+= Operator on list of indices - unexpected behaviour

Hi, I stumbled upon an interesting and unexpected (to me) behaviour when using the += operator on a np.ndarray:

import numpy as np

a = np.zeros(3)

indices = [0, 0, 1, 1, 2]
a[indices] += 1

The thing is, knowing that += avoids creating copies and that slicing also creates views and not copies, I would expect this to add 2 to a[0], 2 to a[1] and 1 to a[2]. Instead I witness the following:

>>> a
array([1., 1., 1.])

Is this a bug? If not, can you please explain this?

Thanks, Michael

[eric-wieser]

Numpy overloads c += b in a way that uses less memory that c = c + b, but python provides no way to overload the combined indexing and augmented assignment.

[mishana]

@mishana: performance differences aren’t really relevant when one of your options doesn’t work 😉

Well in this case it is relevant, since the indices are unique and they basically produce the same result.