SciPython-Docker

This repo contains a Dockerfile to build a foundational scientific Python Docker image. It is intended to be the foundation of a development environment in a context where installing a bespoke set of required packages from the internet may be cumbersome or discouraged. Built images are hosted on Docker Hub. The foundation of the image is the Mambaforge environment management system developed by the Conda-Forge community. Core packages included in the image include:

• CPython (3.12)

Additional packages are included for:

• Documentation (Sphinx)
• Testing (pytest, coverage)
• Linting (flake8, pylint)
• Environment management (python-dotenv)
• Database connectivity (sqlalchemy, pyodbc, pymssql)
• Literate programming (Quarto)

In addition, it includes:

• R (4.4.1);
• RStudio-Server;
• jupyter-rsession-proxy, so you can launch an RStudio session from within Jupyter Notebook/Lab, and a collection of R packages centered around the tidyverse, and literate programming;

Usage

To instantiate an ephemeral container from the image, mount the current directory within the container, and open a bash prompt within the base conda Python environment:

docker run -it --rm -v $(pwd):/home/docker/work blueogive/scipython-docker:latest

By default, you will be running as the (unprivileged) docker user within the container.

Typical Usage

Instantiate a container from the image:

docker run -d --name <container_name> -v $(pwd):/home/docker/work -p 8888:8888 --restart unless-stopped blueogive/scipython-docker:latest /bin/sleep infinity

replacing <container_name> with the name you wish to assign to your container.

At this point, you will have a Bash shell with the base mamba/conda Python virtual environment will be active. You can use the remote development capabilities of VSCode to connect to the container and begin working.

Alternatively, you may wish to use Jupyter Lab and/or RStudio within the container. In that case, you can use the entrypoint argument to have the container set up a starter environment:

docker run -d --name <container_name> -v $(pwd):/home/docker/work -p 8888:8888 --restart unless-stopped --entrypoint /usr/local/bin/start-jupyterlab.sh blueogive/scipython-docker:latest

As the container starts, it will require 2--3 minutes to build the environment. The process should end when the Jupyter Lab process starts and echoes a URL to the terminal. Open the URL in your browser to connect to the Jupyter Lab process in the container. You can start RStudio by clicking the launcher on the Jupyter Lab home screen.

Alternatively, if your requirements are more exacting, you can complete some of the same steps taken by the entrypoint script manually and modify them to fit your needs. Open a shell within the container:

docker exec -it <container_name> /bin/bash

Create a new conda/mamber virtual environment that includes Jupyter (because the base virtual environment does not):

mamba env create -f ~/miniforge3/examples/conda-env-minimal.yml

Activate the new virtual environment:

mamba activate minimal

If you want to use R within Jupyter, install the R kernel:

Rscript -e "IRkernel::installspec()"

Complete the post-install build of Jupyter Lab and start the service:

jupyter lab build
jupyter lab --no-browser --ip 0.0.0.0 --notebook-dir=~/work

As Jupyter Lab starts, it will echo a bit of output to the console ending with statement similar to:

Or copy and paste one of these URLs:
        http://(<container_id> or 127.0.0.1):8888/?token=<token_value>

where <container_id> and <token_value> are hexadecimal strings unique to your instance. On your host, open your preferred browser and point it to http://localhost:8888/?token=<token_value> to connect to the Jupyter Lab instance.

Contributions are welcome.