nirs4all.pipeline.dagml.attested_by_source module
Native multi-output training captures and their exact host-sidecar bindings.
- nirs4all.pipeline.dagml.attested_by_source.bind_source_output_contract(dsl: dict[str, Any], dataset: Any, identity: IdentityMap, sample_ints: list[int], source_names: list[str]) None[source]
Sign the public typed vocabulary and experimental-unit design before compilation.
- nirs4all.pipeline.dagml.attested_by_source.execute_attested_by_source_cv(*, dsl: dict[str, Any], envelope: dict[str, Any], graph: dict[str, Any], spectro: Any, identity: IdentityMap, folds: list[tuple[list[int], list[int]]], source_names: list[str], selection_metric: str, random_state: int | None = None, refit_top_k: int = 1, refit: bool = True, group_by_sample: dict[int, str] | None = None, cli: str | None = None, python: str | None = None, dataset_path: str = '', dataset_pickle: str | None = None, workdir: Path | None = None) dict[str, Any][source]
Execute one signed native training request per selected rank, retaining every original capture.
Each extra rank repeats native candidate CV. Python never ranks scores or synthesizes a TrainingOutcome from another scheduler’s execution bundle.