Docker image with CUDA 8, cuDNN 5.1.10, Python 3, Theano, Keras, OpenCV, Matplotlib and Jupyter Notebook installed. GPU acceleration available if you use nvidia-docker.

• Installed Docker
• If want to use GPU: installed nvidia-docker plugin for Docker and NVIDIA GPU drivers

docker build -t ml-env .

where ml-env can be replaced with any other name.

nvidia-docker run -it --rm -p 8888:8888 --volume $(pwd):/host_dir --workdir /host_dir ml-env bash

This command creates and starts container from previously built image (ml-env) with current host directory mounted as /host_dir and set as working directory.

• -it - redirects std and stdout from container to your terminal.
• --rm - removes container after exit.
• -p 8888:8888 - exposes port 8888 (in:out) from container to host. Required by Jupyter Notebook.
• --volume $(pwd):/host_dir - mounts current host directory ($(pwd)) in container in /host_dir. Instead of $(pwd) you can put any other path.
• --workdir /host_dir - sets current directory in container to previously mounted.
• ml-env - name of image to create container from.

nvidia-docker run -it --rm -p 8888:8888 --volume $(pwd):/host_dir --workdir /host_dir ml-env python3 /host_dir/hello_world.py

or

nvidia-docker run -it --rm -p 8888:8888 --volume $(pwd):/host_dir --workdir /host_dir ml-env jupyter notebook

or any other command at the end instead of python3 or jupyter.

If you need Jupyter secured by password, run command below and enter password:

nvidia-docker run -it --rm -p 8888:8888 --volume $(pwd):/host_dir --workdir /host_dir ml-env run_secured_jupyter.sh