pytorch归一化

torch.nn.functional.normalize

a1=torch.nn.functional.normalize(a,p=1, dim=1)
a1
Out[283]: 
tensor([[[ 0.2748, -0.9671,  0.3415, -0.1457],
         [-0.2564,  0.0282, -0.5989,  0.6267],
         [-0.4687, -0.0047, -0.0596, -0.2275]],
        [[ 0.0069,  0.3461,  0.3227,  0.3084],
         [-0.5909, -0.5238, -0.1853, -0.2525],
         [-0.4022,  0.1301, -0.4920,  0.4391]]])
a1=torch.nn.functional.normalize(a,p=2, dim=0)
a1
Out[285]: 
tensor([[[ 0.9993, -0.8100,  0.6330, -0.7194],
         [-0.2815,  0.0266, -0.9283,  0.9835],
         [-0.6189, -0.0179, -0.0932, -0.7505]],
        [[ 0.0371,  0.5865,  0.7741,  0.6946],
         [-0.9596, -0.9996, -0.3718, -0.1808],
         [-0.7855,  0.9998, -0.9956,  0.6609]]])

softmax

b=torch.randn((2,3,4))

b
Out[345]: 
tensor([[[ 0.9075,  1.3613, -0.9740, -1.2543],
         [ 0.4870, -0.0040, -1.5290, -0.1971],
         [ 0.3198,  2.0391, -0.6837,  0.3684]],
         
        [[ 0.2504, -0.1802, -0.9951,  0.6733],
         [-1.0617, -1.4770, -1.3032,  0.0329],
         [ 1.1119, -0.5354,  0.6401, -0.1070]]])
         
b1=torch.nn.functional.softmax(b, dim=0)
b1
Out[347]: 
tensor([[[0.6586, 0.8237, 0.5053, 0.1270],
         [0.8247, 0.8135, 0.4438, 0.4427],
         [0.3117, 0.9292, 0.2102, 0.6167]],
         
        [[0.3414, 0.1763, 0.4947, 0.8730],
         [0.1753, 0.1865, 0.5562, 0.5573],
         [0.6883, 0.0708, 0.7898, 0.3833]]])

0.6586+0.3414
Out[351]: 1.0

a
Out[354]: 
tensor([[[ 1.,  2.,  3.],
         [ 4.,  5.,  6.]],
         
        [[ 7.,  8.,  9.],
         [10., 11., 12.]]])

a1=torch.nn.functional.softmax(a, dim=0)
a1
Out[356]: 
tensor([[[0.0025, 0.0025, 0.0025],
         [0.0025, 0.0025, 0.0025]],

        [[0.9975, 0.9975, 0.9975],
         [0.9975, 0.9975, 0.9975]]])


a1= torch.nn.functional.softmax(a, dim=1)
a1
Out[337]: 
tensor([[[0.0474, 0.0474, 0.0474],
         [0.9526, 0.9526, 0.9526]],
     
        [[0.0474, 0.0474, 0.0474],
         [0.9526, 0.9526, 0.9526]]])
0.9526+0.0474
Out[338]: 1.0
# 按列求softmax，则每一行之和为1。


a1= torch.nn.functional.softmax(a, dim=2)
a1
Out[333]: 
tensor([[[0.0900, 0.2447, 0.6652],
         [0.0900, 0.2447, 0.6652]],
         
        [[0.0900, 0.2447, 0.6652],
         [0.0900, 0.2447, 0.6652]]]) 
# 按行求softmax，则每一行之和为1。