Given a matrix Z whose dimensions are multiples of k, return the matrix of block sums:

• partition Z into non-overlapping k×k blocks
• the result has shape (Z.shape[0] // k, Z.shape[1] // k)

Input

Z = np.ones((16, 16))
k = 4

Output

[[16. 16. 16. 16.]
 [16. 16. 16. 16.]
 [16. 16. 16. 16.]
 [16. 16. 16. 16.]]