[FIXED] Split a numpy array into nonoverlapping arrays

Issue

I am trying to split a 2D numpy array that is not square into nonoverlapping chunks of smaller 2D numpy arrays. Example - split a 3x4 array into chunks of 2x2:

The array to be split:

[[ 34  15  16  17]
 [ 78  98  99 100]
 [ 23  78  79  80]]

This should output:

[[34 15]
 [78 98]]

[[ 16  17]
 [ 99 100]]

So [23 78 79 80] are dropped because they do not match the 2x2 requirement.

My current code is this:

new_array = np.array([[34,15,16,17], [78,98,99,100], [23,78,79,80]])
window = 2
for x in range(0, new_array.shape[0], window):
    for y in range(0, new_array.shape[1], window):
        patch_im1 = new_array[x:x+window,y:y+window]

This outputs:

[[34 15]
 [78 98]]
[[ 16  17]
 [ 99 100]]
[[23 78]]
[[79 80]]

Ideally, I would like to have the chunks stored in a list.

Solution

This should work on any number of dimensions. Let's take an array:

new_array = np.array(range(25)).reshape((5,5))

Output:

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]])

First calculate the number of last rows/columns which you don't need:

N = 2
rem = np.array(new_array.shape) % N

Output:

array([1, 1])

Then remove the number of rows/columns from the end of your array on each dimension:

for ax, v in enumerate(rem):
if v != 0:
    new_array = np.delete(new_array, range(-1, -v-1, -1), axis=ax)

Output:

array([[ 0,  1,  2,  3],
       [ 5,  6,  7,  8],
       [10, 11, 12, 13],
       [15, 16, 17, 18]])

Then use np.split on each dimension:

arr_list = [new_array]
for ax in range(len(new_array.shape)):
    arr_list_list = [np.split(arr, arr.shape[ax] / N, axis=ax) for arr in arr_list]
    arr_list = [arr for j in arr_list_list for arr in j]

Output:

[array([[0, 1],
        [5, 6]]),
 array([[2, 3],
        [7, 8]]),
 array([[10, 11],
        [15, 16]]),
 array([[12, 13],
        [17, 18]])]

Then transform into a list:

[list(i.reshape(i.size)) for i in arr_list]

Output:

[[0, 1, 5, 6], [2, 3, 7, 8], [10, 11, 15, 16], [12, 13, 17, 18]]

Answered By - AndrzejO