PyTorch学习笔记三

目录

一、损失函数

       理论阶段

        实战阶段

二、优化器 

---

一、损失函数

        作用：

        🟢计算输出值和目标的差距

        🟢为更新输出、调整参数提供依据

       理论阶段

        L1Loss ：输入任意维度，输出相同维度

        MSELoss：均方误差损失函数 

         CrossEntropyLoss：交叉熵，训练分类问题时使用，公式如下

        实战阶段

        首先加载数据集，构建神经网络

import torch
import torchvision.datasets
from torch.nn import Conv2d, MaxPool2d, Linear, Flatten
from torch import nn
from torch.utils.data import DataLoader

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


class deep_nn(nn.Module):
    def __init__(self):
        super().__init__()
        # 第一个卷积层输入通道为3，输出通道为32，卷积核5×5，计算得padding=2
        self.conv1 = Conv2d(3, 32, 5, padding=2)
        # 池化层不改变通道数，默认不填充
        self.maxpool1 = MaxPool2d(2)
        self.conv2 = Conv2d(32, 32, 5, padding=2)
        self.maxpool2 = MaxPool2d(2)
        self.conv3 = Conv2d(32, 64, 5, padding=2)
        self.maxpool3 = MaxPool2d(2)
        self.flatten = Flatten()
        self.liner1 = Linear(1024, 64)
        self.liner2 = Linear(64, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = self.maxpool1(x)
        x = self.conv2(x)
        x = self.maxpool2(x)
        x = self.conv3(x)
        x = self.maxpool3(x)
        x = self.flatten(x)
        x = self.liner1(x)
        x = self.liner2(x)
        return x

Mynn = deep_nn()

for data in loader:
    imgs, targets = data
    output = Mynn(imgs)
    print(output)
    print(targets)

         接着定义损失函数

Myloss = CrossEntropyLoss()

for data in loader:
    imgs, targets = data
    output = Mynn(imgs)
    res_loss = Myloss(output, targets)
    print(res_loss)

 

        运用损失函数获得的梯度进行反向传播

res_loss.backward()

二、优化器 

作用：利用损失函数所获得的梯度对各个神经节点进行调参优化 

        官方文档在torch.opim中，首先先构建一个优化器，传入模型和学习速率，接着调用step（）方法进行调参，在每一个batch-size训练过程中，需将上一个batch-size训练的梯度参数进行清0.

        利用第一节中的损失函数加入优化器，训练10轮

import torch
import torchvision.datasets
from torch.nn import Conv2d, MaxPool2d, Linear, Flatten, CrossEntropyLoss
from torch import nn
from torch.utils.data import DataLoader

dataset = torchvision.datasets.CIFAR10(root='./dataset', train=False, transform=torchvision.transforms.ToTensor(), download=True)
loader = DataLoader(dataset, batch_size=64)


class deep_nn(nn.Module):
    def __init__(self):
        super().__init__()
        # 第一个卷积层输入通道为3，输出通道为32，卷积核5×5，计算得padding=2
        self.conv1 = Conv2d(3, 32, 5, padding=2)
        # 池化层不改变通道数，默认不填充
        self.maxpool1 = MaxPool2d(2)
        self.conv2 = Conv2d(32, 32, 5, padding=2)
        self.maxpool2 = MaxPool2d(2)
        self.conv3 = Conv2d(32, 64, 5, padding=2)
        self.maxpool3 = MaxPool2d(2)
        self.flatten = Flatten()
        self.liner1 = Linear(1024, 64)
        self.liner2 = Linear(64, 10)

    def forward(self, x):
        x = self.conv1(x)
        x = self.maxpool1(x)
        x = self.conv2(x)
        x = self.maxpool2(x)
        x = self.conv3(x)
        x = self.maxpool3(x)
        x = self.flatten(x)
        x = self.liner1(x)
        x = self.liner2(x)
        return x

Mynn = deep_nn()
Myloss = CrossEntropyLoss()

# 构建优化器,学习速率设置为0.01，太大模型不稳定，太小模型训练过慢
optim = torch.optim.SGD(Mynn.parameters(), lr=0.01)

for epoch in range(10):
    total_loss = 0
    for data in loader:
        imgs, targets = data
        output = Mynn(imgs)
        res_loss = Myloss(output, targets)
        # 梯度调为0
        optim.zero_grad()
        # 获取梯度
        res_loss.backward()
        # 进行调优
        optim.step()
        total_loss += res_loss

    print(total_loss)

        可以看出损失在减小