IndexError when extracting noncontiguous columns and rows from matrix

Test case:

>>> A=np.random.rand(4,4)
>>> A[(0,3),(0,2,3)]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
IndexError: shape mismatch: indexing arrays could not be broadcast together with shapes (2,) (3,)

However, A[(0,3),(0,2,3)] is a question with a very definite answer which can be obtained by calling

>>> A[(0,3),:][:,(0,2,3)]

I can’t think of any situation in which this might be ambiguous or indefinite, so if in fact none exists, I propose that numpy should return this result instead of an exception.

Thank you for your work!

[argriffing]

There’s an ix_ function that does something like this.

>>> import numpy as np
>>> a = np.random.rand(4,4)
>>> a[(0,3),:][:,(0,2,3)]
array([[ 0.84060636,  0.69065912,  0.0026522 ],
       [ 0.61781349,  0.98709272,  0.97869428]])
>>> a[np.ix_((0,3),(0,2,3))]
array([[ 0.84060636,  0.69065912,  0.0026522 ],
       [ 0.61781349,  0.98709272,  0.97869428]])

[njsmith]

Yes, this also confused me when I started using Numpy. You just have to memorize that “fancy” indexing uses a totally different rule to combine multiple indexes than slicing does. It’s much more powerful this way, granted, and it’s far too late to change now. But it is confusing.

To get what you originally wanted, you can make it so the indices do broadcast broadcast against each other:

arr[ np.reshape([1, 3, 5], (3, 1)), np.reshape([2, 4, 6], (1, 3)) ] On Feb 16, 2015 7:45 AM, “Andrea Azzini” [e-mail] wrote:

Oh ok, I didn’t know that. Neat. I still feel this is inconsistent with slicing (which uses the same “operator” to extract a submatrix), however there’s absolutely no point in changing anything, and I may be thinking too Matlab-y. Please close this bug, and thanks for your time.

— Reply to this email directly or view it on GitHub https://github.com/numpy/numpy/issues/5574#issuecomment-74527617.