Support for scipy.csr_array/scipy.csc_array

[alam-shahul]

Please make sure these conditions are met

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of scanpy.
• (optional) I have confirmed this bug exists on the main branch of scanpy.

What happened?

It seems like sc.pp.pca doesn't work when .X is a csr_array (but it works fine when it's a csr_matrix); seems like some low level matrix multiplication subroutine is not implemented for csr_arrays.

Minimal code sample

import scanpy as sc
import anndata as ad
import numpy as np
from scipy.sparse import csr_array, csr_matrix

data = np.random.random((2, 5))
dataset = ad.AnnData(
    X=csr_array(data),
    # X=csr_matrix(data),
)
sc.pp.pca(dataset)

Error output

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/env/lib/python3.10/site-packages/scanpy/preprocessing/_pca.py", line 285, in pca
    X_pca, pca_ = _pca_with_sparse(
  File "/env/lib/python3.10/site-packages/scanpy/preprocessing/_pca.py", line 453, in _pca_with_sparse
    total_var = _get_mean_var(X)[1].sum()
  File "/env/lib/python3.10/site-packages/scanpy/preprocessing/_utils.py", line 40, in _get_mean_var
    mean_sq = axis_mean(elem_mul(X, X), axis=axis, dtype=np.float64)
  File "/env/lib/python3.10/functools.py", line 889, in wrapper
    return dispatch(args[0].__class__)(*args, **kw)
  File "/env/lib/python3.10/site-packages/scanpy/_utils/__init__.py", line 546, in elem_mul
    raise NotImplementedError
NotImplementedError

Versions

1.10.4

[flying-sheep]

you forgot to fill out the version information. Please do that. It’s likely that this is true, but I still need that info.

[alam-shahul]

Hmm... well, I just tried it again on my current installation, and I got a different error:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/work/magroup/shahula/bin/miniconda3/envs/popari_distribution/lib/python3.10/site-packages/scanpy/preprocessing/_pca.py", line 200, in pca
    output = _pca_with_sparse(
  File "/work/magroup/shahula/bin/miniconda3/envs/popari_distribution/lib/python3.10/site-packages/scanpy/preprocessing/_pca.py", line 271, in _pca_with_sparse
    mu = X.mean(0).A.flatten()[None, :]
AttributeError: 'numpy.ndarray' object has no attribute 'A'

This is for version 1.9.3. I thought that the previously reported error was also in version 1.9.3, but since the error is different now I'm not sure...

[alam-shahul]

Oh, I just reproduced the original error in 1.10.4 (updated above).

[flying-sheep]

OK, I’ll fix this more generally than just for PCA I think.

I made #3431 to prepare the scene for this fix for now.