构建模型

import torch
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):
    def __init__(self):
        super(Net,self).__init__()
        #定义第一层卷积层，输入维度=1，输出维度=6，卷积核大小3*3
        self.conv1=nn.Conv2d(1,6,3)
        self.conv2=nn.Conv2d(6,16,3)
        self.fc1=nn.Linear(16*6*6,120)
        self.fc2=nn.Linear(120,84)
        self.fc3=nn.Linear(84,10)
        
    def forward(self,x):
        #注意：任意卷积层后面要加激活层，池化层
        x=F.max_pool2d(F.relu(self.conv1(x),(2,2)))
        x=F.max_pool2d(F.relu(self.conv2(x),2))
        x=x.view(-1,self.num_flat_features(x))
        x=F.relu(self.fc1(x))
        x=F.relu(self.fc2(x))
        x=self.fc3(x)
        return x
    
    def num_flat_features(self,x):
        size=x.size()[1:]
        num_features=1
        for s in size:
            num_features*=s
        return num_features
    
net=Net()
print(net)

模型中的可训练参数

params=list(net.parameters())
print(len(params))
print(params[0].size()) #conv1的参数

假设输入尺寸为32*32

input=torch.randn(1,1,32,32) #个数，通道数，长，宽
out=net(input)
print(out)
print(out.size())

注意

损失函数

target=torch.randn(10)
target=target.view(1,-1)
criterion=nn.MSELoss()
loss=criterion(out,target)
print(loss)

print(loss.grad_fn)
print(loss.grad_fn.next_functions[0][0]) #上一层的grad_fn
print(loss.grad_fn.next_functions[0][0].next_functions[0][0]) #上上一层的grad_fn

反向传播

#首先要执行梯度清零的操作
net.zero_grad()

print('conv1.bisa.grad before backward')
print(net.conv1.bias.grad)

#实现一次反向传播
loss.backward()

print('conv1.bisa.grad after backward')
print(net.conv1.bias.grad)

更新网络参数

#导入优化器包
import torch.optim as optim
#构建优化器
optimizer=optim.SGD(net.parameters(),lr=0.01)
#优化器梯度清零
optimizer.zero_grad()
#进行网络计算并计算损失值
output=net(input)
loss=criterion(output,target)
#执行反向传播
loss.backward()
#更新参数
optimizer.step()