nirs4all.pipeline.dagml.structural_topology module

Lower declared early/learned-late choices into native structural campaigns.

This bridge transports declarations and complete raw buffers. DAG-ML owns recipe generation, grouped nested OOF, conditional patches and scoring; Methods owns all learned encoders, branch predictors and the terminal Ridge.

nirs4all.pipeline.dagml.structural_topology.declared_topologies(steps: list[Any], splitter: Any) → list[list[Any]][source]

Validate public grammar without native state or recipe enumeration.

nirs4all.pipeline.dagml.structural_topology.is_topology_choice(steps: list[Any]) → bool[source]

Recognize existing sequence alternatives containing Methods declarations.

nirs4all.pipeline.dagml.structural_topology.lower_topology_choices(steps: list[Any], splitter: Any, schemas: Mapping[str, Any]) → tuple[dict[str, Any], dict[str, Any], list[str]][source]

Lower each public declaration once; native compilation owns expansion.

nirs4all.pipeline.dagml.structural_topology.prepare_topology_structure(steps: list[Any], splitter: Any, dataset: Any, spec: DagMLTuningSpec, run_options: dict[str, Any], native: Any) → dict[str, Any][source]

Bind complete raw identity and ask native DAG-ML to validate every scope.

nirs4all.pipeline.dagml.structural_topology.train_selected_topology(prepared: dict[str, Any], evidence: dict[str, Any], run_options: dict[str, Any]) → Any[source]

Resolve the exact native sink and capture all its portable predecessors.