pytorch之expand和repeat

1. expand

没有重新分配内存,只是原有数据的引用

import torch

x = torch.tensor([[1], [2], [3]])
# 维度(3,1)
new_x = x.expand(-1, 4)  # -1 代表维度不变,即扩展后维度是(3,4)
# 等价于 new_x = x.expand(3, 4) 

print(x)
print(new_x)

x[0][0] = 15
# 修改x的值,new_x对应的值也会改变
print(x)
print(new_x)

输出:

tensor([[1],
        [2],
        [3]])
tensor([[1, 1, 1, 1],
        [2, 2, 2, 2],
        [3, 3, 3, 3]])
tensor([[15],
        [ 2],
        [ 3]])
tensor([[15, 15, 15, 15],
        [ 2,  2,  2,  2],
        [ 3,  3,  3,  3]])

此外,x.expand_as(other) 等价于 x.expand(other.size()).

2. repeat

分配了新的内存


import torch

x = torch.tensor([1, 2, 3])

new_x = x.repeat(2, 1)
# 第一个维度重复两遍,第二个维度不变

print(x)
print(new_x)

x[0] = 15
# x改变,new_x不会改变
print(x)
print(new_x)

输出:

tensor([1, 2, 3])
tensor([[1, 2, 3],
        [1, 2, 3]])

tensor([15,  2,  3])
tensor([[1, 2, 3],
        [1, 2, 3]])

你可能感兴趣的:(pytorch,pytorch,深度学习,人工智能)