@rossbar: FYI, this is the PR that intentionally uses the current behavior of scipy.sparse.isspmatrix as mentioned IRL.

Based on IRL discussions with @rossbar, similar issparray_* functions might be introduced in the future and the isspmatrix_* functios might be deprecated or might have their behavior change in the future.

Based on IRL discussions with @rossbar, similar issparray_* functions might be introduced in the future and the isspmatrix_* functios might be deprecated or might have their behavior change in the future.

There's still discussion to be had, but I would propose to deprecate the isspmatrix_<format> exactly for the reason that it's not API that should be duplicated for arrays.

Instead of e.g. isspmatrix_csc(a), I would argue for the following:

• In the case where you don't care that a is a sparse matrix or sparse array (i.e. you can support both), then I would write this as: issparse(a) and a.format == "csc"
• In the case where you want to make sure you support only matrix semantics (which you wouldn't want to do for new code, but might be necessary to ensure compatibility), then: isinstance(a, spmatrix) and a.format == "csc"

The worst case is when you want to ensure that the input is a sparse array and not a sparse matrix. For this case I think the only current way to do it is with three independent checks: one for sparsity, one to ensure it's not a matrix, and a final one for the format:

>>> issparse(a) and not isinstance(a, spmatrix) and a.format == "csc"

Let's wait for scipy/scipy#18528 to be merged before pursuing.

Another update here - after scipy/scipy#18531, isinstance(..., csr_matrix) now (or will, as of scipy 1.11) return True for csr_matrix and not csr_array.

Yes, this PR needs to be updated to use scipy.sparse.issparse and format instead.

I would be in favor of merging this PR for uniformity and addressing the conversion in another PR. What do you think?