Model Retraining Strategy and Freshness Questions

Design strategies for detecting when models need retraining, how often to retrain, and how to manage retraining infrastructure. Balance between model freshness and computational cost. Strategies: periodic retraining (daily/weekly), trigger-based retraining (when performance degrades), or continuous learning. Handle rollback of bad model versions. Plan for canary deployments where new models are tested on small traffic before full rollout.