DataTerminology.comNEW
A catalog of features with definitions, owners, and versions for reuse across models.
DESCRIPTION
A feature registry centralizes metadata about machine learning features, enabling discovery, reuse, and governance. It captures feature definitions, lineage, owners, versioning, and data sources. Registries help enforce consistency across training and serving, reducing duplication and drift. They integrate with feature stores, model registries, and orchestration tools. Governance policies and access controls ensure compliance and auditability. Strong registries accelerate ML development and facilitate reliable, cross-team collaboration.
ALTERNATIVE DEFINITIONS
Feast
The registry is a centralized metadata store for Feast that keeps track of feature definitions, data sources, and entities.
Defines the registry as the authoritative metadata store within a feature store.
Tecton
The feature registry is the system of record for feature metadata, including feature definitions, versions, owners, and materialization.
Emphasizes system-of-record responsibilities and feature lifecycle metadata.
Databricks
A feature store provides a centralized repository and governance for features used in ML models for both training and inference.
Positions centralized feature governance and reuse as core capabilities of the store and registry.
RELATED TERMS
Feature Store Metadata Management Data Discovery Platform Lineage Governance Feast Tecton Feature Engineering Data Contracts
COMPETING TERMS
Inline Feature Definitions Model Registry
KNOWLEDGE AREAS
MLOps Data Management Metadata Management Data Engineering Platform & Tooling
TARGET AUDIENCE
MLOps Engineer Data Engineer AI Engineer Data Scientist Data Steward Data Governance Manager
EXTERNAL SOURCES
- Wikipedia - Feature store
- Feast - Open feature store documentation
- Tecton - Vendor registry concept
- Databricks - Feature store and registry
TOP QUESTIONS
Why use a Feature Registry?
It prevents duplication, enforces consistency, and enables team-wide reuse. Registry metadata supports governance, lineage, and reliable training-serving parity.
Category: Relevancy
What metadata belongs in a Feature Registry?
Definitions, owners, versions, sources, entities, lineage, quality metrics, and access policies; ideally linked to model registry and data catalog.
Category: Requirements
How does a Feature Registry relate to a Feature Store?
The registry holds metadata; the store manages storage and serving. Together, they enable discovery and operational use of features.
Category: Overview
What is a common mistake with Feature Registries?
Treating it as static documentation; it must be integrated with pipelines, validation, and ownership workflows to stay accurate and actionable.
Category: Mistakes
How do registries support compliance?
They provide audit trails, access controls, and lineage, enabling oversight of sensitive features and ensuring regulatory compliance for ML use.
Category: Risk
What tools implement Feature Registries?
Feast, Tecton, Databricks Feature Store, and custom catalogs integrate registry capabilities with pipelines and model lifecycle management.
Category: Example
What’s the biggest misunderstanding about Feature Registries?
They are not just lists; they are authoritative systems enabling governance, versioning, and production-grade feature lifecycle management.
Category: Misunderstanding
How to start building a Feature Registry?
Define ownership and lifecycle policies, integrate with CI/CD and data catalog, and begin by registering high-impact features used across teams.
Category: Requirements
Can Feature Registries prevent training-serving skew?
Yes, by standardizing definitions and materialization, registries help align feature computation between training datasets and real-time inference.
Category: Opinion