Drift Detection

Overview

Monitor data distribution changes over time to detect concept drift.

Centroid Drift

Detect changes in data center:

-- Detect centroid drift
SELECT centroid_drift_detection(
    'reference_table',
    'current_table',
    'features',
    0.1  -- threshold
);

Distribution Divergence

Measure distribution differences:

-- KL divergence detection
SELECT distribution_divergence(
    'reference_table',
    'current_table',
    'features',
    'kl'  -- divergence type
);

Temporal Monitoring

Track drift over time:

-- Monitor drift over time windows
SELECT temporal_drift_monitor(
    'time_series_table',
    'features',
    'timestamp_column',
    INTERVAL '1 day'  -- window size
);

Learn More

For detailed documentation on drift detection methods, setting thresholds, alerting, and model retraining strategies, visit: Drift Detection Documentation

Related Topics

• Outlier Detection - Detect anomalies
• Quality Metrics - Monitor data quality