Grouping 2D numpy array in average

Question

I am trying to group a numpy array into smaller size by taking average of the elements. Such as take average foreach 5x5 sub-arrays in a 100x100 array to create a 20x20 size array. As I have a huge data need to manipulate, is that an efficient way to do that?

Answer 1

I have tried this for smaller array, so test it with yours:

import numpy as np  nbig = 100 nsmall = 20 big = np.arange(nbig * nbig).reshape([nbig, nbig]) # 100x100  small = big.reshape([nsmall, nbig//nsmall, nsmall, nbig//nsmall]).mean(3).mean(1) 

An example with 6x6 -> 3x3:

nbig = 6 nsmall = 3 big = np.arange(36).reshape([6,6]) array([[ 0,  1,  2,  3,  4,  5],        [ 6,  7,  8,  9, 10, 11],        [12, 13, 14, 15, 16, 17],        [18, 19, 20, 21, 22, 23],        [24, 25, 26, 27, 28, 29],        [30, 31, 32, 33, 34, 35]])  small = big.reshape([nsmall, nbig//nsmall, nsmall, nbig//nsmall]).mean(3).mean(1)  array([[  3.5,   5.5,   7.5],        [ 15.5,  17.5,  19.5],        [ 27.5,  29.5,  31.5]]) 

Answer 2

This is pretty straightforward, although I feel like it could be faster:

from __future__ import division
import numpy as np
Norig = 100
Ndown = 20
step = Norig//Ndown
assert step == Norig/Ndown # ensure Ndown is an integer factor of Norig
x = np.arange(Norig*Norig).reshape((Norig,Norig)) #for testing
y = np.empty((Ndown,Ndown)) # for testing
for yr,xr in enumerate(np.arange(0,Norig,step)):
    for yc,xc in enumerate(np.arange(0,Norig,step)):
        y[yr,yc] = np.mean(x[xr:xr+step,xc:xc+step])

You might also find scipy.signal.decimate interesting. It applies a more sophisticated low-pass filter than simple averaging before downsampling the data, although you'd have to decimate one axis, then the other.