如何在 numpy 中重新排列 ndarray 中的列

Question

I stumbled upon what I think is a weird (or at least unintuitive) behavior of numpy and I would like to understand why it behaves that way. Let's generate a generic array of shape (4, 3, 3).

import numpy as np
arr = np.arange(4*3*3).reshape((4, 3, 3))

Thinking about arr as a list of four three-by-three matrices I want to now swap the first two columns of the first matrix in the list. I can just reorder the columns with an index list:

idx = np.array([1, 0, 2])
m = arr[0]
m[:, idx]
>>> array([[1, 0, 2],
           [4, 3, 5],
           [7, 6, 8]])

I see that i successfully swapped the two columns. However, if I do try to do same directly with arr, I get:

arr[0, :, idx]
>>> array([[1, 4, 7],
           [0, 3, 6],
           [2, 5, 8]])

I guess I'm doing something wrong but I don't understand this behavior.

Answer 1

This weird output is because when you are doing

m = arr[0]
m[:, idx]

then m becomes a whole different "array" with data of "arr" but when you are doing arr[0, :, arr] there arr is a list which has arrays