pytorch中的网络结构

Posted on Edited on In

==首先先给出一些python面向对象的一些基础知识==

子类中初始化父类的相关问题

pytorch中网络构造的深入理解

官方中文文档

pytorch.nn.module

current layer

api:

Container

torch.nn.Sequential

A sequential container.

Modules will be added to it in the order they are passed in the constructor. Alternatively, an ordered dict of modules can also be passed in.

1
2
3
4
5
6
7
# Example of using Sequential
model = nn.Sequential(
          nn.Conv2d(1,20,5),
          nn.ReLU(),
          nn.Conv2d(20,64,5),
          nn.ReLU()
        )

parameters

module 保存所有需要计算的参数

自己在设计网络时，要通过nn.Parameter()加入module的优化器管理

1
2
3
4
5
6
7
8
9
10
11
12
13
class MyLiner(nn.Module):
    
    def __int__(self, inp, outp):
        super(MyLinear , self).__init__()
        
        #
        self.w = nn.Parameter(torch.randn(outp,inp))#这时生成参数不需要 requires_grad =True
        self.b = nn.Parameter(torch.randn(outp))
        
      def forward(self,x ):
        x = x@self.w.t() +self.b
        return x

modules

1
2
3
4
modules

children

GPU

1
2
3
device = torch.device('cuda')
net = Net()
net.to(device)

save and load

save

1
2
3
cn = MyNet()
......
torch.save(cn.state_dict(), "your_model_path.pth")

load

1
2
3
4
5
6
cn = MyNet()

state_dict = torch.load("your_model_path.pth")

#加载训练好的参数
cn.load_state_dict(state_dict)

train/test

1
2
cn = cn.eval()#转换成测试状态
cn = cn.train()#转换成 训练状态

了解基本知识之后，以下从构建自己网络的思路开始讲述

mynet

https://blog.csdn.net/qq_27825451/article/details/90550890

https://blog.csdn.net/qq_27825451/article/details/90705328

貌似是需要自己实验forward过程

常用的网络模块

Flatten

1
2
3
4
5
6
7
class Flatten(nn.Module ):
    
    def __init__(self):
        	super(Flatten, self ).__init__()
     
    def forward(self, input):
        return input.view(input.size(0),-1) #将数据打平作为卷积层和全连接层之间的过渡