Theano and GPU training on Windows

I want to ran existing Theano-based code. It takes forever to train in CPU (of course) but when I try to enable GPU I’m getting error explosion… Fast forward I made it work (on Win 10) and here are couple key steps that were helpful.

  1. I’m using latest Theano\Lasagne bits. See full instructions here.
    pip install --upgrade https://github.com/Theano/Theano/archive/master.zip
    pip install --upgrade https://github.com/Lasagne/Lasagne/archive/master.zip
  2. Once I enabled GPU training via environment variable
  3. THEANO_FLAGS = device=cuda

I got error

Can not use cuDNN on context None: cannot compile with cuDNN. We got this error:
c:\users\olgali\appdata\local\temp\try_flags_zq9bx6.c:4:19: fatal error: cudnn.h: No such file or directory
compilation terminated

I tried lots of things however I think the step below is the one that got me unblocked:

merging cudnn and CUDA directory.

For full instructions go here.

Advertisements

First blog post — Cycle GANs

Well, first blog post have to be cool, right?  I’d so  happens that just few days ago official write up about  my team’s  hackfest with Getty Images got published! Here we got Machine Learning, DNNS, GANs sprinkled with Kubernetes. So here it is. My first blog post will have reference to this awesomeness: Learning Image to Image Translation with CycleGANs. Engoy!