Repository navigation
Wrong axis order when indexing by (int, slice, list) #6370
Description
Advanced indexing sometimes gives the result transposed w.r.t. the expected result.
This occurs when indexing by (int, slice, list), but not (int, slice, slice) or (slice, int, list).
Example:
>>> np.version.full_version
'1.9.3'
>>> a = np.arange(24).reshape((4, 3, 2))
Incorrect:
>>> a[1, :, [0, 1]].shape
(2, 3)
Correct:
>>> a[1, :, :].shape
(3, 2)
>>> a[:, 1, :].shape
(4, 2)
>>> a[:, 1, [0, 1]].shape
(4, 2)
The incorrect result is the transpose of the correct one:
>>> (a[1, :, [0, 1]] == a[1, :, :].T).all()
True
Activity
jaimefrio commentedon Sep 26, 2015
This is actually the documented behavior: the first and last index are considered to be fancy indices, and when more than one non-contiguous axis is fancy-indexed, the result of indexing those axes is grouped before any of the non-fancy indexed axes.
It's confusing, and there is an ongoing discussion to provide a more intuitive indexing, see #6075 and #6256, but it is not a bug, but a feature!
I'm going to close this, but feel free to continue the discussion if you need clarification.