"""
Drift detection configuration for the MedConnect matching model.
Uses Evidently AI to monitor production data against the training baseline.
"""

# Select features to monitor -- not all features are equally important
# Choose features based on their importance to match quality
MONITORED_FEATURES = [
    # Add the features you want to monitor here
    # Consider: which features matter most for the model's predictions?
    # Consider: which features are most likely to change in production?
]

# Configure PSI thresholds -- defaults may not suit healthcare staffing
# The cost of a false alarm vs a missed drift matters for this client
DRIFT_THRESHOLDS = {
    # "feature_name": threshold_value,
    # Default PSI threshold is 0.25 -- is that appropriate for healthcare staffing?
}

# Reference dataset path (training data baseline)
REFERENCE_DATA_PATH = "materials/placement-data-training.csv"

# Current dataset path (recent production data)
CURRENT_DATA_PATH = "materials/placement-data-production.csv"


def configure_drift_report():
    """
    Configure the Evidently AI drift detection report.
    Returns a configured report object.
    """
    # Configure column mapping for the dataset
    # Specify which columns are numerical, categorical, and text

    # Configure the data drift report with DataDriftPreset
    # Set per-feature thresholds where appropriate

    # Add per-region drift analysis
    # Aggregate drift can mask subgroup-specific changes

    pass


def run_drift_detection():
    """
    Run drift detection and return results.
    """
    # Load reference and current datasets

    # Run the configured drift report

    # Extract per-feature drift results

    # Classify severity based on thresholds

    pass


if __name__ == "__main__":
    run_drift_detection()