一种方法是将原始形状转换为(J, N*M)形状的数组：

import numpy as np

J = 2  # size of each subvector
A = np.array([[1, 0], [2, 4], [9, 3], [7, 8]])

# Make an array where each column is a length-J subvector
# order='F' means we fill column by column
subvecs = np.reshape(A, (J, -1), order='F')

# Flag elements equal to subvector max as 1, else 0
maxes = np.max(subvecs, axis=0)
is_max = np.equal(subvecs, maxes).astype(int)

# Reshape back to the original array
B = np.reshape(is_max, A.shape, order='F')
print(B)

给出：

[[0 0]
 [1 1]
 [1 0]
 [0 1]]