問題文
A team wants retraining to start automatically when the quality of the served model degrades. Which pair of triggers matches the mechanisms available on Databricks for this?
選択肢
- An increase in endpoint request volume, and an increase in endpoint latency, because these are the first signs that a model needs retraining, well before any metric moves.
- A schedule that runs weekly, and a manual approval step, because automated triggers based on metrics are not supported, so a person decides each cycle.
- An alert on a drift metric crossing a threshold, and an alert on a computed performance metric falling below a threshold, either of which can start the retraining job.
- A change in the feature table's schema, and the arrival of a new model version in the registry, because these are the events that indicate degradation, and either one can start the retraining job.