res_net_101():
+    deep_res_net(3, 4, 23, 3)
+
+
+def res_net_152():
+    deep_res_net(3, 8, 36, 3)
+
+
+if not is_predict:
+    Inputs("input", "label")
+else:
+    Inputs("input")
+# Outputs("cost-softmax" if not is_predict else "output")
+Outputs("res5_3_branch2c_conv", "res5_3_branch2c_bn")
+
+if layer_num == 50:
+    res_net_50()
+elif layer_num == 101:
+    res_net_101()
+elif layer_num == 152:
+    res_net_152()
+else:
+    print("Wrong layer number.")

demo/quick_start/.gitignore

+*.pyc
+data/dict.txt
+data/dict_all.txt
+data/labels.list
+data/mosesdecoder-master/
+data/reviews_Electronics_5.json.gz
+data/test.list
+data/test.txt
+data/train.list
+data/train.txt
+dataprovider_copy_1.py
+train.log
+output

demo/quick_start/data/get_data.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+DIR="$( cd "$(dirname "$0")" ; pwd -P )"
+cd $DIR
+
+echo "Downloading Amazon Electronics reviews data..."
+# http://jmcauley.ucsd.edu/data/amazon/
+wget http://snap.stanford.edu/data/amazon/productGraph/categoryFiles/reviews_Electronics_5.json.gz
+
+echo "Downloading mosesdecoder..."
+#https://github.com/moses-smt/mosesdecoder
+wget https://github.com/moses-smt/mosesdecoder/archive/master.zip
+
+unzip master.zip
+rm master.zip
+echo "Done."

demo/quick_start/data/pred.list

+./data/pred.txt

demo/quick_start/data/pred.txt

+the device is cute , but that 's just about all that 's good. the specs are what you 'd expect : it 's a wifi mic , with some noise filter options. the app has the option to upload your baby 's name and photo , which is a cutesy touch. but the app is otherwise unstable and useless unless you upgrade for $ 60 / year.set up involves downloading the app , turning on the mic , switching your phone to the wifi network of the mic , telling the app your wifi settings , switching your wifi back to your home router. the app is then directly connected to your mic.the app is adware ! the main screen says " cry notifications on / off : upgrade to evoz premium and receive a text message of email when your baby is crying " .but the adware points out an important limitation , this monitor is only intended to be used from your home network. if you want to access it remotely , get a webcam. this app would make a lot more sense of the premium features were included with the hardware .
+don 't be fooled by my one star rating. if there was a zero , i would have selected it. this product was a waste of my money.it has never worked like the company said it supposed to. i only have one device , an iphone 4gs. after charging the the iphone mid way , the i.sound portable power max 16,000 mah is completely drained. the led light no longer lit up. when plugging the isound portable power max into a wall outlet to charge , it would charge for about 20-30 minutes and then all four battery led indicator lit up showing a full charge. i would leave it on to charge for the full 8 hours or more but each time with the same result upon using. don 't buy this thing. put your money to good use elsewhere .

demo/quick_start/dataprovider_bow.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer.PyDataProvider2 import *
+
+# id of the word not in dictionary
+UNK_IDX = 0
+
+# initializer is called by the framework during initialization.
+# It allows the user to describe the data types and setup the
+# necessary data structure for later use.
+# `settings` is an object. initializer need to properly fill settings.input_types.
+# initializer can also store other data structures needed to be used at process().
+# In this example, dictionary is stored in settings.
+# `dictionay` and `kwargs` are arguments passed from trainer_config.lr.py
+def initializer(settings, dictionary, **kwargs):
+    # Put the word dictionary into settings
+    settings.word_dict = dictionary
+
+    # setting.input_types specifies what the data types the data provider
+    # generates.
+    settings.input_types = [
+        # The first input is a sparse_binary_vector,
+        # which means each dimension of the vector is either 0 or 1. It is the
+        # bag-of-words (BOW) representation of the texts.
+        sparse_binary_vector(len(dictionary)),
+        # The second input is an integer. It represents the category id of the
+        # sample. 2 means there are two labels in the dataset.
+        # (1 for positive and 0 for negative)
+        integer_value(2)]
+
+# Delaring a data provider. It has an initializer 'data_initialzer'.
+# It will cache the generated data of the first pass in memory, so that
+# during later pass, no on-the-fly data generation will be needed.
+# `setting` is the same object used by initializer()
+# `file_name` is the name of a file listed train_list or test_list file given
+# to define_py_data_sources2(). See trainer_config.lr.py.
+@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM)
+def process(settings, file_name):
+    # Open the input data file.
+    with open(file_name, 'r') as f:
+        # Read each line.
+        for line in f:
+            # Each line contains the label and text of the comment, separated by \t.
+            label, comment = line.strip().split('\t')
+
+            # Split the words into a list.
+            words = comment.split()
+
+            # convert the words into a list of ids by looking them up in word_dict.
+            word_vector = [settings.word_dict.get(w, UNK_IDX) for w in words]
+
+            # Return the features for the current comment. The first is a list
+            # of ids representing a 0-1 binary sparse vector of the text,
+            # the second is the integer id of the label.
+            yield word_vector, int(label)
+
+
+def predict_initializer(settings, dictionary, **kwargs):
+    settings.word_dict = dictionary
+    settings.input_types = [
+        sparse_binary_vector(len(dictionary))
+    ]
+
+# Declaring a data provider for prediction. The difference with process
+# is that label is not generated.
+@provider(init_hook=predict_initializer)
+def process_predict(settings, file_name):
+    with open(file_name, 'r') as f:
+        for line in f:
+            comment = line.strip()
+            word_vector = [settings.word_dict.get(w, UNK_IDX) for w in comment]
+            yield word_vector

demo/quick_start/dataprovider_emb.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer.PyDataProvider2 import *
+
+UNK_IDX = 0
+
+def initializer(settings, dictionary, **kwargs):
+    settings.word_dict = dictionary
+    settings.input_types = [
+        # Define the type of the first input as sequence of integer.
+        # The value of the integers range from 0 to len(dictrionary)-1
+        integer_value_sequence(len(dictionary)),
+        # Define the second input for label id
+        integer_value(2)]
+
+
+@provider(init_hook=initializer, cache=CacheType.CACHE_PASS_IN_MEM)
+def process(settings, file_name):
+    with open(file_name, 'r') as f:
+        for line in f:
+            label, comment = line.strip().split('\t')
+            words = comment.split()
+            word_slot = [settings.word_dict.get(w, UNK_IDX) for w in words]
+            yield word_slot, int(label)
+
+
+def predict_initializer(settings, dictionary, **kwargs):
+    settings.word_dict = dictionary
+    settings.input_types = [
+        integer_value(len(dictionary), seq_type=SequenceType.SEQUENCE)
+    ]
+
+
+@provider(init_hook=predict_initializer)
+def process_predict(settings, file_name):
+    with open(file_name, 'r') as f:
+        for line in f:
+            comment = line.strip()
+            word_slot = [settings.word_dict.get(w, UNK_IDX) for w in comment]
+            yield word_slot

demo/quick_start/predict.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+#cfg=trainer_config.lr.py
+#cfg=trainer_config.emb.py
+#cfg=trainer_config.cnn.py
+cfg=trainer_config.lstm.py
+model="output/pass-00003"
+paddle train \
+    --config=$cfg \
+    --use_gpu=false \
+    --job=test \
+    --init_model_path=$model \
+    --config_args=is_predict=1 \
+    --predict_output_dir=. \
+
+mv rank-00000 result.txt

demo/quick_start/preprocess.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+'''
+1. remove HTML before tokensizing 
+2. pos sample : rating score 5; neg sample: rating score 1-2.
+3. size of pos : neg = 1:1.
+4. size of testing set = min(25k, len(all_data) * 0.1), others is traning set.
+5. distinct train set and test set.
+
+Usage:
+    python preprocess.py -i data_file [random seed]
+'''
+
+import sys,os
+import re
+import operator
+import gzip,math
+import random
+import numpy as np
+from bs4 import BeautifulSoup
+from subprocess import Popen, PIPE
+from optparse import OptionParser
+
+def parse(path):
+    """
+    Open .gz file.
+    """
+    g = gzip.open(path, 'r')
+    for l in g:
+        yield eval(l)
+
+def clean(review):
+    """
+    Clean input review: remove HTML, convert words to lower cases.
+    """
+    # Remove HTML
+    review_text = BeautifulSoup(review, "html.parser").get_text()
+
+    # Convert words to lower case
+    review_text = review_text.lower()
+    return review_text
+
+def tokenize(sentences):
+    """
+    Use tokenizer.perl to tokenize input sentences.
+    tokenizer.perl is tool of Moses.
+    sentences : a list of input sentences.
+    return: a list of processed text.
+    """
+    dir = './data/mosesdecoder-master/scripts/tokenizer/tokenizer.perl'
+    tokenizer_cmd = [dir, '-l', 'en', '-q', '-']
+    assert isinstance(sentences, list)
+    text = "\n".join(sentences)
+    tokenizer = Popen(tokenizer_cmd, stdin=PIPE, stdout=PIPE)
+    tok_text, _ = tokenizer.communicate(text)
+    toks = tok_text.split('\n')[:-1]
+    return toks
+
+def create_dict(data, data_dir):
+    """
+    Create dictionary based on data, and saved in data_dir/dict.txt.
+    The first line is unk \t -1. 
+    data: list, input data.
+    data_dir: path to save dict.
+    """
+    word_count = {}
+    for seq in data:
+        try:
+            for w in seq.lower().split():
+                if w not in word_count:
+                    word_count[w] = 1
+                else:
+                    word_count[w] += 1
+        except:
+            sys.stderr.write(seq+"\tERROR\n")
+    f = open(os.path.join(data_dir, 'dict.txt'), 'w')
+    f.write('%s\t%s\n' % ('unk', '-1'))
+    for k, v in sorted(word_count.items(), key=operator.itemgetter(1),\
+                      reverse=True):
+        f.write('%s\t%s\n' % (k, v))
+    f.close()
+
+def save_data(data, data_dir, prefix = ""):
+    file_name = os.path.join(data_dir, "%s.txt" % (prefix))
+    file(file_name,'w').write('\n'.join(data)+'\n')
+    file(os.path.join(data_dir, prefix+'.list'),'w').write('%s\n' % file_name)
+
+def split_data(raw_txt):
+    """
+    Extract positive and negative sample.
+    """
+    pos = []
+    neg = []
+    count = 0
+    dup_cnt = 0
+    sys.stderr.write("extract raw data")
+    for l in raw_txt:
+        rating = l["overall"]
+        text = clean(l["reviewText"])
+        if rating == 5.0 and text:
+            pos.append(text)
+        if rating < 3.0 and text:
+            neg.append(text)
+        count += 1
+        if count % 20000==0:
+            sys.stderr.write(".")
+    sys.stderr.write("\n")
+    return pos, neg
+
+def preprocess(pos_in, neg_in, data_dir, rand_seed):
+    # tokenize
+    sys.stderr.write("tokenize...\n")
+    tmppos = tokenize(pos_in)
+    tmpneg = tokenize(neg_in)
+    cnt = len(tmppos) + len(tmpneg)
+
+    # unique smaples
+    tmppos = list(set(tmppos))
+    tmpneg = list(set(tmpneg))
+    dup_cnt = cnt - len(tmppos) - len(tmpneg)
+    sys.stderr.write("\ntotal size of data set: %d, duplicate data: %d\n" % (cnt, dup_cnt))
+
+    # keep same size of positive and negative sample
+    min_len = min(len(tmppos), len(tmpneg))
+    tmppos = tmppos[0:min_len]
+    tmpneg = tmpneg[0:min_len]
+
+    # creat dictionary
+    sys.stderr.write("create dict with train and test data...\n")
+    all_data = tmppos + tmpneg
+    create_dict(all_data, data_dir)
+
+    # split into train set and test set
+    sys.stderr.write("split data...\n")
+    pos = ["1\t"+i for i in tmppos]
+    neg = ["0\t"+i for i in tmpneg]
+    random.seed(rand_seed)
+    random.shuffle(pos)
+    random.shuffle(neg)
+
+    # split into test set and train set
+    test_len = min(12500, int(min_len * 0.1))
+    test = pos[0:test_len] + neg[0:test_len]
+    train = pos[test_len:] + neg[test_len:]
+
+    # save data
+    sys.stderr.write("save data...\n")
+    save_data(train, data_dir, prefix = 'train')
+    save_data(test, data_dir, prefix = 'test')
+    file(os.path.join(data_dir,'labels.list'),'w').write('neg\t0\npos\t1\n')
+
+def option_parser():
+    parser = OptionParser(usage="usage: python preprcoess.py "\
+                                "-i data_path [options]")
+    parser.add_option("-i", "--data", action="store",
+                      dest="input", help="Input data path.")
+    parser.add_option("-s", "--seed", action="store",
+                      dest="seed", default=1024,
+                      help="Set random seed.")
+    return parser.parse_args()
+
+def main():
+    reload(sys)
+    sys.setdefaultencoding('utf-8')
+    options, args = option_parser()
+    data=options.input
+    seed=options.seed
+    data_dir = os.path.dirname(data)
+    pos, neg = split_data(parse(data))
+    preprocess(pos, neg, data_dir, seed)
+    sys.stderr.write("Done.\n")
+
+if __name__ == '__main__':
+    main()

demo/quick_start/preprocess.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+python preprocess.py -i data/reviews_Electronics_5.json.gz
+
+# use 30k dict
+mv data/dict.txt data/dict_all.txt
+cat data/dict_all.txt | head -n 30001 > data/dict.txt

demo/quick_start/requirements.txt

+beautifulsoup4

demo/quick_start/train.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+cfg=trainer_config.lr.py
+#cfg=trainer_config.emb.py
+#cfg=trainer_config.cnn.py
+#cfg=trainer_config.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'

demo/quick_start/trainer_config.cnn.py

+# edit-mode: -*- python -*-
+
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer_config_helpers import *
+
+dict_file = "./data/dict.txt"
+word_dict = dict()
+with open(dict_file, 'r') as f:
+    for i, line in enumerate(f):
+        w = line.strip().split()[0]
+        word_dict[w] = i
+
+is_predict = get_config_arg('is_predict', bool, False)
+trn = 'data/train.list' if not is_predict else None
+tst = 'data/test.list' if not is_predict else 'data/pred.list'
+process = 'process' if not is_predict else 'process_predict'
+define_py_data_sources2(train_list=trn,
+                        test_list=tst,
+                        module="dataprovider_emb",
+                        obj=process,
+                        args={"dictionary": word_dict})
+
+batch_size = 128 if not is_predict else 1
+settings(
+    batch_size=batch_size,
+    learning_rate=2e-3,
+    learning_method=AdamOptimizer(),
+    regularization=L2Regularization(8e-4),
+    gradient_clipping_threshold=25
+)
+
+data = data_layer(name="word", size=len(word_dict))
+embedding = embedding_layer(input=data, size=128)
+conv = sequence_conv_pool(input=embedding, context_len=3, hidden_size=512)
+output = fc_layer(input=conv, size=2, act=SoftmaxActivation())
+if is_predict:
+    maxid = maxid_layer(output)
+    outputs([maxid, output])
+else:
+    label = data_layer(name="label", size=2)
+    cls = classification_cost(input=output, label=label)
+    outputs(cls)

demo/quick_start/trainer_config.emb.py

+# edit-mode: -*- python -*-
+
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer_config_helpers import *
+
+dict_file = "./data/dict.txt"
+word_dict = dict()
+with open(dict_file, 'r') as f:
+    for i, line in enumerate(f):
+        w = line.strip().split()[0]
+        word_dict[w] = i
+
+is_predict = get_config_arg('is_predict', bool, False)
+trn = 'data/train.list' if not is_predict else None
+tst = 'data/test.list' if not is_predict else 'data/pred.list'
+process = 'process' if not is_predict else 'process_predict'
+define_py_data_sources2(train_list=trn,
+                        test_list=tst,
+                        module="dataprovider_emb",
+                        obj=process,
+                        args={"dictionary": word_dict})
+
+batch_size = 128 if not is_predict else 1
+settings(
+    batch_size=batch_size,
+    learning_rate=2e-3,
+    learning_method=AdamOptimizer()
+)
+
+data = data_layer(name="word", size=len(word_dict))
+embedding = embedding_layer(input=data, size=128)
+avg = pooling_layer(input=embedding, pooling_type=AvgPooling())
+output = fc_layer(input=avg, size=2, act=SoftmaxActivation())
+if is_predict:
+    maxid = maxid_layer(output)
+    outputs([maxid, output])
+else:
+    label = data_layer(name="label", size=2)
+    cls = classification_cost(input=output, label=label)
+    outputs(cls)

demo/quick_start/trainer_config.lr.py

+# edit-mode: -*- python -*-
+
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer_config_helpers import *
+
+dict_file = "./data/dict.txt"
+word_dict = dict()
+with open(dict_file, 'r') as f:
+    for i, line in enumerate(f):
+        w = line.strip().split()[0]
+        word_dict[w] = i
+
+is_predict = get_config_arg('is_predict', bool, False)
+trn = 'data/train.list' if not is_predict else None
+tst = 'data/test.list' if not is_predict else 'data/pred.list'
+process = 'process' if not is_predict else 'process_predict'
+
+# define the data sources for the model.
+# We need to use different process for training and prediction.
+# For training, the input data includes both word IDs and labels.
+# For prediction, the input data only includs word Ids.
+define_py_data_sources2(train_list=trn,
+                        test_list=tst,
+                        module="dataprovider_bow",
+                        obj=process,
+                        args={"dictionary": word_dict})
+
+batch_size = 128 if not is_predict else 1
+settings(
+    batch_size=batch_size,
+    learning_rate=2e-3,
+    learning_method=AdamOptimizer(),
+    regularization=L2Regularization(8e-4),
+    gradient_clipping_threshold=25
+)
+
+# Define the data for text features. The size of the data layer is the number
+# of words in the dictionary.
+data = data_layer(name="word", size=len(word_dict))
+
+# Define a fully connected layer with logistic activation.
+# (also called softmax activation).
+output = fc_layer(input=data, size=2, act=SoftmaxActivation())
+
+if not is_predict:
+    # For training, we need label and cost
+
+    # define the category id for each example.
+    # The size of the data layer is the number of labels.
+    label = data_layer(name="label", size=2)
+
+    # Define cross-entropy classification loss and error.
+    classification_cost(input=output, label=label)
+    cls = classification_cost(input=output, label=label)
+    outputs(cls)
+else:
+    # For prediction, no label is needed. We need to output
+    # We need to output classification result, and class probabilities.
+    maxid = maxid_layer(output)
+    outputs([maxid, output])

demo/quick_start/trainer_config.lstm.py

+# edit-mode: -*- python -*-
+
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from paddle.trainer_config_helpers import *
+
+dict_file = "./data/dict.txt"
+word_dict = dict()
+with open(dict_file, 'r') as f:
+    for i, line in enumerate(f):
+        w = line.strip().split()[0]
+        word_dict[w] = i
+
+is_predict = get_config_arg('is_predict', bool, False)
+trn = 'data/train.list' if not is_predict else None
+tst = 'data/test.list' if not is_predict else 'data/pred.list'
+process = 'process' if not is_predict else 'process_predict'
+define_py_data_sources2(train_list=trn,
+                        test_list=tst,
+                        module="dataprovider_emb",
+                        obj=process,
+                        args={"dictionary": word_dict})
+
+batch_size = 128 if not is_predict else 1
+settings(
+    batch_size=batch_size,
+    learning_rate=2e-3,
+    learning_method=AdamOptimizer(),
+    regularization=L2Regularization(8e-4),
+    gradient_clipping_threshold=25
+)
+
+bias_attr = ParamAttr(initial_std=0.,l2_rate=0.)
+
+data = data_layer(name="word", size=len(word_dict))
+emb = embedding_layer(input=data, size=128)
+fc = fc_layer(input=emb, size=512,
+              act=LinearActivation(),
+              bias_attr=bias_attr,
+              layer_attr=ExtraAttr(drop_rate=0.1))
+lstm = lstmemory(input=fc, act=TanhActivation(),
+                 bias_attr=bias_attr,
+                 layer_attr=ExtraAttr(drop_rate=0.25))
+lstm_last = pooling_layer(input=lstm, pooling_type=MaxPooling())
+output = fc_layer(input=lstm_last, size=2,
+                  bias_attr=bias_attr,
+                  act=SoftmaxActivation())
+if is_predict:
+    maxid = maxid_layer(output)
+    outputs([maxid, output])
+else:
+    label = data_layer(name="label", size=2)
+    cls = classification_cost(input=output, label=label)
+    outputs(cls)

demo/recommendation/.gitignore

+log.txt
+data/meta.bin
+data/ml-1m
+data/ratings.dat.train
+data/ratings.dat.test
+data/train.list
+data/test.list
+dataprovider_copy_1.py
+*.pyc

demo/recommendation/common_utils.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from paddle.trainer.PyDataProvider2 import *
+
+
+def meta_to_header(meta, name):
+    metas = meta[name]['__meta__']['raw_meta']
+    for each_meta in metas:
+        if each_meta['type'] == 'id':
+            yield integer_value(each_meta['max'])
+        elif each_meta['type'] == 'embedding':
+            is_seq = each_meta['seq'] == 'sequence'
+            yield integer_value(len(each_meta['dict']),
+                                seq_type=SequenceType.SEQUENCE if is_seq
+                                else SequenceType.NO_SEQUENCE)
+        elif each_meta['type'] == 'one_hot_dense':
+            yield dense_vector(len(each_meta['dict']))

demo/recommendation/data/config.json

+{
+  "user": {
+    "file": {
+      "name": "users.dat",
+      "delimiter": "::"
+    },
+    "fields": ["id", "gender", "age", "occupation"]
+  },
+  "movie": {
+    "file": {
+      "name": "movies.dat",
+      "delimiter": "::"
+    },
+    "fields": ["id", "title", "genres"]
+  }
+}
+

demo/recommendation/data/config_generator.py

+#!/bin/env python2
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+config_generator.py
+
+Usage:
+    ./config_generator.py <config_file> [--output_format=<output_format>]
+    ./config_generator.py -h | --help
+
+Options:
+    -h --help                           Show this screen.
+    --output_format=<output_format>     Output Config format(json or yaml) [default: json].
+"""
+
+import json
+import docopt
+import copy
+
+DEFAULT_FILE = {
+    "type": "split",
+    "delimiter": ","
+}
+
+DEFAULT_FIELD = {
+    "id": {
+        "type": "id"
+    },
+    "gender": {
+        "name": "gender",
+        "type": "embedding",
+        "dict": {
+            "type": "char_based"
+        }
+    },
+    "age": {
+        "name": "age",
+        "type": "embedding",
+        "dict": {
+            "type": "whole_content",
+            "sort": True
+        }
+    },
+    "occupation": {
+        "name": "occupation",
+        "type": "embedding",
+        "dict": {
+            "type": "whole_content",
+            "sort": "true"
+        }
+    },
+    "title": {
+        "regex": {
+            "pattern": r"^(.*)\((\d+)\)$",
+            "group_id": 1,
+            "strip": True
+        },
+        "name": "title",
+        "type": {
+            "name": "embedding",
+            "seq_type": "sequence",
+        },
+        "dict": {
+            "type": "char_based"
+        }
+    },
+    "genres": {
+        "type": "one_hot_dense",
+        "dict": {
+            "type": "split",
+            "delimiter": "|"
+        },
+        "name": "genres"
+    }
+}
+
+
+def merge_dict(master_dict, slave_dict):
+    return dict(((k, master_dict.get(k) or slave_dict.get(k))
+                 for k in set(slave_dict) | set(master_dict)))
+
+
+def main(filename, fmt):
+    with open(filename, 'r') as f:
+        conf = json.load(f)
+        obj = dict()
+        for k in conf:
+            val = conf[k]
+            file_dict = val['file']
+            file_dict = merge_dict(file_dict, DEFAULT_FILE)
+
+            fields = []
+            for pos, field_key in enumerate(val['fields']):
+                assert isinstance(field_key, basestring)
+                field = copy.deepcopy(DEFAULT_FIELD[field_key])
+                field['pos'] = pos
+                fields.append(field)
+            obj[k] = {
+                "file": file_dict,
+                "fields": fields
+            }
+    meta = {
+        "meta": obj
+    }
+    # print meta
+    if fmt == 'json':
+        def formatter(x):
+            import json
+            return json.dumps(x, indent=2)
+    elif fmt == 'yaml':
+        def formatter(x):
+            import yaml
+            return yaml.safe_dump(x, default_flow_style=False)
+    else:
+        raise NotImplementedError("Dump format %s is not implemented" % fmt)
+
+    print formatter(meta)
+
+
+if __name__ == '__main__':
+    args = docopt.docopt(__doc__, version="0.1.0")
+    main(args["<config_file>"], args["--output_format"])

demo/recommendation/data/meta_config.json

+{
+  "meta": {
+    "movie": {
+      "fields": [
+        {
+          "type": "id", 
+          "pos": 0
+        }, 
+        {
+          "regex": {
+            "pattern": "^(.*)\\((\\d+)\\)$", 
+            "group_id": 1, 
+            "strip": true
+          }, 
+          "type": {
+            "seq_type": "sequence", 
+            "name": "embedding"
+          }, 
+          "dict": {
+            "type": "char_based"
+          }, 
+          "name": "title", 
+          "pos": 1
+        }, 
+        {
+          "type": "one_hot_dense", 
+          "dict": {
+            "delimiter": "|", 
+            "type": "split"
+          }, 
+          "name": "genres", 
+          "pos": 2
+        }
+      ], 
+      "file": {
+        "delimiter": "::", 
+        "type": "split", 
+        "name": "movies.dat"
+      }
+    }, 
+    "user": {
+      "fields": [
+        {
+          "type": "id", 
+          "pos": 0
+        }, 
+        {
+          "type": "embedding", 
+          "dict": {
+            "type": "char_based"
+          }, 
+          "name": "gender", 
+          "pos": 1
+        }, 
+        {
+          "type": "embedding", 
+          "dict": {
+            "sort": true, 
+            "type": "whole_content"
+          }, 
+          "name": "age", 
+          "pos": 2
+        }, 
+        {
+          "type": "embedding", 
+          "dict": {
+            "sort": "true", 
+            "type": "whole_content"
+          }, 
+          "name": "occupation", 
+          "pos": 3
+        }
+      ], 
+      "file": {
+        "delimiter": "::", 
+        "type": "split", 
+        "name": "users.dat"
+      }
+    }
+  }
+}

demo/recommendation/data/ml_data.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+set -ex
+cd "$(dirname "$0")"
+# download the dataset
+wget http://files.grouplens.org/datasets/movielens/ml-1m.zip
+# unzip the dataset
+unzip ml-1m.zip
+# remove the unused zip file
+rm ml-1m.zip

demo/recommendation/data/split.py

+#!/bin/env python2
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Separate movielens 1m dataset to train/test file.
+
+Usage:
+    ./separate.py <input_file> [--test_ratio=<test_ratio>] [--delimiter=<delimiter>]
+    ./separate.py -h | --help
+
+Options:
+    -h --help                       Show this screen.
+    --version                       Show version.
+    --test_ratio=<test_ratio>       Test ratio for separate [default: 0.1].
+    --delimiter=<delimiter>         File delimiter [default: ,].
+"""
+import docopt
+import collections
+import random
+
+
+def process(test_ratio, input_file, delimiter, **kwargs):
+    test_ratio = float(test_ratio)
+    rating_dict = collections.defaultdict(list)
+    with open(input_file, 'r') as f:
+        for line in f:
+            user_id = int(line.split(delimiter)[0])
+            rating_dict[user_id].append(line.strip())
+
+    with open(input_file + ".train", 'w') as train_file:
+        with open(input_file + ".test", 'w') as test_file:
+            for k in rating_dict.keys():
+                lines = rating_dict[k]
+                assert isinstance(lines, list)
+                random.shuffle(lines)
+                test_len = int(len(lines) * test_ratio)
+                for line in lines[:test_len]:
+                    print >> test_file, line
+
+                for line in lines[test_len:]:
+                    print >> train_file, line
+
+
+if __name__ == '__main__':
+    args = docopt.docopt(__doc__, version='0.1.0')
+    kwargs = dict()
+    for key in args.keys():
+        if key != '--help':
+            param_name = key
+            assert isinstance(param_name, str)
+            param_name = param_name.replace('<', '')
+            param_name = param_name.replace('>', '')
+            param_name = param_name.replace('--', '')
+            kwargs[param_name] = args[key]
+    process(**kwargs)

demo/recommendation/dataprovider.py

+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+try:
+    import cPickle as pickle
+except ImportError:
+    import pickle
+
+from paddle.trainer.PyDataProvider2 import *
+import common_utils  # parse
+
+
+def hook(settings, meta, **kwargs):
+    """
+    Init hook is invoked before process data. It will set obj.slots and store
+    data meta.
+
+    :param obj: global object. It will passed to process routine.
+    :type obj: object
+    :param meta: the meta file object, which passed from trainer_config. Meta
+                 file record movie/user features.
+    :param kwargs: unused other arguments.
+    """
+    del kwargs  # unused kwargs
+
+    # Header define slots that used for paddle.
+    #    first part is movie features.
+    #    second part is user features.
+    #    final part is rating score.
+    # header is a list of [USE_SEQ_OR_NOT?, SlotType]
+    headers = list(common_utils.meta_to_header(meta, 'movie'))
+    headers.extend(list(common_utils.meta_to_header(meta, 'user')))
+    headers.append(dense_vector(1))  # Score
+
+    # slot types.
+    settings.input_types = headers
+    settings.meta = meta
+
+
+@provider(init_hook=hook, cache=CacheType.CACHE_PASS_IN_MEM)
+def process(settings, filename):
+    with open(filename, 'r') as f:
+        for line in f:
+            # Get a rating from file.
+            user_id, movie_id, score = map(int, line.split('::')[:-1])
+
+            # Scale score to [-5, +5]
+            score = float(score) * 2 - 5.0
+
+            # Get movie/user features by movie_id, user_id
+            movie_meta = settings.meta['movie'][movie_id]
+            user_meta = settings.meta['user'][user_id]
+
+            outputs = [movie_id - 1]
+
+            # Then add movie features
+            for each_meta in movie_meta:
+                outputs.append(each_meta)
+
+            # Then add user id.
+            outputs.append(user_id - 1)
+
+            # Then add user features.
+            for each_meta in user_meta:
+                outputs.append(each_meta)
+
+            # Finally, add score
+            outputs.append([score])
+            # Return data to paddle
+            yield outputs

demo/recommendation/evaluate.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+function get_best_pass() {
+  cat $1  | grep -Pzo 'Test .*\n.*pass-.*' | sed  -r 'N;s/Test.* cost=([0-9]+\.[0-9]+).*\n.*pass-([0-9]+)/\1 \2/g' | sort | head -n 1
+}
+
+LOG=`get_best_pass log.txt`
+LOG=(${LOG})
+echo 'Best pass is '${LOG[1]}, ' error is '${LOG[0]}, 'which means predict get error as '`echo ${LOG[0]} | python -c 'import math; print math.sqrt(float(raw_input()))/2'`
+
+evaluate_pass="output/pass-${LOG[1]}"
+
+echo 'evaluating from pass '$evaluate_pass

demo/recommendation/prediction.py

+#!/bin/env python2
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from py_paddle import swig_paddle, DataProviderWrapperConverter
+
+from common_utils import *
+from paddle.trainer.config_parser import parse_config
+
+try:
+    import cPickle as pickle
+except ImportError:
+    import pickle
+import sys
+
+if __name__ == '__main__':
+    model_path = sys.argv[1]
+    swig_paddle.initPaddle('--use_gpu=0')
+    conf = parse_config("trainer_config.py", "is_predict=1")
+    network = swig_paddle.GradientMachine.createFromConfigProto(conf.model_config)
+    assert isinstance(network, swig_paddle.GradientMachine)
+    network.loadParameters(model_path)
+    with open('meta.bin', 'rb') as f:
+        meta = pickle.load(f)
+        headers = list(meta_to_header(meta, 'movie'))
+        headers.extend(list(meta_to_header(meta, 'user')))
+        cvt = DataProviderWrapperConverter(True, map(lambda x: x[1], headers))
+        while True:
+            movie_id = int(raw_input("Input movie_id: "))
+            user_id = int(raw_input("Input user_id: "))
+            movie_meta = meta['movie'][movie_id]    # Query Data From Meta.
+            user_meta = meta['user'][user_id]
+            data = [movie_id - 1]
+            data.extend(movie_meta)
+            data.append(user_id - 1)
+            data.extend(user_meta)
+            data = map(lambda (header, val): val if header[0] else [val],
+                       zip(headers, data))
+            print "Prediction Score is %.2f" % ((network.forwardTest(cvt([
+                data]))[0]['value'][0][0] + 5) / 2)

demo/recommendation/preprocess.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+
+cd "$(dirname "$0")"
+delimiter='::'
+dir=ml-1m
+cd data
+echo 'generate meta config file'
+python config_generator.py config.json > meta_config.json
+echo 'generate meta file'
+python meta_generator.py $dir meta.bin --config=meta_config.json
+echo 'split train/test file'
+python split.py $dir/ratings.dat --delimiter=${delimiter} --test_ratio=0.1
+echo 'shuffle train file'
+shuf $dir/ratings.dat.train > ratings.dat.train
+cp $dir/ratings.dat.test .
+echo "./data/ratings.dat.train" > train.list
+echo "./data/ratings.dat.test" > test.list

demo/recommendation/requirements.txt

+PyYAML
+docopt

demo/recommendation/run.sh

+#!/bin/bash
+# Copyright (c) 2016 Baidu, Inc. All Rights Reserved
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+set -e
+paddle train \
+    --config=trainer_config.py \
+    --save_dir=./output \
+    --use_gpu=false \
+    --trainer_count=4\
+    --test_all_data_in_one_period=true \
+    --log_period=100 \
+    --dot_period=1 \
+    --num_passes=50  2>&1 | tee 'log.txt'

demo/semantic_role_labeling/data/test.list

+./data/feature

demo/semantic_role_labeling/data/train.list

+./data/feature