（十三）优化器

1. torch.optim介绍

1.1 如何创建优化器

1.2 step（）方法和zero_grad（）方法

2.代码实战

from torch import nn, optim
from torchvision import datasets, transforms
from torch.utils.data import DataLoader


class My_Module(nn.Module):
    def __init__(self, *args, **kwargs) -> None:
        super().__init__(*args, **kwargs)
        self.seq = nn.Sequential(
            nn.Conv2d(in_channels=3, out_channels=32, kernel_size=5, padding=2),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(in_channels=32, out_channels=32, kernel_size=5, padding=2),
            nn.MaxPool2d(kernel_size=2),
            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=5, padding=2),
            nn.MaxPool2d(kernel_size=2),
            nn.Flatten(),
            nn.Linear(in_features=1024, out_features=64),
            nn.Linear(in_features=64, out_features=10)
        )

    def forward(self, x):
        output = self.seq(x)
        return output


my_module = My_Module()
loss = nn.CrossEntropyLoss()  # 定义损失函数
optimizer = optim.SGD(params=my_module.parameters(), lr=0.01)  # 定义优化器,采用随机梯度下降算法
test_dataset = datasets.CIFAR10(root="datasets", transform=transforms.ToTensor(), download=True)
test_dataloader = DataLoader(dataset=test_dataset, batch_size=1, shuffle=True, drop_last=False)
for data in test_dataloader:
    imgs, targets = data
    output = my_module(imgs)
    result_loss = loss(output, targets)
    optimizer.zero_grad()  # 将模型的上一轮梯度清零，防止对本轮梯度优化产生影响
    result_loss.backward()
    optimizer.step()  # 通过随机梯度下降算法优化模型的梯度

3.代码调试

3.1反向传播前模型梯度

3.2反向传播后模型梯度

3.3随机梯度下降优化后