Exact Paper Model Bank Support

This page documents the helper utilities used by the strict paper-faithful RCIM exact model-bank validation branch. The current RCIM Model-Bank Reproduction paper-faithful surface preserves the recovered original pipeline protocol and feeds the accepted forward/backward archives under models/paper_reference/rcim_track1.

RCIM Model-Bank Reproduction is closed at the full-dataset faithful model-bank level; later all-green or restricted-dataset studies are separate optimization/comparison branches.

Support utilities for the exact RCIM paper model-bank reimplementation.

scripts.paper_reimplementation.rcim_ml_compensation.exact_paper_model_bank.exact_paper_model_bank_support.load_exact_model_bank_config(config_path)[source]

Load one exact paper model-bank configuration file.

Parameters:

config_path (str | Path)

Return type:

dict[str, Any]

scripts.paper_reimplementation.rcim_ml_compensation.exact_paper_model_bank.exact_paper_model_bank_support.build_exact_paper_dataset_bundle(training_config)[source]

Build the exact paper dataframe split bundle.

Parameters:

training_config (dict[str, Any])

Return type:

ExactPaperDatasetBundle

scripts.paper_reimplementation.rcim_ml_compensation.exact_paper_model_bank.exact_paper_model_bank_support.create_exact_paper_base_estimator(family_name)[source]

Create one exact-paper base estimator matching the recovered workflow.

Parameters:

family_name (str)

Return type:

object

scripts.paper_reimplementation.rcim_ml_compensation.exact_paper_model_bank.exact_paper_model_bank_support.fit_exact_family_model_bank(dataset_bundle, enabled_family_list, training_config=None, best_parameter_override_map=None, workflow_stage='search')[source]

Fit the recovered family bank using the configured paper-side strategy.

Parameters:

  • dataset_bundle (ExactPaperDatasetBundle)

  • enabled_family_list (list[str])

  • training_config (dict[str, Any] | None)

  • best_parameter_override_map (dict[str, dict[str, Any]] | None)

  • workflow_stage (str)

Return type:

tuple[dict[str, MultiOutputRegressor], dict[str, dict[str, Any]]]

scripts.paper_reimplementation.rcim_ml_compensation.exact_paper_model_bank.exact_paper_model_bank_support.evaluate_exact_family_model_bank(dataset_bundle, fitted_family_model_dictionary)[source]

Evaluate the recovered family bank on the held-out test split.

Parameters:

  • dataset_bundle (ExactPaperDatasetBundle)

  • fitted_family_model_dictionary (dict[str, MultiOutputRegressor])

Return type:

tuple[list[dict[str, Any]], dict[str, list[dict[str, Any]]]]

scripts.paper_reimplementation.rcim_ml_compensation.exact_paper_model_bank.exact_paper_model_bank_support.export_exact_family_python_and_onnx_bank(dataset_bundle, fitted_family_model_dictionary, training_config, output_directory)[source]

Export one Python plus ONNX model bank per family and target when enabled.

Parameters:

  • dataset_bundle (ExactPaperDatasetBundle) – Prepared paper-faithful train/test bundle.

  • fitted_family_model_dictionary (dict[str, MultiOutputRegressor]) – Fitted family bank keyed by short family name.

  • training_config (dict[str, Any]) – Effective exact-paper configuration.

  • output_directory (Path) – Immutable validation artifact directory.

Returns:

Dictionary summarizing Python and ONNX export status, generated files, and comparison against the recovered ONNX release when configured.

Return type:

dict[str, Any]