In this talk we will introduce zipline, a feature engineering framework developed at airbnb to help data scientists take features to production – safely and quickly.
Zipline is used to
a.) Generate point-in-time correct feature backfills for model training with high efficiency,
b.) Generate online feature serving pipelines that can serve realtime feature aggregates with high availability, while also guaranteeing online-offline consistency.
c.) Incorporate Change Data Streams into features for sub-second data freshness