Fixes for sklearn 1.5

[flying-sheep]

Scikit-learn 1.5.0 will no longer have issparse exported, so let’s import it from where it belongs

[codecov]

Codecov Report

Attention: Patch coverage is 76.47059% with 8 lines in your changes are missing coverage. Please review.

Project coverage is 73.90%. Comparing base (9714250) to head (d314182).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3047      +/-   ##
==========================================
- Coverage   73.98%   73.90%   -0.08%     
==========================================
  Files         117      117              
  Lines       12795    12763      -32     
==========================================
- Hits         9466     9433      -33     
- Misses       3329     3330       +1     

Files	Coverage Δ
scanpy/_utils/__init__.py	66.66% <100.00%> (ø)
scanpy/datasets/_utils.py	100.00% <100.00%> (ø)
scanpy/preprocessing/_utils.py	92.10% <100.00%> (ø)
scanpy/testing/_pytest/fixtures/data.py	100.00% <ø> (ø)
scanpy/testing/_pytest/marks.py	100.00% <100.00%> (ø)
scanpy/tools/_ingest.py	77.23% <100.00%> (ø)
scanpy/tools/_tsne.py	93.02% <100.00%> (ø)
scanpy/external/pp/_magic.py	25.00% <50.00%> (ø)
scanpy/tools/_louvain.py	21.42% <50.00%> (ø)
scanpy/_compat.py	74.00% <60.00%> (ø)
... and 2 more

... and 11 files with indirect coverage changes

[flying-sheep]

Hmm, seems like PCA with svd_solver='arpack' works differently in sklearn 1.5 and flips the coordinates in some tests:

scikit-learn/scikit-learn#28826

E.g. this used to work, now it doesn’t.

from __future__ import annotations

import numpy as np
from sklearn.decomposition import PCA

data = np.asarray([[-1, 2, 0], [3, 4, 0], [1, 2, 0]]).T
expected = np.array(
    [[-1.579575e-15, 1.490712], [-2.44949, -0.745356], [2.44949, -0.745356]],
    dtype=np.float32,
)

pca = PCA(n_components=2, svd_solver="arpack", random_state=np.random.RandomState(0))
transformed = pca.fit_transform(data).astype(np.float32)
np.testing.assert_almost_equal(transformed, expected, decimal=5)