Hello, scikit-learn maintainers!

We're making efforts to better integrate sparse into the PyData ecosystem. For those unaware, sparse provides N-dimensional sparse arrays and aspires to comply as closely as possible to the Array API.

We've already made significant progress in this direction:

• We've already integrated sparse into scipy.sparse:
  • API: Enable pydata/sparse input for csgraph module scipy/scipy#19796
  • MAINT: sparse: Adjust expm and expm_multiply to pydata/sparse input scipy/scipy#20485
• We've added NumPy 2.0 support: issue, PRs
• We now support a significant subset of the the Array API: PRs
• We're also making efforts to transition to a lazy compilation model á la JAX and PyTorch, as well as GPU support and more formats: Announcement post

The next step for us seems to be to target other upstream libraries which would benefit from sparsity, and scikit-learn is at the top of that list. To this end, we'd like to start by adding an upstream CI job for testing sparse, and we'll also make efforts on our end to squash any issues (on either side) that may arise, as ecosystem integration is one of our highest priorities.

#29031 was a step in that direction, but we understand that a path forward needs to be agreed upon with the scikit-learn maintainers before anything lands.

cc @ogrisel @betatim as contributors leading the Array API work in scikit-learn.

xref #26724 as it will likely become relevant