Glossary › Training-Serving Skew
Training-Serving Skew
When the feature values used during training differ from those used at inference time. Causes: different code paths for training vs. serving, different data sources, different preprocessing logic. One of the most common silent failure modes in production ML.
Where Training-Serving Skew shows up
All matching postsFeature Stores Explained: Do You Actually Need One?
Feature stores promise to solve training-serving skew and enable feature reuse. But they add real complexity. Understand what they actually do, when they pay off, and when they do not.
#feature-engineering#data-pipelines#mlops