Dask support for `min`/`max`

[quentinblampey]

Thanks again @flying-sheep for merging the PR about min/max.
While using the new release (1.3.0), I saw two bugs:

from fast_array_utils import stats
import dask.array as da
import numpy as np
from scipy.sparse import csr_matrix
from anndata import AnnData
import anndata

# create multiple data inputs

x_np = np.array([
    [1, 0, 0, 3],
    [0, 5, 6, 0],
    [7, 0, -1, 0],
])

x_sparse = csr_matrix(x_np)
x_dask = da.from_array(x_np)

adata = AnnData(X=x_np)
adata.write_h5ad("test_backed.h5ad")
x_backed = anndata.read_h5ad("test_backed.h5ad", backed="r").X

Bug 1 - mean_var on h5py.Dataset

>>> stats.mean_var(x_backed)
TypeError: unsupported operand type(s) for ** or pow(): 'Dataset' and 'int'

Bug 2 - min/max on dask array

>>> stats.max(x_dask)
TypeError: max() got an unexpected keyword argument 'dtype'

I can have a quick look this week

[flying-sheep]

The first one isn’t actually a bug, mean_var isn’t documented to work on h5py.Datasets.

But the second one definitely is, as we document it to be! We didn’t add tests in your PR and I though the STAT_FUNCS array means that it at least gets run with all array types, but that was wrong.

[quentinblampey]

Is there a reason for not supporting h5py.Dataset for mean_var? If not, I can have a look and make a PR

[flying-sheep]

That would be wonderful! No reason apart from “nobody has done it yet”!