torch.norm()与F.normalize()的区别

两者区别

F.normalize()在torch.norm()的基础上进行了标准化，即：

F.normalize(x) = x / torch.norm(x)

代码示例

import torch
import torch.nn.functional as F


x = torch.tensor([[2.3]]).float()
r1 = x.norm(2)
r2 = F.normalize(x)

## 结果
r1: tensor(3.6056)
r2: tensor([[0.5547, 0.8321]]) = tensor([[2., 3.]]) / tensor(3.6056)