Jupyter notebooks in Git

Working with Jupyter notebooks in Git repositories has always caused some struggles for me with the Git commit history. Jupyter notebooks use the json-format and include some metadata such as the number of times a notebook has been run. So far, I have just cleared the output restarted and run notebooks once to have somewhat clean metadata. This has not always worked and let to some slightly noisy git commits.

Now, I have tried a solution that I remembered when I wanted to commit a Jupyter notebook and run into noisy git diffs again. nbdev_clean removes cell execution related metadata from a notebook.

nbdev_clean --fname . cleans all Juypter notebooks in a repository. I run it on this learning diary Git repository to clean the two notebooks in the snippets folder. I also used it in the project that initially caused me to look into nbdev.

Finally, i used Codespaces to quickly create minimal, reproducible examples of git diffs without and with nbdev_clean. Again, Codespaces are really nice.