df = pd.read_csv('data.csv', parse_dates=["Date"], index_col=[0])    df = pd.DataFrame(df)

test_split=round(len(df)*0.20)df_for_training=df[:-test_split]df_for_testing=df[-test_split:]

scaler = MinMaxScaler(feature_range=(0,1))df_for_training_scaled = scaler.fit_transform(df_for_training)df_for_testing_scaled=scaler.transform(df_for_testing)

train_dataset = TimeSeriesDataset(df_for_training_scaled, seq_len=30, pred_len=1)test_dataset = TimeSeriesDataset(df_for_testing_scaled, seq_len=30, pred_len=1)train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)

class CNN_LSTM_Transformer(nn.Module):    def __init__(self, input_dim=5, cnn_channels=16, lstm_hidden=32, transformer_dim=32,                 transformer_heads=4, transformer_layers=1, pred_len=1):        super().__init__()        # CNN        self.cnn = nn.Conv1d(in_channels=input_dim, out_channels=cnn_channels, kernel_size=3, padding=1)        self.cnn_relu = nn.ReLU()        # LSTM        self.lstm = nn.LSTM(input_size=cnn_channels, hidden_size=lstm_hidden, batch_first=True)        # Transformer Encoder         encoder_layer = nn.TransformerEncoderLayer(d_model=transformer_dim, nhead=transformer_heads, batch_first=True)        self.transformer = nn.TransformerEncoder(encoder_layer, num_layers=transformer_layers)        # Projection layers        self.proj_lstm = nn.Linear(lstm_hidden, transformer_dim)        self.pred_len = pred_len        self.fc_out = nn.Linear(transformer_dim, pred_len)    def forward(self, x):        # x: [batch, seq_len, 1]        batch_size, seq_len, _ = x.shape        # CNN expects [batch, channels, seq_len]        cnn_out = self.cnn_relu(self.cnn(x.transpose(1,2)))  # [B, C, T]        cnn_out = cnn_out.transpose(1,2)  # [B, T, C]        # LSTM        lstm_out, _ = self.lstm(cnn_out)  # [B, T, hidden]        lstm_proj = self.proj_lstm(lstm_out)  # [B, T, transformer_dim]        # Transformer        trans_out = self.transformer(lstm_proj)  # [B, T, transformer_dim]        # 取最后时间步输出预测        out = self.fc_out(trans_out[:, -1, :])  # [B, pred_len]        return out.unsqueeze(-1)  # [B, pred_len, 1]

def train_model(model, dataloader, num_epochs=50, learning_rate=1e-3, device='cpu'):    optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)    criterion = nn.MSELoss()    model.train()    loss_history = []    for epoch in range(num_epochs):        epoch_losses = []        for batch_data, batch_targets in dataloader:            batch_data = batch_data.to(device)            batch_targets = batch_targets.to(device)            optimizer.zero_grad()            outputs = model(batch_data)            loss = criterion(outputs, batch_targets)            loss.backward()            optimizer.step()            epoch_losses.append(loss.item())        avg_loss = np.mean(epoch_losses)        loss_history.append(avg_loss)        if (epoch + 1) % 10 == 0:            print(f"Epoch [{epoch + 1}/{num_epochs}], Loss: {avg_loss:.4f}")    return loss_history

def evaluate_model(model, dataloader, device='cpu'):    model.eval()    preds = []    trues = []    with torch.no_grad():        for batch_data, batch_targets in dataloader:            batch_data = batch_data.to(device)            outputs = model(batch_data)            preds.append(outputs.cpu().numpy())            trues.append(batch_targets.cpu().numpy())    preds = np.concatenate(preds, axis=0).squeeze()    trues = np.concatenate(trues, axis=0).squeeze()    return preds, trues

def visualize_results(loss_history, preds, trues):    sns.set(font_scale=1.2)    plt.rc('font', family=['Times New Roman', 'Simsun'], size=12)    # 图 1：训练损失曲线    # 模型在训练过程中损失的下降情况，说明模型不断优化拟合数据。    plt.plot(loss_history, marker='o', color='dodgerblue', linestyle='-', linewidth=2)    plt.title("Training Loss Curve")    plt.xlabel("Epoch")    plt.ylabel("MSE Loss")    plt.tight_layout()    plt.savefig('output_image1.png', dpi=300, format='png')    plt.show()    # 图 2：真实值与预测值对比曲线    # 对比曲线直观展示模型预测趋势与真实数据的匹配情况，越接近表示模型效果越好。    plt.plot(trues, label="True Values", color='limegreen')    plt.plot(preds, label="Predicted Values", color='crimson')    plt.title("True vs. Predicted Values")    plt.xlabel("Sample Index")    plt.ylabel("Trend Value")    plt.legend()    plt.tight_layout()    plt.savefig('output_image2.png', dpi=300, format='png')    plt.show()

testScore1 = math.sqrt(mean_squared_error(preds_test, trues_test))print('Test Score: %.2f RMSE' % (testScore1))testScore2 = mean_absolute_error(preds_test, trues_test)print('Test Score: %.2f MAE' % (testScore2))testScore3 = r2_score(preds_test, trues_test)print('Test Score: %.2f R2' % (testScore3))testScore4 = mean_absolute_percentage_error(preds_test, trues_test)print('Test Score: %.2f MAPE' % (testScore4))