tensorflow-windows-build

tensorflow wheel file builded on Windows.
The build flow is following the cmake build guide.

tenforlow version is 0.11.0rc2.

How to install

pip install tensorflow-windows-build/cpu/tensorflow-0.11.0rc2_cmake_experimental-py3-none-any.whl

Build environment

• Windows10(x64, Intel Core i7)
• Python 3.5.2
• Visual Studio Community 2015
• You have to choose Visual C++ option, if you want to build tensorflow yourself.
• cmake 3.6.3
• caution! 3.3.1 causes error when build
• numpy: 1.11.2
• swig: 3.0.10
• zlib: 1.2.8

You can create build environment by conda.

conda create -n tensorflow-win swig numpy zlib

(cmake of conda is 3.3.1, and it causes build error, so you have to install cmake yourself from here.)

build script is like below.

cmake .. -A x64 -DCMAKE_BUILD_TYPE=Release ^
-DSWIG_EXECUTABLE=<your_conda_env_path>\Library\bin\swig.exe ^
-DPYTHON_EXECUTABLE=<your_conda_env_path>\python.exe ^
-DPYTHON_LIBRARIES=<your_conda_env_path>\libs\python35.lib

(You need GPU related option if build on GPU)

To build the tenforflow, you need below resource.

• About 2G disk space
• About 1 ~ 1.5 hours to build (8G memory, Core i7)

And You have to change some of step.

• If you can't find vcvarsall.bat, you can use vcvars64.bat instead.
• I have to apply this fix (it fix is already merged at mater branch).
• I have to apply this fix to build on GPU.