scipy.sparse

[jarrodmillman]

Let's work on improving scipy.sparse together! The main focus will be sparse arrays, but we're also interested in compatibility with other scientific python libraries, array API support, performance improvements, bug fixes, and anything else related to sparsity.

[larsoner]

@dschult I opened a PR to move MNE-Python to sparse arrays and it was fairly straightforward. If you have a WIP migration guide I can look I'm happy to review.

[arokem]

You mean this PR?

[larsoner]

Argh yes :)

[arokem]

I also tried to do this on DIPY 😄

[dschult]

Thanks @larsoner and @arokem !
I've put the draft migration guide into draft PR #20926 comments and suggestions are welcome.

A note about migrating indexing: In v1.13 indexing that would typically give a 1D result for np.array raises an exception that support for indexing to 1D sparse arrays is not yet implemented. We are very close to having indexing fully functional.
The current workaround is e.g., A[i, :] -> A[[i], :] which will mimic the spmatrix 2D result for both sparray and spmatrix. Once the indexing PR is merged, the number of options will increase. But that same change will still work.
The other options include:

• using the 1D array and change surrounding code (treat it like np.ndarray)
• using the special index syntax np.newaxis or None to create a 2D result as A[i, None, :]

Please let me know if you have any questions or suggestions about migration and this guide.
Thanks!