Pytorch里面的X.view(-1)操作

Pytorch里面的X.view(-1)操作

import torch
a = torch.randn(3,5,2)
print(a)
print(a.view(-1))

tensor([[[-0.6887,  0.2203],
         [-1.6103, -0.7423],
         [ 0.3097, -2.9694],
         [ 1.2073, -0.3370],
         [-0.5506,  0.4753]],

        [[-1.3605,  1.9303],
         [-1.5382, -1.0865],
         [-0.9208, -0.1754],
         [ 0.1476, -0.8866],
         [ 0.4519,  0.2771]],

        [[ 0.6662,  1.1027],
         [-0.0912, -0.6284],
         [-1.0253, -0.3542],
         [ 0.6909, -1.3905],
         [-2.1140,  1.3426]]])
tensor([-0.6887,  0.2203, -1.6103, -0.7423,  0.3097, -2.9694,  1.2073, -0.3370,
        -0.5506,  0.4753, -1.3605,  1.9303, -1.5382, -1.0865, -0.9208, -0.1754,
         0.1476, -0.8866,  0.4519,  0.2771,  0.6662,  1.1027, -0.0912, -0.6284,
        -1.0253, -0.3542,  0.6909, -1.3905, -2.1140,  1.3426])

结论

X.view(-1)中的-1本意是根据另外一个数来自动调整维度，但是这里只有一个维度，因此就会将X里面的所有维度数据转化成一维的，并且按先后顺序排列。