A 311 team must ensure production enrichment uses the same feature store retrieval path the registered model expects. What should the registered model reference?
Select an answer to reveal the explanation.
Short Explanation
If 311 scores with different features than the model was built on, the answers get weird. Point the registered model at the same feature-store retrieval spec production enrichment uses.
Full Explanation
Referencing the feature store retrieval specification used for production enrichment on the registered model is correct so the 311 scoring path resolves the same features the model was trained to expect. Feature stores and retrieval specs exist to eliminate ad-hoc enrichment drift: neighborhood attributes, rolling complaint rates, and related signals must be fetched consistently online or in batch. Binding that specification to the registered model makes the contract discoverable for production enrichment jobs.
Relying on the feature store offline retrieval specification for production enrichment fails because it defines only the offline batch lookup path, not the authoritative production enrichment retrieval contract the registered model must reference, and would miss online features needed at scoring time. Pointing at the feature store online cache retrieval for production enrichment fails because a cache read is a serving optimization for latency, not the registered retrieval specification, and may return stale or incomplete values that diverge from the production feature contract. Using the feature store exploration retrieval for production enrichment fails because exploration retrieval is ad-hoc and ungoverned for discovery, not the versioned production feature contract the registered model should bind to.
Exam caveat: production retrieval must meet latency and availability needs of 311 scoring; the feature store offline retrieval specification for production enrichment cannot satisfy online scoring because it covers only the offline path. Version retrieval specs with models when feature definitions change. Operational check: configure the registered model to reference the production feature-store retrieval specification, execute a scoring enrichment path, and verify feature names, dtypes, and sample values match the training run’s feature set before go-live.
Prefer to study offline? Printable PDF study book + test engine ($7.50, unofficial). These questions stay free here.