ML Docker Image

This docker image can be used to quickly get into a ML development environment, and also as a base for deploying dockerized ML applications onto AWS services. It is based off the paperspace/fastai image, with additional packages that we need.

Development

Build the docker image with bash ./build.sh

Then you can use something like

sudo docker run --runtime=nvidia -p 127.0.0.1:8888:8888/tcp --shm-size 8G -d --env-file .env -v ~/:/opt/host ml_nn_fastai "source activate fastai && jupyter notebook --ip 0.0.0.0 --port 8888 --no-browser --allow-root --NotebookApp.token=''"

This:

• Starts the docker image with NVIDIA GPU support
• Loads environment variables from a .env file
• Mounts the host filesystem at the root directory, so you can persist Notebooks, cached files etc.
• Boots a Jupyter notebook server and binds it to port 8888 on the host system