ml-feature-engineering

Train-serve parity in ML feature engineering ensures consistent behavior between offline training and online serving. Validating parity prevents feature drift and leakage, maintaining model reliability across development and deployment pipelines.

Define ML feature specs using the feature-spec-template.csv to establish clear lineage and explicit ownership. This catalogs feature engineering workflows, ensuring reproducibility and satisfying requirements for versioned, auditable documentation.

Validate online-offline feature parity by applying rules from online-offline-parity-rules.md against your feature specs. This validation process checks consistency between training and serving paths to avoid data drift.

Model performance drops between training and serving due to broken train-serve parity. Inconsistent feature engineering workflows cause data drift and leakage, requiring explicit feature definition and parity validation plans to resolve.

Yes, this catalogs ML feature engineering workflows focusing on feature definition, lineage, and train-serve parity. It applies across model development and deployment to ensure versioned feature specs and auditable documentation.

Yes, explicit ownership is required for ML feature definitions. Auditable documentation with clear ownership and versioned feature specs ensures reproducibility and maintains consistent online-offline parity across pipelines.