PMLE Practice Question: Which TWO actions are appropriate when you detect…

Which TWO actions are appropriate when you detect that a production model's prediction distribution has shifted significantly from the training distribution?

⚠ Common exam trap

Google Cloud often tests the misconception that immediate rollback or traffic reduction is the correct first action, when in fact the proper response is to investigate the cause before taking corrective action like retraining.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Retrain the model using the most recent data

Option C is correct because when a production model's prediction distribution has drifted from the training distribution, the underlying data distribution has likely changed, so retraining the model on the most recent data is the standard remediation to restore alignment between the model and current conditions. Option D is correct because before applying any fix, you must investigate the root cause of the shift — whether it is genuine data drift, a data pipeline/schema change, or a feature-engineering bug — since the appropriate corrective action depends on the cause. Option A is not appropriate as an automatic first step because rolling back only helps if the previous version is still valid for the current data distribution, and it may not address the underlying drift. Option B is not appropriate because simply increasing logging does not correct the shifted predictions, though it may support diagnosis. Option E is not appropriate because reducing traffic does not fix the drift and may harm service availability without addressing the model's degraded relevance.

Answer analysis

Option-by-option breakdown

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

✗
Immediately roll back to the previous model version
Why it's wrong here
Rolling back addresses the model version, but prediction-distribution shift can stem from changed input data, so reverting may not restore alignment and discards the current model prematurely. It is tempting because rollback is the standard remedy for a bad deployment, and it would be correct if the shift were traced to a faulty model release.
✗
Increase logging for future predictions
Why it's wrong here
Extra logging records future inputs and outputs but does nothing to diagnose or remediate the shift already detected, leaving the degraded model serving traffic. It is tempting because observability is a genuine response step, and it would be correct when the goal is gathering evidence to investigate a suspected but unconfirmed drift.
✓
Retrain the model using the most recent data
Why this is correct
Retraining on recent data lets the model learn the new input-output relationship underlying the shifted distribution, restoring calibration. This is appropriate once drift is confirmed, though pairing it with root-cause investigation avoids retraining on transient or corrupted data.
✓
Investigate the cause of the shift before taking corrective action
Why this is correct
Investigating the shift first distinguishes genuine concept drift from a pipeline bug, upstream data error, or seasonal variation. Acting without diagnosis risks retraining on faulty data or masking an infrastructure fault, so root-cause analysis should precede corrective retraining.
✗
Reduce the traffic to the model to minimize impact
Why it's wrong here
Throttling traffic reduces exposure but leaves the shifted model serving users and does not restore prediction quality, merely limiting blast radius. It is tempting because traffic shaping is a valid mitigation during incidents, and it would be correct if the model were being retired while a replacement is prepared.

Go deeper

PMLE Practice Question: Which TWO actions are appropriate when you detect…

Why each option matters

Option-by-option breakdown

Related to this question