Home / ML Associate practice test / Databricks Machine Learning / Question

Databricks Machine Learning Associate · free practice question

A fraud detection service needs sub-second feature retrieval for real-time scoring, while the data science team also needs full historical feature data for…

Exam-style question on Databricks Machine Learning from the free ML Associate practice test, with the answer and a full explanation below.

Updated November 10, 2026

Databricks Machine Learning

A fraud detection service needs sub-second feature retrieval for real-time scoring, while the data science team also needs full historical feature data for training and evaluation. Which statement best describes the role of offline and online feature storage in Databricks?

  1. AOffline feature tables are intended for dashboarding and reporting, while online feature tables are the source for both training and serving in Databricks model workflows that need low latency
  2. BOffline feature tables are needed when a model is built with AutoML, while online feature tables are intended for manually tuned models that are tracked outside MLflow and Unity Catalog
  3. COnline feature tables are used for long-running historical analysis and training, while offline feature tables are optimized for low-latency key-based retrieval by model serving endpoints
  4. DOffline feature tables store the governed source features used for training and batch workflows, while online feature tables support low-latency access for real-time applications and serving
  5. EOffline feature tables and online feature tables are interchangeable copies of the same data, with identical latency, history, and serving semantics for training and real-time applications
Answer & explanation

Correct answer: D

WHY D: Offline feature tables are the training and batch source of truth, while online feature stores are built for low-latency serving and real-time lookups.

WHY NOT A: Training does not require online stores; offline tables are the training source, not a dashboarding layer.

WHY NOT B: AutoML is unrelated to this distinction.

WHY NOT C: The latency roles are reversed.

WHY NOT E: The two storage modes have different purposes and behaviors.

Take the full ML Associate practice test free →

Would you pass today?

This question is one of 45 in the free, timed Databricks Machine Learning Associate practice test. Take it and get a section-by-section score.

Start the free test →