DEPR: SparseDtype

[jbrockmendel]

Discussed briefly at the sprint. Sparse is not much used, has a bunch of xfailed tests, and adds about 1MB of .so files.

The suggestion here is to deprecate Sparse internally and help [TBD] to implement a 3rd party EA.

IIRC the main user of Sparse is one of the scikits (or pydata/something?) and their actual use case would be better handled with a 2D implementation. Anyone know the appropriate person to ping to get a view on this?

People who use Sparse, please comment so we know you exist!

[dudelstein]

I use it! I find the functionality quite essential for working with high-cardinality categorical features, where it can reduce memory usage 100x or so. I do find the documentation a bit short and some behavior suprising, for example in this question https://stackoverflow.com/questions/77931609/why-does-pandas-sum-give-wrong-answers-for-sparse-dataframe/77932016#77932016.