nirs4all.pipeline.dagml.source_missing module
Explicit missing-source prediction policy shared by live and archive hosts.
- nirs4all.pipeline.dagml.source_missing.append_source_presence(predictions: ndarray, presence: ndarray) ndarray[source]
Append one declared availability feature to one source’s predictions.
- nirs4all.pipeline.dagml.source_missing.build_prediction_availability(dataset: Any, identity: Any, sample_ids: list[int], *, target_values: Any, classification: bool) dict[str, Any][source]
Sign IO masks and encoded class IDs without inventing folds or labels.
- nirs4all.pipeline.dagml.source_missing.checked_prediction_features(spec: dict[str, Any], presence: ndarray) ndarray[source]
Consume native sparse-join features; zero cells are never distributions.
- nirs4all.pipeline.dagml.source_missing.checked_source_presence(resolver: Any, sample_ids: list[str], source_index: int, source_name: str, availability: dict[str, Any], *, fold_label: str | None = None, fitting: bool = False) ndarray[source]
Resolve true IO presence, checking signed Train rows before fitting.
- nirs4all.pipeline.dagml.source_missing.checked_target_validity(resolved: dict[str, Any], sample_ids: list[str], availability: dict[str, Any], *, fitting: bool) ndarray[source]
Verify actual target masks against signed Train authority for fit rows.
- nirs4all.pipeline.dagml.source_missing.predict_present_rows(predict: Callable[[Any], Any], block: Any, presence: ndarray, output_width: int) ndarray[source]
Predict observed rows in original target space; absent rows remain zero.
The callable must include any captured inverse target transform. No encoder or model is called when this source is absent for every requested row.