Splitting numpy array into blocks

Question

I've got a 900 x 650 2D numpy array which I'd like to split into 10 x 10 blocks, which will be checked for nonzero elements. Is there a Pythonic way that I can achieve this with numpy?

I'm looking for functionality similar to the following:

blocks_that_have_stuff = []
my_array = getArray()
my_array.cut_into_blocks((10, 10))
for block_no, block in enumerate(my_array):
    if numpy.count_nonzero(block) > 5:
        blocks_that_have_stuff.append(block_no)

how are your blocks supposed to be set up? is it all possible 10x10 blocks? Just non-overlapping 10x10 blocks. — Jeff Tratner, Jun 29 '13 at 22:18
This stackoverflow answer seems to contain a nice way of splitting a mtrix into blocks: http://stackoverflow.com/a/5078155/204218 — ase, Jun 29 '13 at 22:22

score 8 · Accepted Answer · answered Jun 29 '13 at 23:52

I wrote a routine that cut your matrix in blocks. The example is very easy to understand. I wrote it in an easy form to display the result (only for checking purpose). If you are interested in it, you could include in the output the number of blocks or anything.

import matplotlib.pyplot as plt
import numpy as np

def cut_array2d(array, shape):
    arr_shape = np.shape(array)
    xcut = np.linspace(0,arr_shape[0],shape[0]+1).astype(np.int)
    ycut = np.linspace(0,arr_shape[1],shape[1]+1).astype(np.int)
    blocks = [];    xextent = [];    yextent = []
    for i in range(shape[0]):
        for j in range(shape[1]):
            blocks.append(array[xcut[i]:xcut[i+1],ycut[j]:ycut[j+1]])
            xextent.append([xcut[i],xcut[i+1]])
            yextent.append([ycut[j],ycut[j+1]])
    return xextent,yextent,blocks

nx = 900; ny = 650
X, Y = np.meshgrid(np.linspace(-5,5,nx), np.linspace(-5,5,ny))
arr = X**2+Y**2

x,y,blocks = cut_array2d(arr,(10,10))

n = 0

for x,y,block in zip(x,y,blocks):
    n += 1
    plt.imshow(block,extent=[y[0],y[1],x[0],x[1]],
               interpolation='nearest',origin='lower',
               vmin = arr.min(), vmax=arr.max(),
               cmap=plt.cm.Blues_r)
    plt.text(0.5*(y[0]+y[1]),0.5*(x[0]+x[1]),str(n),
             horizontalalignment='center',
             verticalalignment='center')

plt.xlim([0,900])
plt.ylim([0,650])
plt.savefig("blocks.png",dpi=72)
plt.show()

The output is:

enter image description here

Regards

Note: I think you could optimize this routine using np.meshgrid instead a lot of appends with the xextent & yextent.

Amazing, exactly what I was looking for. I wish I could rep you more for this incredible answer. — Dan Doe, Jun 30 '13 at 00:10
@DanDoe, I believe that you can start a bounty (2 days after you posted) on your question, and award it to the above answer. — Akavall, Jun 30 '13 at 02:19

Splitting numpy array into blocks

1 Answers1