nirs4all.pipeline.dagml.data_provider module

Execute an IO data source through the native preparation scheduler.

nirs4all.pipeline.dagml.data_provider.prepare_data_provider(dataset: Any, *, engine: str, should_stop: Any = None, pipeline: Any = None, refit: Any = None, tuning: Any = None, calibration: Any = None, terminal_predict: Any = None, save_artifacts: Any = None, project: Any = None, runner_kwargs: dict[str, Any] | None = None, session: Any = None, random_state: int | None = None) → Any[source]

Materialize a finite provider once, before folds or scientific fitting.

The provider owns its seed independently of estimator random_state. Native PLAN derives the effective task seed and attests execution; IO owns assembly and validation. Only the realized cohort enters training/replay.