How would you design retraining triggers that do not oscillate?
PICTURE THIS: DATA SPLIT
Fit on train, tune on val, report on test once.
Simple meaning
Combine scheduled retrains with gates: drift plus a statistically significant drop on a delayed metric, plus a minimum data volume.
WHY — Concept Drift instead of guessing?
Why interviewers care about Concept Drift:
question about Concept Drift.
trade-offs, and what you would actually do on a MLOps project - not buzzwords.
Name the idea, why it exists, then one short example.
End with when you use it and one common pitfall.
STEPS — What happens step by step?
Before you speak the answer, walk the interviewer through these steps:
- 1Combine scheduled retrains with
gates: drift plus a statistically significant drop on a delayed metric, plus a minimum data volume.
- 2Require the candidate to
beat the champion on a locked eval set before swap.
- 3Cooldowns after a deploy
stop weekly flip-flops from noisy PSI.
- 4Give an example
One tiny concrete case you can say aloud.
- 5Common mistake
What juniors usually get wrong.
- 6Close
When you pick this over the alternative.
EXAMPLE — See it in action
Here's a short line you can speak, broken into clear beats:
Note: Adapt this scaffold to your own project — keep it under 60–90 seconds.
Key takeaway
Combine scheduled retrains with gates: drift plus a statistically significant drop on a delayed metric, plus a minimum data volume. Require the candidate to beat the champion on a locked eval set before swap.