[人工智能]pytorch view()、transpose()和permute()的区别

transpose()和permute()的不同
1、torch.transpose()是交换指定的两个维度的内容，permute()则可以一次性交换多个维度,代码示例如下

a = torch.tensor([[[1, 2, 3, 4], [4, 5, 6, 7]], [[7, 8, 9, 10], [10, 11, 12, 13]], [[13, 14, 15, 16], [17, 18, 19, 20]]])
print(a, a.shape)

结果输出：tensor([[[ 1,  2,  3,  4],
         			[ 4,  5,  6,  7]],

        			[[ 7,  8,  9, 10],
         			[10, 11, 12, 13]],

        			[[13, 14, 15, 16],
         			[17, 18, 19, 20]]]) torch.Size([3, 2, 4])


b = a.transpose(1,2)
print(b, b.shape)

结果输出：tensor([[[ 1,  4],
         [ 2,  5],
         [ 3,  6],
         [ 4,  7]],

        [[ 7, 10],
         [ 8, 11],
         [ 9, 12],
         [10, 13]],

        [[13, 17],
         [14, 18],
         [15, 19],
         [16, 20]]]) torch.Size([3, 4, 2])

c = a.permute(2, 1, 0)
print(c, c.shape)

结果输出：tensor([[[ 1,  7, 13],
         [ 4, 10, 17]],

        [[ 2,  8, 14],
         [ 5, 11, 18]],

        [[ 3,  9, 15],
         [ 6, 12, 19]],

        [[ 4, 10, 16],
         [ 7, 13, 20]]]) torch.Size([4, 2, 3])

view和 transpose()、permute()的不同
2、如上所述，transpose和permute是将张量的维度进行变换，而view是将张量拉伸成一维，然后根据传入的维度（也就是想要变换的维度），重构出一个新的张量。代码实例如下所示

a = torch.tensor([[[1, 2, 3, 4], [4, 5, 6, 7]], [[7, 8, 9, 10], [10, 11, 12, 13]], [[13, 14, 15, 16], [17, 18, 19, 20]]])
print(a, a.shape)

结果输出：tensor([[[ 1,  2,  3,  4],
         			[ 4,  5,  6,  7]],

        			[[ 7,  8,  9, 10],
         			[10, 11, 12, 13]],

        			[[13, 14, 15, 16],
         			[17, 18, 19, 20]]]) torch.Size([3, 2, 4])
d = a.view(-1, 4)
print(d, d.shape)
结果输出：tensor([[ 1,  2,  3,  4],
        [ 4,  5,  6,  7],
        [ 7,  8,  9, 10],
        [10, 11, 12, 13],
        [13, 14, 15, 16],
        [17, 18, 19, 20]]) torch.Size([6, 4])

通过以上代码可以看出其区别之处