forecasting_workflow

Presets for building foundation-model forecasting workflows from config.

A ForecastingWorkflowConfig declares the model family, the checkpoint that backs it, the requested quantiles/horizons, the target column, and the columns to keep; create_forecasting_workflow() turns it into a CustomForecastingWorkflow with feature-selection preprocessing and quantile-sorting postprocessing. Every selected non-target column is forwarded to the model as a known covariate.

The checkpoint defaults to the published OpenSTEF Chronos-2 model on the HuggingFace Hub, so the minimal config is just:

workflow = create_forecasting_workflow(ForecastingWorkflowConfig())

Pick a different size or variant through Chronos2, or pass a LocalCheckpoint to run a file already on disk.

Functions

create_forecasting_workflow(config)

Build a foundation-model forecasting workflow from a declarative config.

Classes

BackendConfig	A backend configuration.
ForecastingWorkflowConfig(**data)	Declarative configuration for a foundation-model forecasting workflow.
OnnxBackendConfig(**data)	Compute configuration for an ONNX Runtime inference backend.