`sc.pp.scale` leads to `UserWarning: zero-centering a sparse array/matrix densifies it.`

[jpintar]

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?

Running sc.pp.scale(adata) where adata.X is a csr_matrix leads to

.../functools.py:934: UserWarning: zero-centering a sparse array/matrix densifies it.
  return dispatch(args[0].__class__)(*args, **kw)

Minimal code sample

import scanpy as sc
import anndata as ad
import scipy as sp

adata = ad.AnnData(X=sp.sparse.rand(1000, 1000, format="csr"))
sc.pp.scale(adata)

Error output

Versions

Affects versions 1.11.3 and 1.11.2, but not 1.11.1.

Details

scanpy	1.11.3
anndata	0.12.0
scipy	1.15.3
----	----
PyYAML	6.0.2
typing_extensions	4.14.1
threadpoolctl	3.6.0
six	1.17.0
natsort	8.4.0
legacy-api-wrap	1.4.1
zarr	3.1.0
pandas	2.3.1
donfig	0.8.1.post1
pyparsing	3.2.3
setuptools	80.9.0
kiwisolver	1.4.8
packaging	25.0
joblib	1.5.1
llvmlite	0.44.0
numpy	2.2.6
h5py	3.14.0
scikit-learn	1.7.1
pytz	2025.2
matplotlib	3.10.3
numba	0.61.2
crc32c	2.7.1
cycler	0.12.1
session-info2	0.1.2
numcodecs	0.16.1
python-dateutil	2.9.0.post0
pillow	11.3.0
----	----
Python	3.12.11 | packaged by conda-forge | (main, Jun  4 2025, 14:45:31) [GCC 13.3.0]
OS	Linux-4.18.0-425.19.2.el8_7.x86_64-x86_64-with-glibc2.28

[ilan-gold]

This is not a bug but the intended behavior - scale by default zero-centers and you are passing in a sparse matrix.

Please see the docs https://scanpy.readthedocs.io/en/stable/generated/scanpy.pp.scale.html#scanpy.pp.scale

[jpintar]

I know sc.pp.scale densifies the matrix - I was surprised to get a warning exactly because it is expected behaviour. 😆
But I suppose it makes sense to warn users about it anyway...

[ilan-gold]

@jpintar I agree these things might come off as strange, but as someone who has only been maintaining the project for a fraction of the time it has existed, I have come to suspect these sorts of things came about specifically because someone else 8 years ago complained about the opposite - no warning about increased memory usage in the default case!

[flying-sheep]

yeah, everything is a trade-off:

Option	Pro	Con
Default depends on input data type	default always makes sense	default completely changes behavior based on what can be perceived as an implementation detail
Default zero-centers	expected behavior for dense arrays	high memory usage when fed a sparse array
Default doesn’t zero-center	never blows up memory	suboptimal behavior for dense arrays
No default (you have to specify zero_center=True|False)	never unexpected behavior	verbose

I guess for scanpy 2.0 we could maybe change it to the last option 🤔