Zum Inhalt springen
Kitbay
Skill

mle-workflow

MITAktualisiert am 17. Nov. 2026305.644GitHub-Sterne

A skill from the ECC collection for planning and reviewing machine learning features that have to run in production, such as rankers, recommenders, classifiers, embeddings and forecasts.

Was der Skill vermittelt

A skill from the ECC collection for planning and reviewing machine learning features that have to run in production, such as rankers, recommenders, classifiers, embeddings and forecasts. It scales the process to the system instead of assuming labels, GPUs or A/B tests, and walks through a prediction contract, a data contract with point-in-time joins and snapshot IDs, a reproducible pipeline with pinned versions and recorded hashes, evaluation against a baseline with slice metrics and gates that fail closed, packaging for serving with input checks, and monitoring for drift with a rollback to a known good artifact. A twelve-point review checklist and a list of common mistakes, such as random splits on time data, close it out.

Funktionen

  • Define prediction and data contracts
  • Make training runs reproducible
  • Gate model promotion on evaluation results
  • Plan drift monitoring and rollback