forecaster#
Core forecasting model interfaces and configurations.
Provides the fundamental building blocks for implementing forecasting models in OpenSTEF. These mixins establish contracts that ensure consistent behavior across different model types while supporting both single and multi-horizon forecasting scenarios.
Key concepts:
Horizon: The lead time for predictions, accounting for data availability and versioning cutoffs
Quantiles: Probability levels for uncertainty estimation
State: Serializable model parameters that enable saving/loading trained models
Batching: Processing multiple prediction requests simultaneously for efficiency
Multi-horizon forecasting considerations: Some models (like linear models) cannot handle missing data or conditional features effectively, making them suitable only for single-horizon approaches. Other models (like XGBoost) can handle such data complexities and work well for multi-horizon scenarios.
Classes#
|
Base for forecasters that handle multiple horizons simultaneously. |