JupyOps

Install jupyter with pipx

First, issues with other methods I have tried:

• Using the system package manager: some extensions may not be actively maintained, for example python-jupyterlab-vim.
• Using a project virtual environment or conda environment: the setup has to be duplicated in each environment.

What I settled on for now is using pipx to manage a separate virtual environment for everything Jupyter. Here is how I do it.

Install the metapackage that pulls in all Jupyter components:

pipx install jupyter --include-deps
pipx ensurepath

Inject other useful packages into the jupyter environment:

pipx inject jupyter ipywidgets
pipx inject jupyter jupyterlab_widgets
pipx inject jupyter jupyterlab-vim
pipx inject jupyter jupyterlab-git
pipx inject jupyter jupyterlab-lsp 'python-lsp-server[all]'
pipx inject jupyter jupyterlab-code-formatter black

To keep the packages updated, run:

pipx upgrade-all

If the system Python is updated, you may also need to reinstall packages under pipx:

pipx reinstall-all

Manage kernel registry

By default, JupyterLab only has access to the ipython kernel from the same virtual environment. To add another kernel, run the following command using the ipython you would like to register:

<your kernel path>/ipython kernel install --user --name=<your kernel name>

For example, this adds the system ipython kernel to the kernel selection list in the Notebook/Lab UI.

/bin/ipython kernel install --user --name=system

To make ipywidgets work, it also needs to be installed in each kernel environment. For conda, for example:

conda install -c conda-forge ipykernel ipywidgets

To see a list of registered kernels:

jupyter kernelspec list

To remove a registered kernel from the list:

jupyter kernelspec uninstall <unwanted kernel>

uninstall and remove are equivalent here.

Avoid Git repository bloating

When notebooks are added to a Git repository, their output can be filtered out to avoid bloating the repository.

1. Add a filter to the repo-wise Git config. Use --global to make it global. We only need to define the clean command, which is used to convert the contents of a worktree file upon check-in.

   git config filter.strip-notebook-output.clean \
     'jupyter nbconvert --ClearOutputPreprocessor.enabled=True --to=notebook --stdin --stdout --log-level=ERROR'

2. Tell Git to apply this filter to .ipynb files.

   echo '*.ipynb filter=strip-notebook-output' >> .git/info/attributes

3. If the repo has existing unscrubbed notebooks, add them to be renormalized and then commit.

   git add --renormalize .

Note that the filter only cleans checked-in notebooks. The local worktree remains untouched.