I set up a new PC and installed Chainer 1.5 on Windows. When I tried to use the GPU, it was a little troublesome, so I will leave a note. My environment is Windows 8.1 64bit version.
Install CUDA first. If you installed Chainer first without a CUDA environment, it seems that you need to enable CUDA and then install Chainer again.
C: \ Program Files \ NVIDIA GPU Computing Toolkit \ CUDA \ v7.5 \ bin) to ~~ PATH ~~Microsoft Visual C++ Compiler for Python 2.7 It seems to compile the module when installing Chainer, so it is necessary for that.
Apart from the above one, it is necessary to use nvcc (CUDA compiler) when running GPU compatible code. CUDA 7.5 seems to support VS 2010, 2012, 2013. If you have a commercial version of Visual Studio, you can install it normally.
This time it was for home, so I tried to use the free Visual Studio 2013 Express, but CUDA 7.5 doesn't seem to officially support the Express version. The only problem seemed to be the difference between the directory name and file name, so I was able to move it safely by following the steps below.
C: \ Program Files (x86) \ Microsoft Visual Studio 12.0 \ VC \ bin \ x86_amd64 to `C: \ Program Files (x86) \ Microsoft Visual Studio 12.0 \ VC \ bin \ amd64` Copy to. vcvarsx86_amd64.bat in the directory to vcvars64.bat```Python for Windows is distributed in various packages, and I'm not sure which one to choose. I also want to use Jupyter, so this time I decided to install it from Anaconda.
Execute \ Microsoft Visual Studio 12.0 \ vc \ bin \ amd64 \ vcvars64.bat "``` (directory changes depending on VS version)
python train_mnist.py -g 0Recommended Posts