模型训练基本套路

import torch.optim
import torchvision
from torch import nn
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from torchvision.transforms import transforms
from model import *

# 准备数据集
tran_data = torchvision.datasets.CIFAR10('./data', True, transform=transforms.ToTensor(), download=True)
test_data = torchvision.datasets.CIFAR10('./data', False, transform=transforms.ToTensor(), download=True)

# 查看数据集长度
tran_len = len(tran_data)
test_len = len(test_data)
print(f"训练集长度：{tran_len};测试集长度：{test_len}")

# 加载数据集
tran_dataloader = DataLoader(tran_data, batch_size=64)
test_dataloader = DataLoader(test_data, batch_size=64)

# 创建模型
my_model = Model()

# 损失函数
loss_fun = nn.CrossEntropyLoss()

# 优化器
learning_rate = 0.01
optimize = torch.optim.SGD(my_model.parameters(), lr=learning_rate)

# 设置训练网络的一些参数
# 记录训练的次数
total_train_step = 0
total_test_step = 0
# 设置训练次数
epoch = 10

writer = SummaryWriter('./logs_procedure1')

for i in range(epoch):
    print("---------------------")
    print(f"第{i + 1}轮训练开始")
    for data in tran_dataloader:
        imgs, targets = data
        output = my_model(imgs)
        loss = loss_fun(output, targets)

        # 优化模型
        optimize.zero_grad()
        loss.backward()
        optimize.step()

        total_train_step += 1
        if total_train_step % 100 == 0:
            print(f"训练次数：{total_train_step},Loss:{loss.item()}")
            writer.add_scalar('train_loss', loss.item(), total_train_step)

    # 测试数据集开始
    with torch.no_grad():
        total_loss = 0
        for data in test_dataloader:
            imgs, targets = data
            output = my_model(imgs)
            loss = loss_fun(output, targets)
            total_loss = total_loss + loss.item()
        print(f"第{i + 1}轮训练，测试集上的误差为{total_loss}")
        writer.add_scalar('test_loss',total_loss,i+1)
        torch.save(my_model,f'my_model_{i+1}.pth')

writer.close()