block mean of 2D numpy array (in both dimensions)

Question

This question is related to Block mean of numpy 2D array (in fact the title is almost the same!) except that my case is a generalization. I want to divide a 2D array into a sub-blocks in both directions and take the mean over the blocks. (The linked example only divides the array in one dimension).

Thus if my array is this:

import numpy as np 
a=np.arange(16).reshape((4,4))

array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15]])

If my sub-blocks have a size 2x2, then my desired answer is

array([[ 2.5,  4.5],
       [10.5, 12.5]])

The only way I could think of doing this was to carefully reshape on one dimension at a time:

np.mean(np.mean(a.reshape((2,2,-1)),axis=1).reshape((-1,2,2)),axis=2)

This gives the correct solution but is a bit of a convoluted mess, and I was wondering if there is a cleaner easier code to do the same thing, maybe some numpy blocking function that I am unaware of ?

score 2 · Accepted Answer · edited Jul 22 '22 at 09:25

2

You can do:

# sample data
a=np.arange(24).reshape((6,4))

rows, cols = a.shape    
a.reshape(rows//2, 2, cols//2, 2).mean(axis=(1,-1))

Output:

array([[ 2.5,  4.5],
       [10.5, 12.5],
       [18.5, 20.5]])

edited Jul 22 '22 at 09:25

divenex

15,176
9
55
55

answered Feb 11 '21 at 18:39

Quang Hoang

146,074
10
56
74

double axis mean, of course (doh!) - nice the double forward slash for int(n/m) - I didn't know that either... thanks! – ClimateUnboxed Feb 11 '21 at 19:47
1

See here for a generalization to N-dimensional arrays https://stackoverflow.com/a/73078468/3753826 – divenex Jul 22 '22 at 10:07
@Quang my function `blockwise_average_ND` was generalized for any dimension as stated, not just 3D as you write. But if you found a bug please let me know. – divenex Jul 22 '22 at 11:57
1

Ah, I see. I was looking at the accepted answer. – Quang Hoang Jul 22 '22 at 11:58

block mean of 2D numpy array (in both dimensions)

1 Answers1

Linked