Jupyter Notebook .gitignore template

.ipynb_checkpoints is the autosave directory. Every time the editor saves a checkpoint it writes a full copy of the notebook there, so the directory ends up holding a stale duplicate of every notebook in the folder. Both rules in this template target it: the first matches it at any depth, the second matches its contents in a subdirectory, which is belt and braces for older tooling.

profile_default and ipython_config.py hold interactive shell configuration for one user: startup scripts, display settings, extension loads. .virtual_documents is written by editor language server integrations, which extract the code from a notebook into plain files so a language server can analyse them.

The problem this template cannot solve is the notebook format itself. A notebook is JSON containing both your source and the output of every cell, and the output includes execution counts that change on every run, base64 encoded images, and whatever the last run printed. Two people running the same notebook produce completely different files, so the diff is unreadable and the merge conflicts are unresolvable by hand.

The usual answers are to strip output before committing, using a filter that runs on staging, or to pair each notebook with a plain script export and review that instead. Neither is an ignore rule, which is why a repository that only adds this template still ends up with noisy history.

The other risk is what those outputs contain. A cell that printed the head of a dataframe has embedded that data in the notebook file, and a cell that echoed an environment variable while debugging has embedded a credential. Committing a notebook is closer to committing a screenshot of your session than to committing code.