iaindillingham
commented
3 years ago Diffable notebooks are hard and, I'd argue, not the solution. That said, tools such as nbdime can help.
I said they're "not the solution", which is a bold claim given that I don't know the problem 🙂 However, I think the problem is that notebooks aren't good places to store analytic code: they work well for exploring and communicating, but work poorly for reasoning. Cookiecutter Data Science suggest that after we've explored, we move analytic code to a Python module, where it can be tested. Python modules are much easier to diff, too.
Add ability to use diffable notebooks and push to existing repos