What is training-serving skew?
mediumAnswer
- Training-serving skew is a mismatch between how features are computed at training and at serving time — different preprocessing code, different data sources, different schemas, different aggregations.
- Result: production performance is worse than offline eval.
- Prevent with a shared feature pipeline (feature store), unit tests that verify equivalent outputs, schema validation, and shadow-mode deployments before going live.
Check yourself — multiple choice
- A feature store increases training-serving skew
- Shared preprocessing code and feature stores reduce skew
- Skew only affects deep-learning models
- Shadow-mode deployment does not help detect skew
Feature stores + shared preprocessing minimize skew.
#mlops#deployment
Practise MLOps & Data Quality
215 interview questions in this topic.