Learning objectives
- • Separate data drift, concept drift, and performance degradation
- • Choose alerting thresholds that drive action rather than noise
- • Design retraining triggers and rollback criteria
MLOps
Monitoring, Drift, and Retraining
Build an operational view of production models that covers data quality, drift, business outcomes, and recovery.
Recommended on day 3490 minutesIntermediate
Interview prompts
- • What do you monitor first after launching a new ranking model?
- • How do you know when drift requires retraining versus investigation?
Prerequisites