完整的模型训练

• train.py

import torchvision
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from model import *
train_data = torchvision.datasets.CIFAR10(root="download_data", train=True,
                                          transform=torchvision.transforms.ToTensor(), download=True)
test_data = torchvision.datasets.CIFAR10(root="download_data", train=False,
                                         transform=torchvision.transforms.ToTensor(), download=True)
data_size = len(train_data)
test_data_size = len(test_data)
print(data_size)
print(test_data_size)
# 格式化字符串
print("数据集长度：{}".format(data_size))
data_loader = DataLoader(train_data,batch_size=64)
test_loader = DataLoader(test_data,batch_size=64)
# 创建网络模型
l = Lcy()
# 定义损失函数
loss_function = nn.CrossEntropyLoss()
# 优化器
# learning_rate = 0.01
# 1e-2 : 1 x (10)^(-2) = 1 x 0.01 = 0.01
# 两种写法表达的数字相同
learning_rate = 1e-2
optimizer = torch.optim.SGD(l.parameters(),lr=learning_rate)
# 设置初始化参数
# 记录训练的次数
total_train_step = 0
# 记录测试的次数
total_test_step = 0
# 训练的轮数
epoch = 10
writer = SummaryWriter("logs")
for i in range(epoch):
    print("-------第{}轮训练-------".format(i+1))
    # 训练的步骤
    for data in data_loader:
        img, targets = data
        output = l(img)
        loss = loss_function(output,targets)
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        total_test_step += 1
        if total_test_step % 100 == 0:
            print("训练次数{},loss:{}".format(total_test_step,loss))
            writer.add_scalar("train_loss",loss, total_test_step)
    #测试步骤：
    total_test_loss = 0
    with torch.no_grad():
        for data in test_loader:
            img, targets = data
            output = l(img)
            loss = loss_function(output,targets)
            total_test_loss += loss.item()
    print("整体测试上的损失：{}".format(total_test_loss))
    writer.add_scalar("test_loss",total_test_loss,total_test_step)
writer.close()

• model.py

import torch
from torch import nn
class Lcy(nn.Module):
    def __init__(self):
        super(Lcy,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,input):
        output = self.model(input)
        return output
if __name__ == '__main__':
    l = Lcy()
    input = torch.ones((64,3,32,32))
    output = l(input)
    print(output.shape)