Open riedgar-ms opened 1 year ago
we should support like, last two years or something, not more.
So PR #1287 updates the SciKit-Learn lower bound (assuming we merge) but that still leaves pandas. And that has a lot of deprecation warnings
yeah, and we need to make sure we work with latest pandas, which is very different. Also, we might need to do changes related to the upcoming numpy=2 release.
Looks like we also have issues with things being deprecated in numpy
as well
As mentioned on Discord, there's a problem with pandas. Python 3.8 is only supported up to pandas 2.0. Version 2.1 of pandas deprecates DataFrame.applymap()
and introduces DataFrame.map()
as a replacement.
This means that #1289 is actually incompatible with Python 3.8, and should perhaps be reverted. The issue it addressed can be hushed (for how) in pyproject.toml
we could also do 3.9+ which would be 3.9, 3.10, 3.11, and the upcoming 3.12, which should be enough
Thoughts @fairlearn/fairlearn-maintainers ? I wouldn't object to bumping to >3.9, but this should probably be more widely considered.
I also don't mind bumping to 3.9+ (and it's a good incentive for me to update my own installation, lol)
Thanks.... @MiroDudik @mmadaio @alliesaizan ?
Maybe a year after removing python 3.8 is ok? :) Happy to make the change.
Yeah I think we can remove now. Pythin 3.13 is out already, and we need to test for that.
The current bounds listed for
pandas
andscikit-learn
in requirements.txt are laughably out of date, and we're getting test failures due to deprecation warnings.