Repository navigation
Axis order from "fancy" indexing with dask.array does not match NumPy #7313
Description
This indexing operations from Dask and NumPy should result in arrays with the same shape:
import dask.array as da
import numpy as np
print(np.ones((2, 3, 4))[0, :, [0, 1]].shape) # (2, 3)
print(da.ones((2, 3, 4))[0, :, [0, 1]].shape) # (3, 2)This is the "Combing advanced and basic indexing" case from NumPy: https://numpy.org/doc/stable/reference/arrays.indexing.html#combining-advanced-and-basic-indexing
It's also the same counter-intuitive example from NEP-21: https://numpy.org/neps/nep-0021-advanced-indexing.html#mixed-indexing
It would be best to fix this inconsistency, or to raise an error pointing users to vindex.
Note: this is somewhat related to but distinct from #4157
Environment:
- Dask version: 2.12.0
Reacted by Josh Moore and abergou
Activity
ncclementi commentedon Aug 11, 2021
Retested and this issue still persists on
Dask 2021.07.2,@shoyer Would you like to take a crack at this?
shoyer commentedon Aug 11, 2021