Docker notes

Using Docker for deep learning

John followed the directions here. What he had to do:

Preparing for the container:

1. We already had CUDA installed, so no problem there. If not already installed, you'd need to do it.

2. Install Docker CE using directions here. (Note that the lab machines are x86_64, so follow the command corresponding to amd64 in the directions.

3. Install nvidia-docker

4. I also needed to

   sudo apt install nvidia-modprobe
   

Building the container:

1. Unfortunately, as explained here, if you want GPU support, you need to build the image locally. This was straightforward, but took a while long time.

   Edit: OpenCV build failed when I tried to run this. Alternatives below.

2. For Theano/Lasagne, there's a docker image

   • Because of an error with an outdated version of the Python six package (see here) needed to upgrade that.