train_models.py

# -*- coding: utf-8 -*-
"""
Script to pre-train and save K-Means models for the Gradio app.
Run this once to generate models/ folder with trained models.
"""

import sys
sys.path.insert(0, '../src')

from utils.data_loader import DataLoader
from utils.clustering_models import ClusteringModels
import os


def main():
    """Train and save models."""
    
    print("=" * 70)
    print("TRAINING K-MEANS MODELS FOR GRADIO APP")
    print("=" * 70)
    
    # Load data
    print("\n[1/4] Loading data...")
    data_loader = DataLoader("./data/processed")
    scaled_features = data_loader.scaled_features
    original_features = data_loader.original_features
    
    print(f"  Scaled features shape: {scaled_features.shape}")
    print(f"  Original features shape: {original_features.shape}")
    
    # Initialize clustering models
    print("\n[2/4] Initializing clustering models...")
    models_dir = "./models"
    os.makedirs(models_dir, exist_ok=True)
    
    cm = ClusteringModels(scaled_features, original_features, models_dir)
    
    # Train models
    print("\n[3/4] Training K-Means models (k=2 to k=10)...")
    cm.train_models(k_range=range(2, 11))
    
    # Apply PCA
    print("\n[4/4] Applying PCA for visualization...")
    cm.apply_pca(n_components=None)  # Keep all components
    
    # Save everything
    print("\n[5/5] Saving models to disk...")
    cm.save_models()
    
    print("\n" + "=" * 70)
    print("TRAINING COMPLETED SUCCESSFULLY!")
    print("=" * 70)
    
    # Print summary
    print("\nSummary:")
    print(f"  Models saved: {len(cm.kmeans_models)} (k=2 to k={max(cm.kmeans_models.keys())})")
    print(f"  PCA components: {cm.pca_features.shape[1]}")
    
    print("\n✓ Checking models...")
    print("\nSilhouette Scores by K:")
    for k, score in zip(range(2, 11), cm.silhouette_scores):
        print(f"  k={k}: {score:.4f}")
    
    best_k = range(2, 11)[cm.silhouette_scores.index(max(cm.silhouette_scores))]
    print(f"\nBest K (by Silhouette Score): {best_k}")


if __name__ == "__main__":
    main()