Encoding, scaling, splits that do not leak, and pipelines that produce the same features in training and serving.
10h · 7 lessons · 1 challenge
Before this: Dataframes & Exploratory Analysis, SQL & Data Modeling
Training/serving skew is the defect that survives every code review and kills the model in production. It is nearly always a feature computed one way offline and another way online.
This module builds features the way that avoids it: one transformation defined once, split correctly, validated, versioned. It is also where the leakage material from the dataframes module becomes a discipline rather than a warning.
Offline metrics are excellent, online is a disaster. Diff the two feature paths and find the divergence in each of eight scenarios.
Worth 500 XP.
