How-to Guides¶

Task-oriented recipes for getting a specific job done with Yohou-MLflow. These assume you already know the basics and want to accomplish something concrete.

• How to Check That a Registered Model Will Load Before Deploying: This guide shows you how to find out, before a deployment, whether an environment can load a registered forecaster. Use it when you upgrade yohou, scikit-learn or polars in the environment that runs scheduled forecasts.
• How to Forecast Through MLflow's Generic Predict Interface: This guide shows you how to forecast with a yohou model loaded through mlflow.pyfunc.load_model. Use it when your code loads MLflow models generically rather than through yohou_mlflow.load_model.
• How to Save a Forecaster That Contains Your Own Estimator Classes: This guide shows you how to save and load a forecaster whose estimators come from your own code or from a third-party library. Use it when save_model or log_model raises UntrustedTypesError.
• How to Keep the Registry Small When Every Run Registers a Version: This guide shows you how to delete old versions of a registered forecaster, and the files behind them, after each scheduled run. Use it when a job registers a new version on every run, since each version stores a full copy of the forecaster.
• How to Move Pickled Forecasters to the Yohou Flavour: This guide shows you how to re-save forecasters you stored with pickle, for example through MLflow's generic Python model, as yohou-flavour models. Use it when a job still logs yohou forecasters with mlflow.sklearn or a custom mlflow.pyfunc.PythonModel.
• Contributing to Yohou-MLflow: Thank you for your interest in contributing to Yohou-MLflow! This document provides guidelines for contributing to the project.

If you are still finding your feet, start with the Tutorials.