Machine Learning - Monitoring

Key areas

• System: latency, throughput, error rates, saturation.
• Model: score distributions, confidence, data/label drift.
• Business: acceptance rate, conversions, risk thresholds.

Logging predictions

# Pseudocode for a logging middleware
# log = {"request_id": rid, "features": x_hash, "prediction": y, "score": p, "ts": now()}
# emit(log)