Chronos2Forecaster#

class openstef_foundation_models.models.forecasting.Chronos2Forecaster(*, quantiles: list[~openstef_core.types.Quantile] = [0.5], horizons: list[~openstef_core.types.LeadTime], supports_batching: bool = True, backend: ~openstef_foundation_models.inference.backend.InferenceBackend, hyperparams: ~openstef_foundation_models.models.forecasting.chronos2_forecaster.Chronos2HyperParams = <factory>)[source]

Bases: Forecaster

Zero-shot probabilistic forecaster backed by a Chronos-2 checkpoint.

The forecaster composes an InferenceBackend (built once and reused across an entire backtest) and translates between OpenSTEF datasets and the model’s tensor interface. Prediction is batch-first: predict_batch() runs the backend once over a stack of series and predict() is a batch-of-one wrapper.

Parameters:
HyperParams

alias of Chronos2HyperParams

backend: InferenceBackend
hyperparams: Chronos2HyperParams
supports_batching: bool
property hparams: Chronos2HyperParams

Model hyperparameters for training and prediction.

Concrete forecasters implement this by returning their narrowed hyperparams field, giving callers a polymorphic read-only view while each subclass keeps full type safety on its own field.

Overrides:

Forecaster.hparams

property is_fitted: bool

Check if the predictor has been fitted.

Overrides:

Predictor.is_fitted

fit(data: ForecastInputDataset, data_val: ForecastInputDataset | None = None) → None[source]

Fit the forecaster.

Chronos-2 is pretrained and zero-shot, so there is nothing to fit. The method exists only to satisfy the forecaster contract.

Parameters:
Return type:

None

Overrides:

Predictor.fit

predict(data: ForecastInputDataset) → ForecastDataset[source]

Forecast a single series.

Parameters:

data (ForecastInputDataset) – Input dataset whose target history provides the model context.

Return type:

ForecastDataset

Returns:

Probabilistic forecast for the requested quantiles and horizon.

Overrides:

Predictor.predict

predict_batch(data: list[ForecastInputDataset]) → BatchResult[ForecastDataset][source]

Forecast a batch of series in a single backend call.

Parameters:

data (list[ForecastInputDataset]) – Input datasets to forecast. Each provides its own target history and forecast start.

Return type:

BatchResult[ForecastDataset]

Returns:

One forecast per input dataset, in the same order.

Overrides:

BatchPredictor.predict_batch

model_config: ClassVar[ConfigDict] = {'arbitrary_types_allowed': True, 'extra': 'ignore', 'protected_namespaces': ()}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].