python - How to split a numpy array in fixed size chunks with and without overlap?

Question

Lets say I have an array:

>>> arr = np.array(range(9)).reshape(3, 3)
>>> arr
array([[0, 1, 2],
       [3, 4, 5],
       [6, 7, 8]])

I would like to create a function f(arr, shape=(2, 2)) that takes the array and a shape, and splits the array into chunks of the given shape without padding. Thus, by overlapping certain parts if necessary. For example:

>>> f(arr, shape=(2, 2))
array([[[[0, 1],
         [3, 4]],

        [[1, 2],
         [4, 5]]],

       [[[3, 4],
         [6, 7]],

        [[4, 5],
         [7, 8]]]])

I managed to creates to output above with np.lib.stride_tricks.as_strided(arr, shape=(2, 2, 2, 2), strides=(24, 8, 24, 8)). But I don't know how to generalize this for to all arrays and all chunk sizes.

Preferably, for 3D arrays.

If no overlap is necessary, it should avoid that. Another example:

>>> arr = np.array(range(16).reshape(4,4)
>>> arr
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15]])
>>> f(arr, shape=(2,2))
array([[[[0, 1],
         [4, 5]],

        [[2, 3],
         [6, 7]]],

       [[[8, 9],
         [12, 13]],

        [[10, 11],
         [14, 15]]]])

skimage.util.view_as_blocks comes close, but requires that the array and block shape are compatible.

Answer 1

There's a builtin in scikit-image as view_as_windows for doing exactly that -

from skimage.util.shape import view_as_windows

view_as_windows(arr, (2,2))

Sample run -

In [40]: arr
Out[40]: 
array([[0, 1, 2],
       [3, 4, 5],
       [6, 7, 8]])

In [41]: view_as_windows(arr, (2,2))
Out[41]: 
array([[[[0, 1],
         [3, 4]],

        [[1, 2],
         [4, 5]]],


       [[[3, 4],
         [6, 7]],

        [[4, 5],
         [7, 8]]]])

---

For the second part, use its cousin from the same family/module view_as_blocks -

from skimage.util.shape import view_as_blocks

view_as_blocks(arr, (2,2))