Question 25
Domain 1: Databricks Machine LearningWhich use case most clearly calls for an offline feature table rather than an online feature store?
Correct answer: A
Explanation
An offline feature table fits when features are used for “training a model on historical data” and “scheduled batch scoring,” because those workflows do not need low-latency lookups. The exam guide distinguishes “online and offline feature tables,” and online stores are for serving features at inference time, not batch-only processing.
Why each option is right or wrong
A. Training a model on historical data and running scheduled batch scoring
The exam guide’s feature-store objective explicitly separates online from offline feature tables, and batch-only workflows belong on the offline side because they do not require low-latency retrieval at request time. In this scenario, the model is fit on past records and then scored on a schedule, so the feature values can be read from stored historical data rather than served in milliseconds from an online store.
B. Serving a loan decision inside a live API request
Live API requests need realtime inference and low-latency feature access.
C. Returning a fraud score in milliseconds at checkout
Millisecond checkout scoring is a realtime serving use case for an online feature store.
D. Fetching features during an interactive mobile session
Interactive mobile sessions typically need online, low-latency feature retrieval.