model.py

import torch
from torch import nn

# 搭建神经网络
class MyNN(nn.Module):
    def __init__(self):
        super(MyNN, self).__init__()
        self.model = nn.Sequential(
            nn.Conv2d(3, 32, 5, 1, 2),
            nn.MaxPool2d(2),
            nn.Conv2d(32, 32, 5, 1, 2),
            nn.MaxPool2d(2),
            nn.Conv2d(32, 64, 5, 1, 2),
            nn.MaxPool2d(2),
            nn.Flatten(),
            nn.Linear(64 * 4 * 4, 64),
            nn.Linear(64, 10)
        )

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

if __name__ == '__main__':
    # 验证网络的正确性
    mynn = MyNN()
    input = torch.ones(64,3,32,32)
    output = mynn(input)
    print(output)

运行结果：torch.Size([64,10])

返回64行数据，每一行数据有10个数据，代表每一张图片在10个类别中的概率

train.py

import torchvision
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
# model.py必须和train.py在同一个文件夹下
from model import *

# 准备数据集（CIFAR10 数据集是PIL Image，要转换为tensor数据类型）
train_data = torchvision.datasets.CIFAR10(root="../datasets",train=True,transform=torchvision.transforms.ToTensor(),download=False)
test_data = torchvision.datasets.CIFAR10(root="../datasets",train=False,transform=torchvision.transforms.ToTensor(),download=False)

# 获得数据集的长度
train_data_size = len(train_data)
test_data_size = len(test_data)
print("训练数据集的长度为：{}".format(train_data_size))
print("测试数据集的长度为：{}".format(test_data_size))

# 利用dataloader来加载数据集
train_dataloader = DataLoader(train_data,batch_size=64)
test_dataloader = DataLoader(test_data,batch_size=64)

# 创建网络模型
mynn = MyNN()
# 损失函数
loss_function = nn.CrossEntropyLoss()
# 优化器
learning_rate = 1e-2
optimizer = torch.optim.SGD(mynn.parameters(), lr=learning_rate) # SGD 随机梯度下降

# 设置训练网络的一些参数
total_train_step = 0 # 记录训练次数
total_test_step = 0 # 记录测试次数
epoch = 10 # 训练的轮数

# 添加tensorboard
writer = SummaryWriter("../logs_train")

for i in range(epoch):
    print("----------第{}轮训练开始----------".format(i+1))
    # 训练步骤开始
    mynn.train()
    for data in train_dataloader:
        imgs,targets = data
        outputs = mynn(imgs)
        loss = loss_function(outputs, targets)
        # 优化器优化模型
        optimizer.zero_grad() # 首先要梯度清零
        loss.backward() # 反向传播得到每一个参数节点的梯度
        optimizer.step() # 对参数进行优化

        total_train_step += 1
        # 训练步骤逢百才打印记录
        if total_train_step % 100 == 0:
            print("训练次数：{}，loss:{}".format(total_train_step, loss.item()))
            writer.add_scalar("train_loss",loss.item(),total_train_step)

    # 测试步骤开始
    mynn.eval()
    total_test_loss = 0
    total_accuracy = 0
    # 无梯度，不进行调优
    with torch.no_grad():
        for data in test_dataloader:
            imgs,targets = data
            outputs = mynn(imgs)
            loss = loss_function(outputs, targets)
            total_test_loss += loss
            # 即便得到整体测试集上的 loss，也不能很好说明在测试集上的表现效果
            # 在分类问题中可以用正确率表示
            accuracy = (outputs.argmax(1) == targets).sum()
            total_accuracy += accuracy
    print("整体测试集上的loss:{}".format(total_test_loss))
    print("整体测试集上的正确率:{}".format(total_accuracy/test_data_size))
    writer.add_scalar("test_loss",total_test_loss,total_test_step)
    writer.add_scalar("test_accuracy",total_accuracy/test_data_size,total_test_step)
    total_test_step += 1

    # 保存每一轮训练的模型
    torch.save(mynn,"mynn_{}.pth".format(i))
    # torch.save(mynn.state_dict(),"mynn_{}.pth".format(i))
    print("模型已保存")

writer.close()

关于正确率的计算：

方式1：

import torch

outputs = torch.tensor([[0.1,0.2],
                       [0.3,0.4]])
target = torch.tensor([0,1])

predict = outputs.argmax(1)
print(predict)

print(predict == target)
print((predict == target).sum())

方式2：

import torch

outputs = torch.tensor([[0.1,0.2],
                       [0.3,0.4]])
target = torch.tensor([0,1])

predict = torch.max(outputs, dim=1)[1]
print(predict)

print(torch.eq(predict,target))
print(torch.eq(predict,target).sum())
print(torch.eq(predict,target).sum().item())