Split ndarray into smaller ndarray stored in a list

Question

I currently looking for method in which i can split a ndarray into smaller ndarrays.

example: given ndarray with shape (78,1440,3), from which i want to extract a list of smaller ndarrays of the size (78,72,3), that would be 20 smaller sub ndarrays.

I tried using numpy.split .

numpy.split(matrix,72,axis=1)

which generates a list of length 72 and the first entry has the shape (78,20,3)..

Why am I not able to extract the size i need?

Answer 1

The 72 in the split is the number of elements to split it into, not the size of the splitted dimensions (according to the axis).

You can however use:

numpy.split(matrix,,axis=1)

to split it into 20 elements of length 72 (for your given case). Note that you have to ensure that the shape[1] is dividable by 72 otherwise a ValueError will be raised.

Answer 2

Approach #1 : You can use np.hsplit made exactly for this task -

np.hsplit(arr,20) # creates list of 20 arrays 

Sample run -

1) Input array :

In [52]: a = np.random.randint(0,9,(2,6,3))

In [53]: a
Out[53]: 
array([[[7, 8, 8],
        [7, 7, 1],
        [1, 6, 4],
        [6, 3, 8],
        [4, 7, 4],
        [0, 6, 3]],

       [[0, 8, 5],
        [2, 2, 8],
        [6, 0, 7],
        [5, 4, 6],
        [4, 3, 1],
        [8, 6, 6]]])

2) Split axis=1 into 3 parts , thus each part/subarray would be of length (2,2,3) shape. Thus, we would get a list of those 3 arrays :

In [54]: b = np.hsplit(a,3)

3) Manually verify those parts :

In [55]: b[0]
Out[55]: 
array([[[7, 8, 8],
        [7, 7, 1]],

       [[0, 8, 5],
        [2, 2, 8]]])

In [56]: b[1]
Out[56]: 
array([[[1, 6, 4],
        [6, 3, 8]],

       [[6, 0, 7],
        [5, 4, 6]]])

In [57]: b[2]
Out[57]: 
array([[[4, 7, 4],
        [0, 6, 3]],

       [[4, 3, 1],
        [8, 6, 6]]])

Approach #2 : Another tool for this task would be np.array_split -

np.array_split(arr,20,axis=1)