一、损失函数的介绍

参考：
损失函数（loss function）
pytorch loss-functions 文档

作用：
1.计算实际输出和目标之间的差距
2.为我们更新输出提供一定的依据（反向传播)

1.L1Loss类的使用

from torch.nn import L1Loss

class torch.nn.L1Loss(size_average=None, reduce=None, reduction='mean')

参考：
torch.nn.L1Loss

作用：创建一个衡量输入x(模型预测输出)和目标y之间差的绝对值的平均值的标准。

reduction = ‘mean’：

代码实现

import torch
from torch.nn import L1Loss


input = torch.tensor([1, 2, 3], dtype=torch.float32)  # 创建一个一阶张量(3个元素)
targets = torch.tensor([1, 2, 5], dtype=torch.float32)  # 创建一个一阶张量(3个元素)

# 转化为四阶张量（不转化也可以）
inputs = torch.reshape(input, (1, 1, 1, 3))
targets = torch.reshape(targets, (1, 1, 1, 3))

loss1 = L1Loss()  # 创建一个实例
loss2 = L1Loss(reduction="sum")  # 创建一个实例
loss3 = L1Loss(reduction="none")  # 创建一个实例
result1 = loss1(inputs, targets)
result2 = loss2(inputs, targets)
result3 = loss3(inputs, targets)

print(result1)
print(result2)
print(result3)

输出：

tensor(0.6667)
tensor(2.)
tensor([[[[0., 0., 2.]]]])

2.MSELoss类的使用

from torch.nn import MSELoss

class torch.nn.MSELoss(size_average=None, reduce=None, reduction='mean')

参考：
torch.nn.MSELoss

作用：创建一个衡量输入x(模型预测输出)和目标y之间均方误差标准。

reduction = ‘mean’：

# -*- coding: UTF-8 -*-
# 开发团队  ：桂林电子科技大学 - 人工智能学院 - chen
# 开发人员  ：Chen
# 开发时间  ：2023/1/15 21:46
# 开发名称  ：18.nn.loss.py
# 开发工具  ：PyCharm
import torch
from torch.nn import MSELoss


input = torch.tensor([1, 2, 3], dtype=torch.float32)  # 创建一个一阶张量(3个元素)
targets = torch.tensor([1, 2, 5], dtype=torch.float32)  # 创建一个一阶张量(3个元素)

# 转化为四阶张量（不转化也可以）
inputs = torch.reshape(input, (1, 1, 1, 3))
targets = torch.reshape(targets, (1, 1, 1, 3))

loss1 = MSELoss()  # 创建一个实例
loss2 = MSELoss(reduction="sum")  # 创建一个实例
loss3 = MSELoss(reduction="none")  # 创建一个实例
result1 = loss1(inputs, targets)
result2 = loss2(inputs, targets)
result3 = loss3(inputs, targets)

print(result1)
print(result2)
print(result3)

输出：

tensor(1.3333)
tensor(4.)
tensor([[[[0., 0., 4.]]]])

3.损失函数在模型中的实现

import torch
import torchvision
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear, Sequential
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter


class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()  # 初始化父类属性
        self.model1 = Sequential(
            Conv2d(3, 32, 5, stride=1, padding=2),
            MaxPool2d(2),
            Conv2d(32, 32, 5, padding=2),
            MaxPool2d(2),
            Conv2d(32, 64, 5, padding=2),
            MaxPool2d(2),
            Flatten(),
            Linear(1024, 64),
            Linear(64, 10)
        )

    def forward(self, x):
        x = self.model1(x)
        return x


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

loss = nn.CrossEntropyLoss()  # 交叉熵
model = Model()  # 创建一个实例
for data in dataloader:
    imgs, targets = data
    output = model(imgs)
    result_loss = loss(output, targets)
    print(result_loss)  # 输出误差

输出：

Files already downloaded and verified
tensor(2.4058, grad_fn=<NllLossBackward0>)
tensor(2.2368, grad_fn=<NllLossBackward0>)
tensor(2.2519, grad_fn=<NllLossBackward0>)
...
...

二、反向传播

参考：
深度学习 | 反向传播详解

import torch
import torchvision
from torch import nn
from torch.nn import Conv2d, MaxPool2d, Flatten, Linear, Sequential
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter


class Model(nn.Module):
    def __init__(self) -> None:
        super().__init__()  # 初始化父类属性
        self.model1 = Sequential(
            Conv2d(3, 32, 5, stride=1, padding=2),
            MaxPool2d(2),
            Conv2d(32, 32, 5, padding=2),
            MaxPool2d(2),
            Conv2d(32, 64, 5, padding=2),
            MaxPool2d(2),
            Flatten(),
            Linear(1024, 64),
            Linear(64, 10)
        )

    def forward(self, x):
        x = self.model1(x)
        return x


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

loss = nn.CrossEntropyLoss()  # 交叉熵
model = Model()  # 创建一个实例
for data in dataloader:
    imgs, targets = data
    output = model(imgs)
    result_loss = loss(output, targets)
    result_loss.backward()  # 反向传播
    print(result_loss)  # 输出误差

Bebug模型：查看卷积核的梯度
model -> Protected Attributes -> _models -> 0 -> weight -> grad