rkoren-kitchen 1.0.0

Reusable MLOps framework: data ingestion, experiment tracking, serving, and monitoring

rkoren-kitchen

A reusable MLOps framework for competition and prediction-modeling projects: data ingestion, experiment tracking, model serving, monitoring, and cloud provisioning — driven by one menu.yaml manifest and one kitchen CLI.

It is intentionally generic: the same workflow helpers and CLI cover a large surface of Kaggle competitions and prediction-modeling projects with minimal per-project configuration.

Install

pip install rkoren-kitchen

Everything ships in the base install — the CLI, the training/serving/monitoring stack, and the recipes provisioning sub-package (YAML → Terraform → AWS).

What's in the box

• Modeling helpers — train_val_split, classification_metrics/regression_metrics, time_series_cv/loto_cv, calibration and ensembling utilities.
• Experiment tracking — an MLflow wrapper (experiment(), Tracker) with a champion registry and one-command auto-promotion.
• Pipeline — FeatureBuilder / Trainer / Evaluator stages plus a DataStore, run from the kitchen CLI (kitchen run …) or a menu.yaml (kitchen menu run).
• Serving & monitoring — a FastAPI serving scaffold (Lambda/ECR) and in-house drift reports.
• Provisioning — kitchen recipes … (a.k.a. the recipes CLI) turns a YAML spec into Terraform for S3/ECR/IAM/Lambda/RDS.

Quickstart

kitchen init my-project        # scaffold a project
kitchen run train              # train, track, and (optionally) auto-promote a champion
kitchen menu run               # or run the whole pipeline from menu.yaml

Stability

The public Python API is the top-level kitchen.__all__; the kitchen CLI and the menu.yaml schema (kitchen menu schema) are the other supported surfaces. The package follows SemVer.

See the project repository for full docs.

License

Public domain — released under The Unlicense. Feel free to use, modify, and contribute to this platform, however you like.