nirs4all.pipeline.dagml.training_compiler module

Prepared DAG-ML training-contract compiler for the native estimator.

P3-R0 deliberately does not lower arbitrary nirs4all pipeline syntax yet. It closes the next safe seam: an already-built DAG-ML training contract bundle can be injected into DagMLPipelineEstimator through a small compiler object, with fit identity diagnostics added in one place before the native binding call.

class nirs4all.pipeline.dagml.training_compiler.DagMLPreparedTrainingContracts(request: Any, data_envelopes: Mapping[str, Any], relations: Any, training_influence: Any, op_callback: Callable[[Any], Any] | None, outcome_id: str, run_id: str, bundle_id: str, warnings: Sequence[str] = (), diagnostics: Mapping[str, Any] | None = None, methods_inputs: Mapping[str, Any] | None = None, methods_library_path: str | None = None)[source]

Bases: object

Opaque DAG-ML training contracts already prepared by a trusted builder.

The fields map one-to-one to DagMLTrainingExecution and ultimately to dag_ml.execute_training(...). This type performs only seam-level validation: object identity, required text ids, callback callability, and diagnostics normalization. DAG-ML remains the authority for graph, relation, envelope, scoring, and influence semantics.

bundle_id: str

data_envelopes: Mapping[str, Any]

diagnostics: Mapping[str, Any] | None = None

methods_inputs: Mapping[str, Any] | None = None

methods_library_path: str | None = None

op_callback: Callable[[Any], Any] | None

outcome_id: str

relations: Any

request: Any

run_id: str

training_influence: Any

warnings: Sequence[str] = ()

class nirs4all.pipeline.dagml.training_compiler.DagMLTrainingContractFactoryCompiler(factory: ~collections.abc.Callable[[...], ~nirs4all.pipeline.dagml.training_compiler.DagMLPreparedTrainingContracts], include_fit_identity_diagnostics: bool = True, additional_diagnostics: ~collections.abc.Mapping[str, ~typing.Any] = <factory>)[source]

Bases: object

Compiler adapter around a callable that builds prepared contracts.

This is the intended handoff point for P3-R1 lowerers: the lowerer owns pipeline syntax, dataset envelopes, relations, influence, and callbacks; the shared seam below still validates and annotates the resulting native contract before the estimator calls DAG-ML.

additional_diagnostics: Mapping[str, Any]

compile_fit(estimator: DagMLPipelineEstimator, X: Any, y: Any, *, sample_ids: Any = None, groups: Any = None, metadata: Any = None, identity_frame: DagMLFitIdentityFrame) → DagMLTrainingExecution[source]: Build, validate and package contracts for one estimator fit call.

factory: Callable[[...], DagMLPreparedTrainingContracts]

include_fit_identity_diagnostics: bool = True

class nirs4all.pipeline.dagml.training_compiler.DagMLTrainingRequestCompiler(contracts: DagMLTrainingRequestContracts, include_fit_identity_diagnostics: bool = True, additional_diagnostics: Mapping[str, ~typing.Any]=<factory>, dagml_module: str | None = None)[source]

Bases: object

Compiler adapter for a request-spec plus executable DAG-ML contracts.

additional_diagnostics: Mapping[str, Any]

compile_fit(estimator: DagMLPipelineEstimator, X: Any, y: Any, *, sample_ids: Any = None, groups: Any = None, metadata: Any = None, identity_frame: DagMLFitIdentityFrame) → DagMLTrainingExecution[source]: Assemble a signed request and return native execution inputs.

contracts: DagMLTrainingRequestContracts

dagml_module: str | None = None

include_fit_identity_diagnostics: bool = True

class nirs4all.pipeline.dagml.training_compiler.DagMLTrainingRequestContracts(request_spec: DagMLTrainingRequestSpec, data_envelopes: Mapping[str, Any], relations: Any, training_influence: Any, op_callback: Callable[[Any], Any] | None, outcome_id: str, run_id: str, bundle_id: str, warnings: Sequence[str] = (), diagnostics: Mapping[str, Any] | None = None, methods_inputs: Mapping[str, Any] | None = None, methods_library_path: str | None = None)[source]

Bases: object

Executable contract bundle whose request still needs local assembly.

bundle_id: str

data_envelopes: Mapping[str, Any]

diagnostics: Mapping[str, Any] | None = None

methods_inputs: Mapping[str, Any] | None = None

methods_library_path: str | None = None

op_callback: Callable[[Any], Any] | None

outcome_id: str

relations: Any

request_spec: DagMLTrainingRequestSpec

run_id: str

to_prepared() → DagMLPreparedTrainingContracts[source]: Assemble the signed request and return prepared execution contracts.

training_influence: Any

warnings: Sequence[str] = ()

class nirs4all.pipeline.dagml.training_compiler.PreparedDagMLTrainingCompiler(contracts: DagMLPreparedTrainingContracts, include_fit_identity_diagnostics: bool = True, additional_diagnostics: Mapping[str, ~typing.Any]=<factory>)[source]

Bases: object

Compiler adapter for pre-built DAG-ML training contracts.

additional_diagnostics: Mapping[str, Any]

compile_fit(estimator: DagMLPipelineEstimator, X: Any, y: Any, *, sample_ids: Any = None, groups: Any = None, metadata: Any = None, identity_frame: DagMLFitIdentityFrame) → DagMLTrainingExecution[source]: Return native execution inputs for one estimator fit call.

contracts: DagMLPreparedTrainingContracts

include_fit_identity_diagnostics: bool = True

nirs4all.pipeline.dagml.training_compiler.compile_prepared_training_contracts(contracts: DagMLPreparedTrainingContracts, *, identity_frame: DagMLFitIdentityFrame, include_fit_identity_diagnostics: bool = True, additional_diagnostics: Mapping[str, Any] | None = None) → DagMLTrainingExecution[source]: Validate and package prepared DAG-ML training contracts for execution.