Skip to main content

Jupyter Notebooks

JupyterLab is an interactive web-based environment for data science and scientific computing. On the DSRI it runs as a container based on the official Jupyter docker stacks, with sudo privileges so you can install anything you need using pip, conda, or apt.

Deploy​

Find the JupyterLab template in the DSRI Catalog (make sure the Templates checkbox is checked) and instantiate it with the following parameters:

  • Password to access the notebook
  • Docker image to use (see available images below)
  • Git repository URL (optional) - will be automatically cloned at startup; if a requirements.txt is present, packages will be installed automatically with pip
  • Git username and email to automatically configure git

Persistent storage​

A persistent volume is automatically created and mounted at /home/jovyan/work - the default working folder in JupyterLab. Data stored here survives pod restarts. You can find it in the DSRI web UI under Administrator view > Storage > Persistent Volume Claims.

Available images​

The default image is ghcr.io/maastrichtu-ids/jupyterlab:latest, a custom image for data science on the DSRI with additional kernels (Java), conda integration, VisualStudio Code, and Python autocomplete.

You can also use any image from the official Jupyter docker stack:

  • ghcr.io/maastrichtu-ids/jupyterlab:knowledge-graph - for working with knowledge graphs, includes a SPARQL kernel and OpenRefine
  • jupyter/scipy-notebook - science packages pre-installed
  • jupyter/datascience-notebook - includes Julia kernel
  • jupyter/tensorflow-notebook - TensorFlow pre-installed
  • jupyter/r-notebook - for working with R
  • jupyter/pyspark-notebook - for connecting to a Spark cluster
  • jupyter/all-spark-notebook - for running Spark locally in the notebook

To build your own image, use this repository as a starting point.

Manage dependencies with conda​

With the ghcr.io/maastrichtu-ids/jupyterlab:latest image you can use conda environments as notebook kernels, as long as nb_conda_kernels and ipykernel are installed in those environments.

You can pass a git repository URL containing an environment.yml file at startup - the conda environment will be installed automatically and appear in the JupyterLab Launcher. See this repository for an example.

Or install it directly in a running JupyterLab using mamba (faster than conda):

mamba env create -f environment.yml

Wait 1-2 minutes for the new environment to appear in the Launcher.

Example environment.yml for a Python 3.9 environment:

name: custom-env
channels:
  - defaults
  - conda-forge
  - anaconda
dependencies:
  - python=3.9
  - ipykernel
  - nb_conda_kernels
  - pip
  - pip:
    - matplotlib
caution

You cannot use conda activate in a Docker container. Use conda run instead to run scripts in a specific environment:

conda run -n custom-env python --version

Use git​

You can always use git from the terminal. Before pushing to GitHub or GitLab, configure your username and email:

git config --global user.name "Jean Dupont"
git config --global user.email jeandupont@gmail.com

To save your password for 15 minutes:

git config credential.helper cache

Or store it in a plain text file:

git config --global credential.helper 'store --file ~/.git-credentials'
tip

We recommend using SSH instead of HTTPS where possible. See GitHub's guide on generating SSH keys.

You can also use the JupyterLab Git extension to manage repositories directly from the UI. It will prompt for credentials if the repository is private.