This is my take on Dockerized Jupyter. I use it for Keras and Tensorflow projects. It assumes notebooks are in <current working directory>/notebooks/ and datasets are in <current working directory>/notebooks/datasets.
Collaboration is installed by default (via jupyter_collaboration)! So if you install this on a server feel free to work it with your team. See jupyter_collaboration docs.
Use chmod +x run.sh to make the builds the Dockerfile in setup/ and opens a Jupyter notebook which mounts the notebooks/ directory for reading and writing and data/ for reading. notebooks/ will show up as your working directory, and there will be a subdirectory data/ with your data.
Pass cpu, gpu, or help to run.sh.
gpu: run the Docker image with all system GPUs attached.
cpu: run the Docker image with just your CPU.
help: prints a help message.
Comments, questions, and suggestions are always welcome. Please open an issue for any of them.