Sparse masking throws error

[ilan-gold]

Describe the issue:
Computing a masked version of a sparse array throws an error on compute. I am aware than sparse support may not be official but figured I'd report anyway.

Minimal Complete Verifiable Example:

from scipy import sparse
import dask.array as da

X = sparse.random(10_000, 1_000, format="csr", density=0.1)
X_dask = da.from_array(X)
X_dask[X_dask > 0].compute()
X_dask[X_dask > 0] = 1000
X_dask.compute()

Anything else we need to know?:

I am happy to help with this (and other things we are loosely aware of surrounding sparse arrays) if you would accept a PR or investigation at least.

Environment:

• Dask version: 2024.2.1
• Python version: 3.9.18
• Operating System: Mac
• Install method (conda, pip, source): pip

[ilan-gold]

So the first issue seems to have a host of issues (X_dask[X_dask > 0].compute()) including lack of 1D support in scipy.sparse and also a general incompatibility with dask and np/sp.matrix but the second issue has a more direct cause because there is no np.where equivalent in scipy.sparse. This can be easily reproduced with

np.where(X>0, 10, X)

But this sort of setting does work in general

mask = X>0
X[X>0] = 10
X[mask] # all 10's