pytorch中 .view()函数讲解

一、普通用法 (手动调整size)

view()相当于reshape、resize，重新调整Tensor的形状。

import torch
a1 = torch.arange(0,16)
print(a1)
# tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15])

a2 = a1.view(8, 2)
a3 = a1.view(2, 8)
a4 = a1.view(4, 4)
print(a2)
#tensor([[ 0,  1],
#        [ 2,  3],
#        [ 4,  5],
#        [ 6,  7],
#        [ 8,  9],
#        [10, 11],
#        [12, 13],
#        [14, 15]])
print(a3)
#tensor([[ 0,  1,  2,  3,  4,  5,  6,  7],
#        [ 8,  9, 10, 11, 12, 13, 14, 15]])
print(a4)
#tensor([[ 0,  1,  2,  3],
#        [ 4,  5,  6,  7],
#        [ 8,  9, 10, 11],
#        [12, 13, 14, 15]])

二、特殊用法：参数-1 (自动调整size)

view中一个参数定为-1，代表自动调整这个维度上的元素个数，以保证元素的总数不变。

v1 = torch.arange(0,16)
print(v1)
# tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15])
v2 = v1.view(-1, 16)
v2
# tensor([[ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15]])
v2 = v1.view(-1, 8)
v2
# tensor([[ 0,  1,  2,  3,  4,  5,  6,  7],
#         [ 8,  9, 10, 11, 12, 13, 14, 15]])
v2 = v1.view(-1, 4)
v2
#tensor([[ 0,  1,  2,  3],
#        [ 4,  5,  6,  7],
#        [ 8,  9, 10, 11],
#        [12, 13, 14, 15]])
v2 = v1.view(-1, 2)
v2
#tensor([[ 0,  1],
#        [ 2,  3],
#        [ 4,  5],
#        [ 6,  7],
#        [ 8,  9],
#        [10, 11],
#        [12, 13],
#        [14, 15]])

v3 = v1.view(4*4, -1)
v3
# tensor([[ 0],
#         [ 1],
#         [ 2],
#         [ 3],
#         [ 4],
#         [ 5],
#         [ 6],
#         [ 7],
#         [ 8],
#         [ 9],
#         [10],
#         [11],
#         [12],
#         [13],
#         [14],
#         [15]])