How to efficiently slide numpy array?

Question

I would like to slide a ND numpy array. Currently, the code below do the work.

import numpy as np
arr = np.array([np.arange(0,16), np.arange(17,33),np.arange(33,49)])
window_size=4
expected_opt=[arr [:, i:i+window_size]for i in range(0,16,window_size)]

But I curious whether there is more efficient way to achieve similar objective.

one might suggest 1, but the solution gives different output.

Quang Hoang · Accepted Answer · 2020-11-03T15:46:14.100

1

You are looking at a reshape, not rolling:

arr.reshape(arr.shape[0],-1,window_size).transpose(1,0,2)

Output:

array([[[ 0,  1,  2,  3],
        [17, 18, 19, 20],
        [33, 34, 35, 36]],

       [[ 4,  5,  6,  7],
        [21, 22, 23, 24],
        [37, 38, 39, 40]],

       [[ 8,  9, 10, 11],
        [25, 26, 27, 28],
        [41, 42, 43, 44]],

       [[12, 13, 14, 15],
        [29, 30, 31, 32],
        [45, 46, 47, 48]]])

edited Nov 03 '20 at 15:46

answered Nov 03 '20 at 15:31

Quang Hoang

146,074
10
56
74

Hi @Quang, thanks for the suggestion. But,the result is inconsistent if, say I changed the window_size to 2. By right, the shape should be of (8,3,2) – mpx Nov 03 '20 at 15:44
1

@balandongiv thanks for pointing that out. I updated the answer. – Quang Hoang Nov 03 '20 at 15:46
Great, now window_size=2,4,8 give the intended result – mpx Nov 03 '20 at 16:04

How to efficiently slide numpy array?

1 Answers1