FEATURE-STORE-EDI

Versioned ML feature repository for EDI.

Definition

Leading Feature Stores: Feast (open-source, Tecton-backed), Tecton (managed), Databricks Feature Store, Sagemaker Feature Store, Vertex AI Feature Store. Solve the training-serving skew: same transformations in batch (Spark) and online (Redis). In EDI: features = partner average delay, % syntax errors, D-30 volume.

Origin

Concept popularised by Uber Michelangelo in 2017 ; Feast open-sourced in 2019.

Example in context

An EDI hub feeds Feast with partner behavioural features to predict INVOIC delays.

Related terms

• MLflow — experiment tracking.