The difference between torch.nn and torch.nn.functional in pytorch

Link to the original text

torch.nn

pytorch Chinese document links ：torch.nn

stay __init__() The function defines , It defines a class ：

torch.nn.functional

pytorch Chinese document links ：torch.nn.functional

stay __forward()__ The function defines , What is defined is a function ：

The difference between the two

torch.nn Is a defined class , With class xx To define the , You can extract varying learning parameters .

torch.nn.functional It's a function , from def function( ) Definition , Is a fixed formula .

In the deep learning, there will be many weights that are constantly updated , So we need to adopt the method of class , To ensure that we can still use the operation steps we set before when the parameters change . So if the model has learnable parameters , You should use nn.Module, Otherwise there is no difference between the two . But simple calculation does not need to create a new class to do , So use nn.functional Define a function .

namely ： Within the layer variable For nn Definition , Otherwise, use nn.functional Definition .

Learn more about programming , Please pay attention to my official account ：

The way of code