Feature Stores

Feature Store Systems

Centralized feature management for ML.

Purpose

Feature store: storage and serving of engineered features. Ensures consistency between training and serving.

Offline store: historical features for training. Online store: real-time features for inference.

Components

Feature registry: feature definitions. Backfill: compute historical features. Point-in-time correctness.

Tools

Feast: open-source feature store. Tecton, Vertex AI Feature Store: managed solutions.

Key Takeaways

1. Feature stores ensure consistency
2. Offline and online serving
3. Important for production ML