Feature Store
A centralized data system that manages, stores, and serves the features (input variables) used by machine learning models in both training and production.
Why Feature Stores?
ML features often require complex computation (aggregations, joins, transformations). Without a feature store, teams duplicate work, features drift between training and production, and debugging is difficult.
Key Capabilities
Feature computation and storage, versioning, point-in-time correctness (preventing data leakage), low-latency serving for real-time inference, and feature sharing across teams and models.
Tools
Feast (open-source), Tecton, Databricks Feature Store, SageMaker Feature Store, and Hopsworks. Feature stores are a core component of mature MLOps infrastructure.