etechoptimist

dvc stage add -n feature_engineering_analysis_oct_2025 \
  -d data/raw/analysis_oct_2025/analysis.csv \
  -d src/feature_engineering.py \
  -o data/features/engineered_features_analysis_oct_2025.parquet \
  python src/feature_engineering.py \
    --input data/raw/analysis_oct_2025/analysis.csv \
    --output data/features/engineered_features_analysis_oct_2025.parquet

etechoptimist

dvc add data/raw/analysis_oct_2025/analysis.csv
git commit -m "October 2025 monitoring dataset"
git add dvc.yaml dvc.lock src/feature_engineering.py data\features\.
dvc add data/raw/analysis_oct_2025/analysis.csv
dvc commit
git add data/raw/analysis_oct_2025/analysis.csv.dvc data/raw/analysis_oct_2025/.gitignore
git commit -m "October 2025 monitoring dataset"
git tag -a v-oct2025-monitoring -m "Monitoring dataset for October 2025"
dvc push

etechoptimist

dvc stage add -n monitoring_oct_2025 
-d data/features/engineered_features_analysis_oct_2025.parquet 
-d data/features/filtered_features.parquet -d models/ensemble_model.joblib 
-d src/monitoring_nannyml.py -o reports/monitoring/oct_2025 python src/monitoring_nannyml.py 
--input data/features/engineered_features_analysis_oct_2025.parquet 
--output reports/monitoring/oct_2025/performance_estimation.html

etechoptimist

# ----------------------------------------------------------------------------
# UNIVARIATE DRIFT + RCA
# ----------------------------------------------------------------------------
def run_univariate_drift(reference_df, analysis_df, selected_features, perf_results, output_dir: Path):
    logger.info("Running univariate drift + RCA...")

    uv = nml.UnivariateDriftCalculator(
        column_names=selected_features,
        continuous_methods=["kolmogorov_smirnov"],  # Best for continuous: detects distribution shape and location changes
        categorical_methods=["jensen_shannon"],  # Best for categorical: information-theoretic, symmetric distance

etechoptimist

# ----------------------------------------------------------------------------
# MULTIVARIATE DRIFT (PCA)
# ----------------------------------------------------------------------------
def run_multivariate_drift(reference_df, analysis_df, selected_features, output_dir: Path):
    logger.info("Running PCA multivariate drift...")
    

etechoptimist

# ----------------------------------------------------------------------------
# PERFORMANCE ESTIMATION (CBPE ONLY ON ANALYSIS DATA)
# ----------------------------------------------------------------------------
def run_performance_estimation(reference_df, analysis_df, output_dir: Path):
    import time
    logger.info("Running CBPE performance estimation...")
    logger.info(f"Analysis dataset: {len(analysis_df)} records")

    print(reference_df["y_true"].nunique())
    

etechoptimist

# ----------------------------------------------------------------------------
# MODEL SCORING
# ----------------------------------------------------------------------------
def score_model(df: pd.DataFrame, selected_features, model_path: str, is_reference: bool = False):
    logger.info(f"Scoring model: {model_path}")
    model = load(model_path)

    X = df[selected_features].fillna(0)

    df["y_pred_proba"] = model.predict_proba(X)[:, 1]

etechoptimist

# ----------------------------------------------------------------------------
# LOAD ENGINEERED MONITORING DATA AND FILTER BY selected_features
# ----------------------------------------------------------------------------
def load_and_filter_engineered_monitoring(engineered_path: str, selected_features):
    logger.info(f"Loading engineered monitoring dataset: {engineered_path}")
    df = pd.read_parquet(engineered_path)

    # Ensure timestamp exists
    if "timestamp" in df.columns:
        df["timestamp"] = pd.to_datetime(df["timestamp"])

etechoptimist

  
# ----------------------------------------------------------------------------
# LOAD TRAINING FILTERED FEATURES → DETERMINE SELECTED FEATURE LIST
# ----------------------------------------------------------------------------
def load_selected_features(filtered_training_path: str):
    df = pd.read_parquet(filtered_training_path)
    # all model features EXCEPT identifiers, typology, and transaction_type
    selected = [
        col for col in df.columns
        if col not in ["session_id", "table_id", "patron_id", "timestamp", "typology", "transaction_type"]