KerasSetup.md

Here is how I set up my environment for running Keras with GPU

Install Python 3.5 and pip if you do not already have it

If you want GPU, you'll need CUDA and cuDNN

Install CUDA 8.0 Install cuDNN 5

Install virtualenv if you don't already have it

pip install virtualenv

Create a location for virtualenvs. I prefer to put mine at ~/ve/, this location is not critical

mkdir ~/ve
cd ~/ve

Create a VE for Keras install. Use the -p flag to specify python3.

Note: depending on your env, you may need to specify python3.5 explicitly

virtualenv -p `which python3` keras

Install essentials

pip install numpy scipy pandas jupyter

Install Tensorflow.

For GPU users:

pip install tensorflow-gpu

For CPU Users:

pip install tensorflow

Note: I was able to do this with the very simple pip call. The Tensorflow install instructions used to call for pointed to a specific TF_BINARY_URL, but it seems that is no longer necessary

Verify CUDA connection

ipython

In [1]: import tensorflow

I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:135] successfully opened CUDA library libcurand.so.8.0 locally

In [2]: import keras
Using TensorFlow backend.

After this, I also like to pull up a small Keras model and train it while watching my GPU to make sure it hasn't defaulted to CPU mode (it does this sometimes and it drives me nuts!)

watch -n1 nvidia-smi