1 torch.reshape

形式

torch.reshape(input, shape)

功能

返回一个与输入张量数据和元素数相同的,但是形状为shape的张量。

参数

  • input:需要被重新定义形状的输入张量
  • shape:新的形状

使用示例

import torch

if __name__ == '__main__':
    a = torch.arange(0, 16)
    print(a)

    b = torch.reshape(a, shape=(4, 4))
    print(b)

输出

tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15])
tensor([[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11],
        [12, 13, 14, 15]])

2 torch.view

形式

Tensor.view(*shape)

功能

返回一个新的张量,其数据与自张量相同,但形状不同。

返回的张量与输入向量共享相同的数据,具有相同数量的元素,但是可能拥有不同的形状。

参数

  • shape:需要修改的形状

使用示例

import torch

if __name__ == '__main__':
    a = torch.arange(0, 16)
    print(a)

    b = a.view(4, 4)
    print(b)

输出

tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15])
tensor([[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11],
        [12, 13, 14, 15]])

3 reshape和view二者的区别

Pytorchreshapeview这两个函数的作用是一模一样,但是两者还是有区别的。

在Pytorch的关于view的官方文档有这么一段话

When it is unclear whether a view() can be performed, it is advisable to use reshape(), which returns a view if the shapes are compatible, and copies (equivalent to calling contiguous()) otherwise.

意思是当你搞不清是不是可以使用view时,那么建议使用reshape

区别:

  • view只能用在contiguous的张量上,但是reshape没有这个限制;

  • 如果在view之前使用了transpose或者permute等操作,那么需要接一个.contiguous()操作返回一个contiguous copy,才能使用view算子。

  • view输出的Tensor与输入Tensor共享内存。

简单的说,tensor.reshape可以看做是tensor.contiguous().view,所以尽可能的使用reshape,而不是view