pytorch --反向传播和优化器

news2024/9/25 23:20:26

1. 反向传播

计算当前张量的梯度

Tensor.backward(gradient=None, retain_graph=None, create_graph=False, inputs=None)

计算当前张量相对于图中叶子节点的梯度。

使用反向传播，每个节点的梯度，根据梯度进行参数优化，最后使得损失最小化

代码：

import torch
import torchvision
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

dataset = torchvision.datasets.CIFAR10('data',train=False,transform=torchvision.transforms.ToTensor(),download=True)
dataloader = DataLoader(dataset,batch_size=1)



class Tudui(nn.Module):
    def __init__(self):
        super().__init__()
        # 另一种写法
        self.model1 = nn.Sequential(
            nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Flatten(),
            nn.Linear(in_features=1024, out_features=64),
            nn.Linear(in_features=64, out_features=10)
        )

    def forward(self,x):
        # sequential方式
        x = self.model1(x)
        return x
loss = nn.CrossEntropyLoss()
tudui = Tudui()
for data in dataloader:
    imgs,target = data
    outputs= tudui(imgs)
    result_loss = loss(outputs,target)
    result_loss.backward() # 梯度
    print(result_loss)

2.优化器（以随机梯度下降算法为例）
将上一步的梯度清零
params ,lr（学习率）
随机梯度下降SGD
torch.optim.SGD(params,
lr=,
momentum=0, dampening=0, weight_decay=0, nesterov=False, *, maximize=False, foreach=None, differentiable=False)

代码：

import torch
import torchvision
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter

dataset = torchvision.datasets.CIFAR10('data',train=False,transform=torchvision.transforms.ToTensor(),download=True)
dataloader = DataLoader(dataset,batch_size=1)



class Tudui(nn.Module):
    def __init__(self):
        super().__init__()
        # 另一种写法
        self.model1 = nn.Sequential(
            nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, stride=1, padding=2),
            nn.MaxPool2d(2),
            nn.Flatten(),
            nn.Linear(in_features=1024, out_features=64),
            nn.Linear(in_features=64, out_features=10)
        )

    def forward(self,x):
        # sequential方式
        x = self.model1(x)
        return x
loss = nn.CrossEntropyLoss()
tudui = Tudui()
optim = torch.optim.SGD(tudui.parameters(),lr=0.01) # params,lr
for epoch in range(5):# 在整个数据集上训练5次
    running_loss = 0
    #对数据进行一轮学习
    for data in dataloader:
        imgs,target = data
        outputs= tudui(imgs)
        result_loss = loss(outputs,target)
        optim.zero_grad() # 将上一步的梯度清零
        result_loss.backward() # 梯度
        optim.step() # 根据梯度修改参数
        # print(result_loss)
        running_loss = running_loss + result_loss
    print(running_loss)