Feature Engineering Pipelines for Real-Time ML
Use same transformation code in training and serving when possible.
$$ ext{skew} = lVert f_{train}(x) - f_{serve}(x) Vert $$
Track skew continuously and block rollout if thresholds are exceeded.
post://feature-engineering-pipelines-realtime-ml
Feature Engineering Pipelines for Real-Time ML
author: Swadhin Biswas • • read: 1 min
