Python containerized development environment

A docker image that comes with common packages for science, especially for machine learning with Nvidia + CUDA.

Packages

tensorflow
keras
numpy
matplotlib
h5py
cmcrameri
scipy
seaborn
pyyaml
tqdm
ipython
ipykernel
jupyterlab

Requirements

docker
docker-compose (optional)
nvidia-container-toolkit configured for docker

$ nvidia-ctk runtime configure --runtime=docker
$ sudo systemctl restart docker

Using the bundled Jupyter Lab

$ git pull github.com/mstcl/pycde
$ cd pycde
$ docker compose up -d --build

You can change the bind mount inside docker-compose.yml to mount your project folder.

If you want to use standalone Docker instead:

$ docker build -t pycde
$ docker run -p 127.0.0.1:8888:8888 -v /srv/projects:/projects --restart unless-stopped --security-opt=no-new-privileges --log-opt max-size=1g --gpus 1 pycde

Running as a standalone kernel (to use with your own editor)

Minimum requirements for the connecting host

ipykernel
ipython

Creating a custom ipython kernelspec

$ python -m ipykernel install --user --name=docker

Then edit ~/.local/share/jupyter/kernels/docker/kernel.json to look something like (also in examples/kernel.json ):

{
  "argv": [
    "/usr/bin/docker",
    "run",
    "--rm",
    "--network=host",
    "--gpus",
    "all",
    "-v",
    "{connection_file}:/connection-spec",
    "-v",
    "/projects:/project",
    "pycde-pycde",
    "python",
    "-m",
    "ipykernel_launcher",
    "-f",
    "/connection-spec"
  ],
  "display_name": "docker",
  "language": "python"
}

In VSCode, Jupyter Lab, or whatever, you can select the "docker" kernel instead.