torch.zeros_like() 和 torch.zeros()的区别

• torch.zeros_like(tensor)：是根据给定张量，生成与其形状相同的全0张量

• torch.zeros(size)：其形状由变量参数size定义，返回一个由标量值0填充的张量

例子：

import torch
input = torch.rand(2, 3)
print(input)
# 生成与input形状相同、元素全为0的张量
a = torch.zeros_like(input)
print(a)

# 报错：类型错误TypeError
b = torch.zeros(input)
print(b)

# 生成大小为【3】的0张量
c = torch.zeros(3)
print(c)

# 生成大小为【2，3】的0张量
d = torch.zeros([2,3])
print(d)

结果：

# input = torch.rand(2, 3)
tensor([[0.3517, 0.1117, 0.2550],
        [0.7272, 0.2602, 0.9998]])
        
# a = torch.zeros_like(input)
tensor([[0., 0., 0.],
        [0., 0., 0.]])
        
# torch.zeros(input)
Traceback (most recent call last):
  File "", line 9, in <module>
TypeError: zeros(): argument 'size' (position 1) must be tuple of ints, not Tensor
    
# c = torch.zeros(3)    
tensor([0., 0., 0.])

# d = torch.zeros([2,3])
tensor([[0., 0., 0.],
        [0., 0., 0.]])
        

参考链接：

1.https://blog.csdn.net/tszupup/article/details/108130721

2.https://blog.csdn.net/weixin_43979572/article/details/86482618