import numpy as np
a = np.random.random_sample((3,1,3))
b  = np.random.random_sample((2,3))
c = a-b
c = np.square(c)
c = np.sum(c,axis=2)
c= np.sqrt(c)
a1 = a[1,:,:]
b1 = b[1,:]
print(a1,'5')
print(b1,'6')
print(np.square(a1-b1).shape)
print(np.sum(np.square(a1-b1),axis=1),'7')
print(np.sqrt(np.sum(np.square(a1-b1),axis=1)))

a.shape = (1,2,3,4)

numpy.sum(a,axis = 0).shape = (2,3,4)
numpy .sum(a, axis =1).shape = (1,3,4)

numpy.sum(a, axis=None, dtype=None, out=None, keepdims=<no value>, initial=<no value>, where=<no value>)
'''
keepdimsbool, optional
If this is set to True, the axes which are reduced are left in the result as dimensions with size one. With this option, the result will broadcast correctly against the input array.
If the default value is passed, then keepdims will not be passed through to the sum method of sub-classes of ndarray, however any non-default value will be. If the sub-class' method does not implement keepdims any exceptions will be raised.
'''

a = np.array([[0, 0, 0],
       [0, 1, 0],
       [0, 2, 0],
       [1, 0, 0],
       [1, 1, 0]])
print(a)
'''
輸出：
[[0 0 0]
 [0 1 0]
 [0 2 0]
 [1 0 0]
 [1 1 0]]
'''

a_sum_true = np.sum(a, keepdims=True)
print(a_sum_true)
print(a_sum_true.shape)
a_sum_false = np.sum(a, keepdims=False)
print(a_sum_false)
print(a_sum_false.shape)
'''
輸出：
[[6]]
(1, 1)
6
()
'''

a_sum_axis1_true = np.sum(a, axis=1, keepdims=True)
print(a_sum_axis1_true)
print(a_sum_axis1_true.shape)
a_sum_axis1_false = np.sum(a, axis=1, keepdims=False)
print(a_sum_axis1_false)
print(a_sum_axis1_false.shape)
'''
輸出：
[[0]
 [1]
 [2]
 [1]
 [2]]
(5, 1)
[0 1 2 1 2]
(5,)
'''

a_sum_axis0_true = np.sum(a, axis=0, keepdims=True)
print(a_sum_axis0_true)
print(a_sum_axis0_true.shape)
a_sum_axis0_false = np.sum(a, axis=0, keepdims=False)
print(a_sum_axis0_false)
print(a_sum_axis0_false.shape)
'''
輸出：
[[2 4 0]]
(1, 3)
[2 4 0]
(3,)
'''