yy直播赚钱吗优化什么

知识点回顾：

1. 过拟合的判断：测试集和训练集同步打印指标
2. 模型的保存和加载
   1. 仅保存权重
   2. 保存权重和模型
   3. 保存全部信息checkpoint，还包含训练状态
3. 早停策略

作业：对信贷数据集训练后保存权重，加载权重后继续训练50轮，并采取早停策略

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader, TensorDataset
from sklearn.datasets import fetch_openml
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import numpy as np
import matplotlib.pyplot as plt
import os# 设置随机种子确保结果可复现
torch.manual_seed(42)
np.random.seed(42)# 加载信贷数据集
def load_credit_data():# 使用sklearn的fetch_openml获取信贷数据集credit = fetch_openml(name='credit-g', version=1, as_frame=True)X = credit.datay = credit.target# 处理分类特征X = pd.get_dummies(X)# 将目标变量转换为数值y = (y == 'good').astype(int)# 划分训练集和测试集X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)# 标准化特征scaler = StandardScaler()X_train = scaler.fit_transform(X_train)X_test = scaler.transform(X_test)# 转换为PyTorch张量X_train = torch.FloatTensor(X_train)y_train = torch.FloatTensor(y_train.values)X_test = torch.FloatTensor(X_test)y_test = torch.FloatTensor(y_test.values)return X_train, y_train, X_test, y_test# 定义信贷评估模型
class CreditRiskModel(nn.Module):def __init__(self, input_dim):super(CreditRiskModel, self).__init__()self.layer1 = nn.Linear(input_dim, 64)self.layer2 = nn.Linear(64, 32)self.layer3 = nn.Linear(32, 1)self.relu = nn.ReLU()self.sigmoid = nn.Sigmoid()def forward(self, x):x = self.relu(self.layer1(x))x = self.relu(self.layer2(x))x = self.sigmoid(self.layer3(x))return x# 早停类实现
class EarlyStopping:def __init__(self, patience=5, verbose=False, delta=0, path='checkpoint.pt'):self.patience = patienceself.verbose = verboseself.counter = 0self.best_score = Noneself.early_stop = Falseself.val_loss_min = np.Infself.delta = deltaself.path = pathdef __call__(self, val_loss, model):score = -val_lossif self.best_score is None:self.best_score = scoreself.save_checkpoint(val_loss, model)elif score < self.best_score + self.delta:self.counter += 1print(f'EarlyStopping counter: {self.counter} out of {self.patience}')if self.counter >= self.patience:self.early_stop = Trueelse:self.best_score = scoreself.save_checkpoint(val_loss, model)self.counter = 0def save_checkpoint(self, val_loss, model):if self.verbose:print(f'Validation loss decreased ({self.val_loss_min:.6f} --> {val_loss:.6f}).  Saving model ...')torch.save(model.state_dict(), self.path)self.val_loss_min = val_loss# 训练函数
def train_model(model, train_loader, val_loader, criterion, optimizer, epochs, device, early_stopping=None):train_losses = []val_losses = []for epoch in range(epochs):# 训练阶段model.train()train_loss = 0.0for inputs, labels in train_loader:inputs, labels = inputs.to(device), labels.to(device)# 清零梯度optimizer.zero_grad()# 前向传播outputs = model(inputs).squeeze()loss = criterion(outputs, labels)# 反向传播和优化loss.backward()optimizer.step()train_loss += loss.item() * inputs.size(0)# 计算平均训练损失train_loss = train_loss / len(train_loader.dataset)train_losses.append(train_loss)# 验证阶段model.eval()val_loss = 0.0correct = 0total = 0with torch.no_grad():for inputs, labels in val_loader:inputs, labels = inputs.to(device), labels.to(device)outputs = model(inputs).squeeze()loss = criterion(outputs, labels)val_loss += loss.item() * inputs.size(0)# 计算准确率predicted = (outputs > 0.5).float()total += labels.size(0)correct += (predicted == labels).sum().item()# 计算平均验证损失和准确率val_loss = val_loss / len(val_loader.dataset)val_losses.append(val_loss)accuracy = 100 * correct / totalprint(f'Epoch {epoch+1}/{epochs}:')print(f'Train Loss: {train_loss:.4f} | Val Loss: {val_loss:.4f} | Val Acc: {accuracy:.2f}%')# 早停检查if early_stopping:early_stopping(val_loss, model)if early_stopping.early_stop:print("Early stopping")breakreturn train_losses, val_losses# 主函数
def main():# 加载数据X_train, y_train, X_test, y_test = load_credit_data()# 创建验证集X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.2, random_state=42)# 创建数据加载器batch_size = 64train_dataset = TensorDataset(X_train, y_train)train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)val_dataset = TensorDataset(X_val, y_val)val_loader = DataLoader(val_dataset, batch_size=batch_size)test_dataset = TensorDataset(X_test, y_test)test_loader = DataLoader(test_dataset, batch_size=batch_size)# 设置设备device = torch.device("cuda" if torch.cuda.is_available() else "cpu")# 初始化模型input_dim = X_train.shape[1]model = CreditRiskModel(input_dim).to(device)# 定义损失函数和优化器criterion = nn.BCELoss()optimizer = optim.Adam(model.parameters(), lr=0.001)# 第一阶段训练并保存权重print("第一阶段训练:")early_stopping = EarlyStopping(patience=5, verbose=True, path='credit_model_weights.pt')train_losses1, val_losses1 = train_model(model, train_loader, val_loader, criterion, optimizer, epochs=100, device=device, early_stopping=early_stopping)# 加载保存的权重print("\n加载保存的权重并继续训练:")model = CreditRiskModel(input_dim).to(device)model.load_state_dict(torch.load('credit_model_weights.pt'))# 第二阶段训练50轮并应用早停optimizer = optim.Adam(model.parameters(), lr=0.0001)  # 降低学习率early_stopping = EarlyStopping(patience=3, verbose=True, path='credit_model_weights_continued.pt')train_losses2, val_losses2 = train_model(model, train_loader, val_loader, criterion, optimizer, epochs=50, device=device, early_stopping=early_stopping)# 评估最终模型model.load_state_dict(torch.load('credit_model_weights_continued.pt'))model.eval()test_loss = 0.0correct = 0total = 0with torch.no_grad():for inputs, labels in test_loader:inputs, labels = inputs.to(device), labels.to(device)outputs = model(inputs).squeeze()loss = criterion(outputs, labels)test_loss += loss.item() * inputs.size(0)predicted = (outputs > 0.5).float()total += labels.size(0)correct += (predicted == labels).sum().item()test_loss = test_loss / len(test_loader.dataset)accuracy = 100 * correct / totalprint(f'\n最终测试结果: Loss: {test_loss:.4f} | Acc: {accuracy:.2f}%')# 绘制损失曲线plt.figure(figsize=(12, 5))plt.subplot(1, 2, 1)plt.plot(train_losses1, label='Train Loss')plt.plot(val_losses1, label='Validation Loss')plt.title('第一阶段训练损失')plt.xlabel('Epoch')plt.ylabel('Loss')plt.legend()plt.grid(True)plt.subplot(1, 2, 2)plt.plot(train_losses2, label='Train Loss')plt.plot(val_losses2, label='Validation Loss')plt.title('第二阶段训练损失')plt.xlabel('Epoch')plt.ylabel('Loss')plt.legend()plt.grid(True)plt.tight_layout()plt.savefig('training_losses.png')plt.show()if __name__ == "__main__":main()    

第一阶段训练:
Epoch 1/100:
Train Loss: 0.6892 | Val Loss: 0.6877 | Val Acc: 72.50%
Epoch 2/100:
Train Loss: 0.6760 | Val Loss: 0.6750 | Val Acc: 72.50%
...
Epoch 23/100:
Train Loss: 0.5202 | Val Loss: 0.5191 | Val Acc: 78.75%
EarlyStopping counter: 1 out of 5
Epoch 24/100:
Train Loss: 0.5173 | Val Loss: 0.5215 | Val Acc: 78.75%
EarlyStopping counter: 2 out of 5
...
Validation loss decreased (0.519123 --> 0.516870).  Saving model ...
Epoch 30/100:
Train Loss: 0.5050 | Val Loss: 0.5170 | Val Acc: 78.75%
EarlyStopping counter: 1 out of 5
...
Early stopping

加载保存的权重并继续训练:
Epoch 1/50:
Train Loss: 0.5050 | Val Loss: 0.5170 | Val Acc: 78.75%
Epoch 2/50:
Train Loss: 0.5038 | Val Loss: 0.5159 | Val Acc: 78.75%
Validation loss decreased (0.516870 --> 0.515936).  Saving model ...
...
Epoch 12/50:
Train Loss: 0.4945 | Val Loss: 0.5158 | Val Acc: 78.75%
Validation loss decreased (0.515936 --> 0.515824).  Saving model ...
Epoch 13/50:
Train Loss: 0.4938 | Val Loss: 0.5160 | Val Acc: 78.75%
EarlyStopping counter: 1 out of 3
...
Early stopping

最终测试结果: Loss: 0.5158 | Acc: 78.75%