wrong nanmean with complex64

[dcherian]

Describe the bug np.nanmean of sparse array does not match numpy result

To Reproduce

import numpy as np
import sparse as sp

array = np.array([0. +0.j, 0.+np.nan * 1j], dtype=np.complex64)
sparray = sp.COO.from_numpy(array)

print("numpy:", np.nanmean(array), "sparse:", np.nanmean(sparray))

numpy: 0j sparse: (nan+nanj)

Expected behavior Match the numpy result

System

• OS and version: [e.g. Windows 10] macOS latest
• sparse version (sparse.__version__) 0.13.0
• NumPy version (np.__version__) 1.22.4
• Numba version (numba.__version__) 0.55.2

Additional context Seen when using hypothesis to test out array wrapping in xarray (https://github.com/pydata/xarray/pull/4972)

[hameerabbasi]

Thanks for the report, I'll look into this.

[dcherian]

Thanks!