作业：

对于day'41的简单cnn，看看是否可以借助调参指南进一步提高精度。

import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
import numpy as np# 设置中文字体支持
plt.rcParams["font.family"] = ["SimHei"]
plt.rcParams['axes.unicode_minus'] = False  # 解决负号显示问题# 1. 改进数据预处理 - 增加数据增强
transform_train = transforms.Compose([transforms.RandomCrop(32, padding=4),  # 随机裁剪transforms.RandomHorizontalFlip(),     # 随机水平翻转transforms.ToTensor(),transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))  # 使用CIFAR-10的真实统计数据
])transform_test = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])# 2. 加载CIFAR-10数据集
train_dataset = datasets.CIFAR10(root='./data',train=True,download=True,transform=transform_train
)test_dataset = datasets.CIFAR10(root='./data',train=False,transform=transform_test
)# 3. 增加Batch Size - 从64增加到128
batch_size = 128
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=2)
test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False, num_workers=2)# 4. 改进模型结构 - 更深、更宽的网络
class ImprovedMLP(nn.Module):def __init__(self):super(ImprovedMLP, self).__init__()self.flatten = nn.Flatten()# 增加网络宽度和深度self.layer1 = nn.Linear(3072, 1024)self.bn1 = nn.BatchNorm1d(1024)  # 添加批归一化self.relu1 = nn.ReLU()self.dropout1 = nn.Dropout(0.3)self.layer2 = nn.Linear(1024, 1024)self.bn2 = nn.BatchNorm1d(1024)self.relu2 = nn.ReLU()self.dropout2 = nn.Dropout(0.3)self.layer3 = nn.Linear(1024, 512)self.bn3 = nn.BatchNorm1d(512)self.relu3 = nn.ReLU()self.dropout3 = nn.Dropout(0.3)self.layer4 = nn.Linear(512, 256)self.bn4 = nn.BatchNorm1d(256)self.relu4 = nn.ReLU()self.dropout4 = nn.Dropout(0.3)self.layer5 = nn.Linear(256, 10)def forward(self, x):x = self.flatten(x)x = self.layer1(x)x = self.bn1(x)x = self.relu1(x)x = self.dropout1(x)x = self.layer2(x)x = self.bn2(x)x = self.relu2(x)x = self.dropout2(x)x = self.layer3(x)x = self.bn3(x)x = self.relu3(x)x = self.dropout3(x)x = self.layer4(x)x = self.bn4(x)x = self.relu4(x)x = self.dropout4(x)x = self.layer5(x)return x# 检查GPU是否可用
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")# 初始化模型
model = ImprovedMLP()
model = model.to(device)# 5. 优化器与学习率调度 - 使用学习率预热和余弦退火
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001, weight_decay=5e-4)  # 添加L2正则化
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=200)  # 余弦退火调度器# 6. 早停机制
class EarlyStopping:def __init__(self, patience=10, delta=0):self.patience = patienceself.delta = deltaself.counter = 0self.best_score = Noneself.early_stop = Falsedef __call__(self, val_acc):score = val_accif self.best_score is None:self.best_score = scoreelif score < self.best_score + self.delta:self.counter += 1if self.counter >= self.patience:self.early_stop = Trueelse:self.best_score = scoreself.counter = 0return self.early_stop# 7. 训练模型（改进版，记录训练和验证准确率）
def train(model, train_loader, test_loader, criterion, optimizer, scheduler, device, epochs):model.train()# 记录每个 epoch 的准确率和损失train_acc_history = []train_loss_history = []test_acc_history = []test_loss_history = []# 早停实例early_stopping = EarlyStopping(patience=15)for epoch in range(epochs):running_loss = 0.0correct = 0total = 0for batch_idx, (data, target) in enumerate(train_loader):data, target = data.to(device), target.to(device)optimizer.zero_grad()output = model(data)loss = criterion(output, target)loss.backward()optimizer.step()running_loss += loss.item()_, predicted = output.max(1)total += target.size(0)correct += predicted.eq(target).sum().item()# 每100个批次打印一次训练信息if (batch_idx + 1) % 100 == 0:print(f'Epoch: {epoch+1}/{epochs} | Batch: {batch_idx+1}/{len(train_loader)} 'f'| 损失: {loss.item():.4f} | 准确率: {100.*correct/total:.2f}%')# 计算当前epoch的平均训练损失和准确率epoch_train_loss = running_loss / len(train_loader)epoch_train_acc = 100. * correct / totaltrain_loss_history.append(epoch_train_loss)train_acc_history.append(epoch_train_acc)# 测试阶段model.eval()test_loss = 0correct_test = 0total_test = 0with torch.no_grad():for data, target in test_loader:data, target = data.to(device), target.to(device)output = model(data)test_loss += criterion(output, target).item()_, predicted = output.max(1)total_test += target.size(0)correct_test += predicted.eq(target).sum().item()epoch_test_loss = test_loss / len(test_loader)epoch_test_acc = 100. * correct_test / total_testtest_loss_history.append(epoch_test_loss)test_acc_history.append(epoch_test_acc)print(f'Epoch {epoch+1}/{epochs} 完成 | 训练准确率: {epoch_train_acc:.2f}% | 测试准确率: {epoch_test_acc:.2f}%')# 更新学习率scheduler.step()# 检查早停if early_stopping(epoch_test_acc):print(f"早停触发！在 epoch {epoch+1} 停止训练")break# 绘制训练和测试准确率曲线plot_accuracy(train_acc_history, test_acc_history, epochs)# 绘制训练和测试损失曲线plot_loss(train_loss_history, test_loss_history, epochs)return epoch_test_acc, epoch_test_loss# 8. 绘制准确率曲线
def plot_accuracy(train_acc, test_acc, epochs):plt.figure(figsize=(10, 5))plt.plot(range(1, len(train_acc)+1), train_acc, 'b-', label='训练准确率')plt.plot(range(1, len(test_acc)+1), test_acc, 'r-', label='测试准确率')plt.xlabel('Epoch')plt.ylabel('准确率 (%)')plt.title('训练和测试准确率')plt.legend()plt.grid(True)plt.tight_layout()plt.show()# 9. 绘制损失曲线
def plot_loss(train_loss, test_loss, epochs):plt.figure(figsize=(10, 5))plt.plot(range(1, len(train_loss)+1), train_loss, 'b-', label='训练损失')plt.plot(range(1, len(test_loss)+1), test_loss, 'r-', label='测试损失')plt.xlabel('Epoch')plt.ylabel('损失')plt.title('训练和测试损失')plt.legend()plt.grid(True)plt.tight_layout()plt.show()# 10. 执行训练和测试
epochs = 200  # 增加训练轮次
print("开始训练模型...")
final_accuracy, final_loss = train(model, train_loader, test_loader, criterion, optimizer, scheduler, device, epochs)
print(f"训练完成！最终测试准确率: {final_accuracy:.2f}% | 最终测试损失: {final_loss:.4f}")# 保存模型
torch.save(model.state_dict(), 'cifar10_improved_mlp_model.pth')
print("模型已保存为: cifar10_improved_mlp_model.pth")

训练完成！最终测试准确率: 93.98%