Chapter 2: Data Drift Detection

Types of Drift

Type	Description	Detection
Covariate Drift	Input data distribution changes	Statistical tests on features
Concept Drift	Relationship between X and Y changes	Monitor prediction accuracy
Label Drift	Target distribution changes	Compare prediction distributions

Statistical Tests for Drift

from scipy import stats
import numpy as np

def detect_drift(reference_data: np.ndarray, current_data: np.ndarray, threshold: float = 0.05):
    """Detect drift using Kolmogorov-Smirnov test."""
    drift_results = {}
    
    for i, feature in enumerate(feature_names):
        statistic, p_value = stats.ks_2samp(
            reference_data[:, i],
            current_data[:, i]
        )
        drift_results[feature] = {
            'statistic': statistic,
            'p_value': p_value,
            'drift_detected': p_value < threshold
        }
    
    return drift_results

Drift Monitoring Dashboard

# Grafana dashboard config
panels:
  - title: "Feature Drift Score"
    type: timeseries
    targets:
      - expr: model_drift_score{feature="age"}
      - expr: model_drift_score{feature="income"}
    thresholds:
      - value: 0.1
        color: yellow
      - value: 0.2
        color: red

---

Next: Chapter 3: Alerting and Incident Response