mlflow_storage_callback

MLflow integration for tracking and storing forecasting workflows.

Provides a single callback for logging model training runs, artifacts, and metrics to MLflow. The callback is model-agnostic — it delegates to polymorphic methods on BaseForecastingModel and ModelFitResult so it works unchanged for both single-model and ensemble workflows.

Key behaviours:

• Logs model hyperparameters, plus per-component hyperparameters via model.component_hyperparams (e.g. per-forecaster in an ensemble).

• Stores training data, plus per-component datasets via result.component_fit_results.

• Collects evaluation metrics via result.metrics_to_flat_dict(); subclasses embed child metrics automatically.

• Stores feature-importance plots for every explainable component exposed by model.get_explainable_components().

• Supports model reuse (skip re-fit if a recent run exists) and model selection (keep the better model based on a configurable metric with a bias-towards-newer penalty).

Classes

MLFlowStorageCallback(**data)

MLFlow callback for logging forecasting workflow events.