maint: Adapt to breaking changes in scipy.sparse.isspmatrix #598
Conversation
This used to return True with sparse arrays as well, but now it only returns True for sparse matrices. To keep the old behavior, we use `scipy.sparse.issparse` instead.
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| @@ -65,7 +65,7 @@ | |||
| sort, | |||
| zeros, | |||
| ) | |||
| from scipy.sparse import csr_matrix, hstack, identity, isspmatrix | |||
| from scipy.sparse import csr_matrix, hstack, identity, issparse | |||
Check notice
Code scanning / CodeQL
Unused import
|
this won't work for older scipy's, you need to check scipy version before doing this, preferably importing into a unified name depending on version. |
|
Also, all of sisl still uses sparse matrices, but I agree it would be nice if users could do sparse arrays. |
|
Note this could be collected into a |
|
I have an application that uses sisl with arrays, that's how I noticed the bug. I checked up to the minimum version that sisl supports now (1.5.0), and |
|
Sorry, I missed it! |
|
FYI I could also locate it in 0.10, so I think we are safe! Thanks! |
|
Perfect :) |
|
By the way I think now the |
|
yeah, they should just be removed. |
|
These got corrected, now a little more to go and we are done! :) |
In this PR: scipy/scipy#18528 scipy introduced a breaking change which has already been released in
1.11.isspmatrixno longer returnsTrueif you pass a sparse array (instead of a sparse matrix).To keep the old behavior, we should use
scipy.sparse.issparseinstead.isspmatrix_fmt(e.g.isspmatrix_csr) becomes more cumbersome because there is no equivalent function in scipy to check for sparse arrays as well, and you have to goissparse(matrix) and matrix.format == "fmt".These changes keep the old behavior of sisl even with
scipy >= 1.11and don't introduce backwards incompatibilities, so no need to bump thescipydependency (I think).