iddaai-be/ai-engine/scripts/train_cards_model.py

"""
Card Market XGBoost Model Trainer
==================================
Kart (4.5 Alt/Üst, 5.5 Alt/Üst) için XGBoost modeli eğitir.

Usage:
    python3 scripts/train_cards_model.py
"""

import os
import sys
import pickle
import numpy as np
import pandas as pd
import xgboost as xgb
from sklearn.model_selection import train_test_split, StratifiedKFold
from sklearn.metrics import accuracy_score, log_loss, roc_auc_score, classification_report

# Config
AI_ENGINE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DATA_PATH = os.path.join(AI_ENGINE_DIR, "data", "training_data_cards.csv")
MODELS_DIR = os.path.join(AI_ENGINE_DIR, "models", "xgboost")

os.makedirs(MODELS_DIR, exist_ok=True)

# Feature columns
FEATURES = [
    # Referee features
    "ref_matches",
    "ref_avg_yellow",
    "ref_avg_red",
    "ref_avg_total",
    
    # Team features
    "home_team_matches",
    "home_team_avg_cards",
    "away_team_matches",
    "away_team_avg_cards",
    
    # League features
    "league_avg_cards",
    "league_match_count",
    
    # Derived
    "combined_team_avg",
    "ref_team_combined",
]


def load_data():
    if not os.path.exists(DATA_PATH):
        print(f"❌ Data file not found: {DATA_PATH}")
        print("   Run extract_card_training_data.py first!")
        sys.exit(1)
    
    print(f"📦 Loading data from {DATA_PATH}...")
    df = pd.read_csv(DATA_PATH)
    df.fillna(0, inplace=True)
    print(f"   Shape: {df.shape}")
    return df


def train_card_model(df, target_col, model_name):
    """Kart modeli eğit"""
    
    print(f"\n🚀 Training {model_name} (Target: {target_col})...")
    
    # Filter valid rows
    valid_df = df[df[target_col].notna()].copy()
    if valid_df.empty:
        print(f"   ⚠️ No valid data for {target_col}, skipping.")
        return None
    
    X = valid_df[FEATURES]
    y = valid_df[target_col].astype(int)
    
    print(f"   Target distribution: {dict(y.value_counts())}")
    
    # Split
    X_train, X_test, y_train, y_test = train_test_split(
        X, y, test_size=0.2, random_state=42, stratify=y
    )
    
    # Model params
    params = {
        'objective': 'binary:logistic',
        'eval_metric': 'logloss',
        'eta': 0.05,
        'max_depth': 5,
        'subsample': 0.8,
        'colsample_bytree': 0.8,
        'min_child_weight': 3,
        'nthread': 4,
        'seed': 42
    }
    
    # Train with cross-validation
    skf = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
    cv_scores = []
    
    for fold, (train_idx, val_idx) in enumerate(skf.split(X_train, y_train)):
        X_t, X_v = X_train.iloc[train_idx], X_train.iloc[val_idx]
        y_t, y_v = y_train.iloc[train_idx], y_train.iloc[val_idx]
        
        dtrain = xgb.DMatrix(X_t, label=y_t, feature_names=FEATURES)
        dval = xgb.DMatrix(X_v, label=y_v, feature_names=FEATURES)
        
        model = xgb.train(
            params,
            dtrain,
            num_boost_round=500,
            evals=[(dval, 'eval')],
            early_stopping_rounds=30,
            verbose_eval=False
        )
        
        preds = model.predict(dval)
        auc = roc_auc_score(y_v, preds)
        cv_scores.append(auc)
        print(f"   Fold {fold+1} AUC: {auc:.4f}")
    
    print(f"   Mean CV AUC: {np.mean(cv_scores):.4f} (+/- {np.std(cv_scores):.4f})")
    
    # Train final model on all training data
    dtrain_full = xgb.DMatrix(X_train, label=y_train, feature_names=FEATURES)
    dtest = xgb.DMatrix(X_test, label=y_test, feature_names=FEATURES)
    
    final_model = xgb.train(
        params,
        dtrain_full,
        num_boost_round=300,
        verbose_eval=False
    )
    
    # Evaluate
    test_preds = final_model.predict(dtest)
    test_pred_class = (test_preds > 0.5).astype(int)
    
    acc = accuracy_score(y_test, test_pred_class)
    auc = roc_auc_score(y_test, test_preds)
    
    print(f"\n📊 Test Results:")
    print(f"   Accuracy: {acc:.4f}")
    print(f"   AUC: {auc:.4f}")
    print(classification_report(y_test, test_pred_class))
    
    # Feature importance
    importance = final_model.get_score(importance_type='gain')
    print(f"\n🔍 Top Features:")
    sorted_importance = sorted(importance.items(), key=lambda x: x[1], reverse=True)[:5]
    for feat, score in sorted_importance:
        print(f"   {feat}: {score:.2f}")
    
    # Save model
    model_path = os.path.join(MODELS_DIR, f"xgb_{model_name.lower()}.json")
    final_model.save_model(model_path)
    print(f"\n💾 Model saved to: {model_path}")
    
    return final_model


def main():
    df = load_data()
    
    # Train multiple card models
    models = []
    
    # 1. Cards Over 4.5
    model_45 = train_card_model(df, "label_cards_over45", "cards45")
    models.append(("cards_over_45", model_45))
    
    # 2. Cards Over 3.5
    model_35 = train_card_model(df, "label_cards_over35", "cards35")
    models.append(("cards_over_35", model_35))
    
    # 3. Cards Over 5.5
    model_55 = train_card_model(df, "label_cards_over55", "cards55")
    models.append(("cards_over_55", model_55))
    
    print("\n" + "="*60)
    print("✅ All card models trained successfully!")
    print(f"📁 Models saved to: {MODELS_DIR}")
    
    # List saved files
    import glob
    card_files = glob.glob(os.path.join(MODELS_DIR, "xgb_cards*.json"))
    for f in card_files:
        print(f"   - {os.path.basename(f)}")


if __name__ == "__main__":
    main()