PMLE Practice Question: Collaborating Within and Across Teams to Manage Data and Models

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Enable point-in-time retrieval on the feature view.

  Why this is correct

  Point-in-time retrieval returns feature values as they existed at the supplied timestamp, joining each training row to its historical feature state. This directly prevents label leakage, satisfying the requirement that retrieved features exactly match values at a specific historical timestamp rather than current values.

• ✗

  Use the offline store without point-in-time and rely on data ordering.

  Why it's wrong here

  Relying on data ordering in the offline store without point-in-time retrieval returns the latest feature values, causing label leakage because training rows receive values recorded after the label timestamp. It is tempting because offline stores hold historical data, and would be correct only when training on current snapshots where temporal correctness is irrelevant.

• ✗

  Use the online store with a timestamp filter.

  Why it's wrong here

  The online store serves only the latest feature values, so a timestamp filter cannot reconstruct historical values and label leakage persists. It is tempting because the online store supports low-latency lookups, and would be correct for real-time inference serving, not for point-in-time correct training datasets.

• ✗

  Create a new feature view with only historical data.

  Why it's wrong here

  Historical data alone cannot serve point-in-time queries; the feature view must enable offline serving with a timestamp, which retrieves values as of that instant. A historical-only view is tempting for batch training exports, but it lacks the temporal lookup mechanism needed to prevent label leakage.

Go deeper

Related to this question