nirs4all.pipeline.dagml.methods_classification module

Declarations and label transport for native Methods multimodal classifiers.

All encoders, PLS-logistic fitting, probability calculation, grouped folds and metrics remain in Methods/DAG-ML. This module only signs and decodes identities.

nirs4all.pipeline.dagml.methods_classification.checked_vocabulary(value: Any) → dict[str, Any]

Attest the exact signed contiguous-ID/original-label correspondence.

nirs4all.pipeline.dagml.methods_classification.classification_vocabulary(values: Any, train_rows: list[int]) → dict[str, Any]

Declare a sorted typed vocabulary using training rows alone.

nirs4all.pipeline.dagml.methods_classification.classifier_controller_for_graph(graph: Mapping[str, Any], cohort: Any, *, allow_fit: bool, node_params: Any = None, binding_source_ids: Any = None, binding_controller_id: str | None = None) → Any

Transport encoded identities to native owners with the signed graph edges.

nirs4all.pipeline.dagml.methods_classification.classifier_head(model: Any) → dict[str, Any]

Read one genuine native PLS-logistic declaration without fitting it.

nirs4all.pipeline.dagml.methods_classification.classifier_recipe(model: Any, *, allow_source_selection: bool = False) → dict[str, Any]

Reuse raw source declarations with the distinct native classifier head.

nirs4all.pipeline.dagml.methods_classification.decode_labels(values: Any, vocabulary: Any) → ndarray

Decode only exact native class IDs, preserving original scalar types.

nirs4all.pipeline.dagml.methods_classification.encode_labels(values: Any, vocabulary: Any) → ndarray

Encode target identities; a heldout-only or differently typed label fails.

nirs4all.pipeline.dagml.methods_classification.fit_declared_methods_classifier(model: Any, blocks: list[Any], y: Any, *, source_schemas: Any) → None

Fit genuine native state; a failed replacement leaves the old state intact.

nirs4all.pipeline.dagml.methods_classification.graph_vocabulary(graph: Mapping[str, Any]) → dict[str, Any]

Require one common signed vocabulary on all classifier producers.