Transformer-time-series-forecasting/evaluate.py at main · zareenrahman/Transformer-time-series-forecasting · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36

import matplotlib.pyplot as plt
from sklearn.metrics import mean_absolute_error, mean_squared_error
import math
import torch

def evaluate_model(model, val_loader, device='cpu', pred_len=6, feature_idx=0, n_plot=3):
    model.load_state_dict(torch.load("best_transformer_model.pth"))
    model.eval()

    X_val_batch, y_val_batch = next(iter(val_loader))
    X_val_batch = X_val_batch.to(device)
    y_val_batch = y_val_batch.to(device)

    with torch.no_grad():
        y_pred_batch = model(X_val_batch)

    y_true = y_val_batch.cpu().numpy()
    y_pred = y_pred_batch.cpu().numpy()

    mae = mean_absolute_error(y_true.flatten(), y_pred.flatten())
    rmse = math.sqrt(mean_squared_error(y_true.flatten(), y_pred.flatten()))

    print(f"MAE: {mae:.4f}")
    print(f"RMSE: {rmse:.4f}")

    for i in range(n_plot):
        plt.figure(figsize=(10, 4))
        plt.plot(range(pred_len), y_true[i, :, feature_idx], label='True')
        plt.plot(range(pred_len), y_pred[i, :, feature_idx], label='Predicted')
        plt.title(f"Sample {i} - Feature {feature_idx}")
        plt.xlabel("Prediction step")
        plt.ylabel("Value (kW)")
        plt.legend()
        plt.grid(True)
        plt.show()