DAY 52 神经网络调参指南
知识点回顾：
- 随机种子
- 内参的初始化
- 神经网络调参指南
- 参数的分类
- 调参的顺序
- 各部分参数的调整心得
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
import torchvision
import torchvision.transforms as transforms
import numpy as np
import matplotlib.pyplot as plt
import random
from tqdm import tqdm# 设置随机种子确保结果可复现
def set_seed(seed=42):random.seed(seed)np.random.seed(seed)torch.manual_seed(seed)if torch.cuda.is_available():torch.cuda.manual_seed(seed)torch.cuda.manual_seed_all(seed)torch.backends.cudnn.deterministic = Truetorch.backends.cudnn.benchmark = False# 定义CNN模型
class SimpleCNN(nn.Module):def __init__(self):super(SimpleCNN, self).__init__()# 第一个卷积层：输入通道1，输出通道32，卷积核3x3self.conv1 = nn.Conv2d(1, 32, kernel_size=3, padding=1)self.bn1 = nn.BatchNorm2d(32)self.relu1 = nn.ReLU()self.pool1 = nn.MaxPool2d(2)# 第二个卷积层：输入通道32，输出通道64，卷积核3x3self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1)self.bn2 = nn.BatchNorm2d(64)self.relu2 = nn.ReLU()self.pool2 = nn.MaxPool2d(2)# 全连接层self.fc1 = nn.Linear(64 * 7 * 7, 128)self.dropout = nn.Dropout(0.5)self.fc2 = nn.Linear(128, 10)def forward(self, x):# 卷积层1x = self.conv1(x)x = self.bn1(x)x = self.relu1(x)x = self.pool1(x)# 卷积层2x = self.conv2(x)x = self.bn2(x)x = self.relu2(x)x = self.pool2(x)# 展平x = x.view(-1, 64 * 7 * 7)# 全连接层x = self.fc1(x)x = self.relu1(x)x = self.dropout(x)x = self.fc2(x)return x# 参数初始化
def init_weights(m):if isinstance(m, nn.Conv2d):nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')if m.bias is not None:nn.init.constant_(m.bias, 0)elif isinstance(m, nn.BatchNorm2d):nn.init.constant_(m.weight, 1)nn.init.constant_(m.bias, 0)elif isinstance(m, nn.Linear):nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')nn.init.constant_(m.bias, 0)# 训练函数
def train(model, train_loader, criterion, optimizer, device):model.train()train_loss = 0correct = 0total = 0for batch_idx, (inputs, targets) in enumerate(tqdm(train_loader)):inputs, targets = inputs.to(device), targets.to(device)optimizer.zero_grad()outputs = model(inputs)loss = criterion(outputs, targets)loss.backward()optimizer.step()train_loss += loss.item()_, predicted = outputs.max(1)total += targets.size(0)correct += predicted.eq(targets).sum().item()acc = 100. * correct / totalreturn train_loss / len(train_loader), acc# 测试函数
def test(model, test_loader, criterion, device):model.eval()test_loss = 0correct = 0total = 0with torch.no_grad():for batch_idx, (inputs, targets) in enumerate(tqdm(test_loader)):inputs, targets = inputs.to(device), targets.to(device)outputs = model(inputs)loss = criterion(outputs, targets)test_loss += loss.item()_, predicted = outputs.max(1)total += targets.size(0)correct += predicted.eq(targets).sum().item()acc = 100. * correct / totalreturn test_loss / len(test_loader), acc# 主函数
def main():# 设置随机种子set_seed(42)# 设置设备device = 'cuda' if torch.cuda.is_available() else 'cpu'print(f"使用设备: {device}")# 数据预处理transform_train = transforms.Compose([transforms.RandomRotation(10),transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,))])transform_test = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,))])# 加载数据集train_dataset = torchvision.datasets.MNIST(root='./data', train=True, download=True, transform=transform_train)test_dataset = torchvision.datasets.MNIST(root='./data', train=False, download=True, transform=transform_test)# 创建数据加载器batch_size = 128 # 调参参数1train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=2)test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=100, shuffle=False, num_workers=2)# 初始化模型model = SimpleCNN().to(device)model.apply(init_weights)# 定义损失函数和优化器criterion = nn.CrossEntropyLoss()learning_rate = 0.01 # 调参参数2optimizer = optim.Adam(model.parameters(), lr=learning_rate, weight_decay=1e-5) # 调参参数3scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min', patience=3, factor=0.5)# 训练模型epochs = 15 # 调参参数4train_losses = []train_accs = []test_losses = []test_accs = []best_acc = 0for epoch in range(epochs):print(f"\nEpoch: {epoch+1}/{epochs}")train_loss, train_acc = train(model, train_loader, criterion, optimizer, device)test_loss, test_acc = test(model, test_loader, criterion, device)train_losses.append(train_loss)train_accs.append(train_acc)test_losses.append(test_loss)test_accs.append(test_acc)print(f"Train Loss: {train_loss:.4f} | Train Acc: {train_acc:.2f}%")print(f"Test Loss: {test_loss:.4f} | Test Acc: {test_acc:.2f}%")# 学习率调整scheduler.step(test_loss)# 保存最佳模型if test_acc > best_acc:best_acc = test_acctorch.save(model.state_dict(), 'best_model.pth')print(f"Model saved with accuracy: {best_acc:.2f}%")# 加载最佳模型model.load_state_dict(torch.load('best_model.pth'))_, final_acc = test(model, test_loader, criterion, device)print(f"\nFinal Test Accuracy: {final_acc:.2f}%")# 绘制训练过程plt.figure(figsize=(12, 5))plt.subplot(1, 2, 1)plt.plot(train_losses, label='Train Loss')plt.plot(test_losses, label='Test Loss')plt.xlabel('Epoch')plt.ylabel('Loss')plt.legend()plt.title('Loss Curves')plt.subplot(1, 2, 2)plt.plot(train_accs, label='Train Accuracy')plt.plot(test_accs, label='Test Accuracy')plt.xlabel('Epoch')plt.ylabel('Accuracy (%)')plt.legend()plt.title('Accuracy Curves')plt.tight_layout()plt.savefig('training_curves.png')plt.show()if __name__ == '__main__':main()