split a numpy array both horizontally and vertically

Here's one approach staying within NumPy environment -

def view_as_blocks(arr, BSZ):
    # arr is input array, BSZ is block-size
    m,n = arr.shape
    M,N = BSZ
    return arr.reshape(m//M, M, n//N, N).swapaxes(1,2).reshape(-1,M,N)

Sample runs

1) Actual big case to verify shapes :

In [41]: aa = np.reshape(np.arange(270),(18,15))

In [42]: view_as_blocks(aa, (9,5)).shape
Out[42]: (6, 9, 5)

2) Small case to manually verify values:

In [43]: aa = np.reshape(np.arange(36),(6,6))

In [44]: aa
Out[44]: 
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]])

In [45]: view_as_blocks(aa, (2,3)) # Blocks of shape (2,3)
Out[45]: 
array([[[ 0,  1,  2],
        [ 6,  7,  8]],

       [[ 3,  4,  5],
        [ 9, 10, 11]],

       [[12, 13, 14],
        [18, 19, 20]],

       [[15, 16, 17],
        [21, 22, 23]],

       [[24, 25, 26],
        [30, 31, 32]],

       [[27, 28, 29],
        [33, 34, 35]]])

If you are willing to work with other libraries, scikit-image could be of use here, like so -

from skimage.util import view_as_blocks as viewB

out = viewB(aa, tuple(BSZ)).reshape(-1,*BSZ)

Runtime test -

In [103]: aa = np.reshape(np.arange(270),(18,15))

# @EFT's soln
In [99]: %timeit split_2d(aa, (2,3))
10000 loops, best of 3: 23.3 µs per loop

# @glegoux's soln-1
In [100]: %timeit list(get_chunks(aa, 2,3))
100000 loops, best of 3: 3.7 µs per loop

# @glegoux's soln-2
In [111]: %timeit list(get_chunks2(aa, 9, 5))
100000 loops, best of 3: 3.39 µs per loop

# Proposed in this post
In [101]: %timeit view_as_blocks(aa, (9,5))
1000000 loops, best of 3: 1.86 µs per loop

Please note that I have used (2,3) for split_2d and get_chunks as by their definitions, they are using that as the number of blocks. In my case with view_as_blocks, I have the parameter BSZ indicating the block size. So, I have (9,5) there. get_chunks2 follows the same format as view_as_blocks. The outputs should represent the same there.


You could use np.split & np.concatenate, the latter to allow the second split to be conducted in a single step:

def split_2d(array, splits):
    x, y = splits
    return np.split(np.concatenate(np.split(array, y, axis=1)), x*y)

ab = split_2d(aa,(2,3))

ab[0].shape
Out[95]: (9, 5)

len(ab)
Out[96]: 6

This also seems like it should be relatively straightforward to generalize to the n-dim case, though I haven't followed that thought all the way through just yet.

Edit:

For a single array as output, just add np.stack:

np.stack(ab).shape
Out[99]: (6, 9, 5)

To cut, this matrix (18,15) :

+-+-+-+
+     +
+-+-+-+

in 2x3 blocks (9,5) like it :

+-+-+-+
+-+-+-+
+-+-+-+

Do:

from pprint import pprint
import numpy as np

M = np.reshape(np.arange(18*15),(18,15))

def get_chunks(M, n, p):
    n = len(M)//n
    p = len(M[0])//p
    for i in range(0, len(M), n):
        for j in range(0, len(M[0]), p):
            yield M[i:i+n,j:j+p]

def get_chunks2(M, n, p):
        for i in range(0, len(M), n):
            for j in range(0, len(M[0]), p):
                yield M[i:i+n,j:j+p]

# list(get_chunks2(M, 9, 5)) same result more faster
chunks = list(get_chunks(M, 2, 3))

pprint(chunks)

Output:

[array([[  0,   1,   2,   3,   4],
       [ 15,  16,  17,  18,  19],
       [ 30,  31,  32,  33,  34],
       [ 45,  46,  47,  48,  49],
       [ 60,  61,  62,  63,  64],
       [ 75,  76,  77,  78,  79],
       [ 90,  91,  92,  93,  94],
       [105, 106, 107, 108, 109],
       [120, 121, 122, 123, 124]]),
 array([[  5,   6,   7,   8,   9],
       [ 20,  21,  22,  23,  24],
       [ 35,  36,  37,  38,  39],
       [ 50,  51,  52,  53,  54],
       [ 65,  66,  67,  68,  69],
       [ 80,  81,  82,  83,  84],
       [ 95,  96,  97,  98,  99],
       [110, 111, 112, 113, 114],
       [125, 126, 127, 128, 129]]),
 array([[ 10,  11,  12,  13,  14],
       [ 25,  26,  27,  28,  29],
       [ 40,  41,  42,  43,  44],
       [ 55,  56,  57,  58,  59],
       [ 70,  71,  72,  73,  74],
       [ 85,  86,  87,  88,  89],
       [100, 101, 102, 103, 104],
       [115, 116, 117, 118, 119],
       [130, 131, 132, 133, 134]]),
 array([[135, 136, 137, 138, 139],
       [150, 151, 152, 153, 154],
       [165, 166, 167, 168, 169],
       [180, 181, 182, 183, 184],
       [195, 196, 197, 198, 199],
       [210, 211, 212, 213, 214],
       [225, 226, 227, 228, 229],
       [240, 241, 242, 243, 244],
       [255, 256, 257, 258, 259]]),
 array([[140, 141, 142, 143, 144],
       [155, 156, 157, 158, 159],
       [170, 171, 172, 173, 174],
       [185, 186, 187, 188, 189],
       [200, 201, 202, 203, 204],
       [215, 216, 217, 218, 219],
       [230, 231, 232, 233, 234],
       [245, 246, 247, 248, 249],
       [260, 261, 262, 263, 264]]),
 array([[145, 146, 147, 148, 149],
       [160, 161, 162, 163, 164],
       [175, 176, 177, 178, 179],
       [190, 191, 192, 193, 194],
       [205, 206, 207, 208, 209],
       [220, 221, 222, 223, 224],
       [235, 236, 237, 238, 239],
       [250, 251, 252, 253, 254],
       [265, 266, 267, 268, 269]])]