Original ONNX Release Parity Runner

This page documents the evaluation-only parity runner used to compare the recovered paper-original ONNX release against the current repository rcim_original forward archive.

Validate recovered original ONNX release parity against repo archives.

class scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.OnnxTargetEntry(family_name, target_kind, harmonic_order, target_name, onnx_model_path)[source]

Bases: object

One recovered ONNX target model resolved from the release tree.

Parameters:

  • family_name (str)

  • target_kind (str)

  • harmonic_order (int)

  • target_name (str)

  • onnx_model_path (Path)

family_name: str
target_kind: str
harmonic_order: int
target_name: str
onnx_model_path: Path
scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.format_project_path(path_value)[source]

Format a path relative to the repository when possible.

Parameters:

path_value (Path)

Return type:

str

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.resolve_target_name(target_kind, harmonic_order)[source]

Resolve the recovered exact-paper target name.

Parameters:

  • target_kind (str)

  • harmonic_order (int)

Return type:

str

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.load_yaml_dictionary(yaml_path)[source]

Load one YAML dictionary.

Parameters:

yaml_path (Path)

Return type:

dict[str, Any]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.write_yaml_dictionary(yaml_path, payload)[source]

Write one YAML dictionary with stable formatting.

Parameters:

  • yaml_path (Path)

  • payload (dict[str, Any])

Return type:

None

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.build_onnx_release_manifest(onnx_release_root)[source]

Build a deterministic manifest for the recovered ONNX release.

Parameters:

onnx_release_root (Path)

Return type:

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

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.predict_onnx_model(onnx_model_path, feature_matrix)[source]

Run one ONNX target model on the CPU provider.

Parameters:

  • onnx_model_path (Path)

  • feature_matrix (DataFrame | ndarray)

Return type:

ndarray

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.predict_python_model(python_model_path, family_name, feature_matrix)[source]

Run one archived Python target model.

Parameters:

  • python_model_path (Path)

  • family_name (str)

  • feature_matrix (DataFrame)

Return type:

ndarray

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.build_repo_reference_entry_lookup(repo_archive_root)[source]

Build a lookup for current repo archived original forward models.

Parameters:

repo_archive_root (Path)

Return type:

dict[tuple[str, str, int], dict[str, Any]]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.compute_target_metric_dictionary(truth_vector, prediction_vector)[source]

Compute target-level regression metrics.

Parameters:

  • truth_vector (ndarray)

  • prediction_vector (ndarray)

Return type:

dict[str, float]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.compute_curve_mean_centering_metric_dictionary(truth_curve_deg, predicted_curve_deg)[source]

Compute raw and mean-centered metrics for one TE Curve Verification Pipeline curve.

Parameters:

  • truth_curve_deg (ndarray)

  • predicted_curve_deg (ndarray)

Return type:

dict[str, float]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.compute_improvement_percent(raw_metric_value, adjusted_metric_value)[source]

Compute percentage improvement from a raw metric to an adjusted metric.

Parameters:

  • raw_metric_value (float)

  • adjusted_metric_value (float)

Return type:

float

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.summarize_mean_centering_metric_dictionary(metric_dictionary_list)[source]

Summarize one list of mean-centering metric dictionaries.

Parameters:

metric_dictionary_list (list[dict[str, float]])

Return type:

dict[str, float]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.build_table_parity_result(exact_config_path, source_dataframe_path, onnx_manifest_dictionary, repo_reference_entry_lookup)[source]

Evaluate ONNX and repo original archives on the exact-paper split.

Parameters:

  • exact_config_path (Path)

  • source_dataframe_path (Path)

  • onnx_manifest_dictionary (dict[tuple[str, str, int], OnnxTargetEntry])

  • repo_reference_entry_lookup (dict[tuple[str, str, int], dict[str, Any]])

Return type:

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

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.build_onnx_prediction_dictionary_for_track2(curve_record_list, family_name, onnx_manifest_dictionary, selected_harmonic_list)[source]

Predict all ONNX targets needed by one TE Curve Verification Pipeline family.

Parameters:

  • curve_record_list (list[HarmonicCurveRecord])

  • family_name (str)

  • onnx_manifest_dictionary (dict[tuple[str, str, int], OnnxTargetEntry])

  • selected_harmonic_list (list[int])

Return type:

dict[str, ndarray]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.evaluate_track2_onnx_family(curve_record_list, family_name, onnx_manifest_dictionary, selected_harmonic_list, percentage_error_denominator)[source]

Evaluate one recovered ONNX family through TE Curve Verification Pipeline curve reconstruction.

Parameters:

  • curve_record_list (list[HarmonicCurveRecord])

  • family_name (str)

  • onnx_manifest_dictionary (dict[tuple[str, str, int], OnnxTargetEntry])

  • selected_harmonic_list (list[int])

  • percentage_error_denominator (str)

Return type:

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

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.build_track2_parity_result(track2_config_path, onnx_manifest_dictionary)[source]

Evaluate recovered ONNX release and repo original archive in TE Curve Verification Pipeline.

Parameters:

  • track2_config_path (Path)

  • onnx_manifest_dictionary (dict[tuple[str, str, int], OnnxTargetEntry])

Return type:

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

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.summarize_table_parity(target_result_list)[source]

Build family-level Tables 2-5 parity summary.

Parameters:

target_result_list (list[dict[str, Any]])

Return type:

list[dict[str, Any]]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.write_target_parity_csv(csv_path, target_result_list)[source]

Write target-level parity rows to CSV.

Parameters:

  • csv_path (Path)

  • target_result_list (list[dict[str, Any]])

Return type:

None

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.write_track2_offset_diagnostic_csv(csv_path, offset_entry_list)[source]

Write TE Curve Verification Pipeline per-curve raw and mean-centered diagnostic rows.

Parameters:

  • csv_path (Path)

  • offset_entry_list (list[dict[str, Any]])

Return type:

None

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.build_parity_report_markdown(validation_summary)[source]

Build the Markdown report for original ONNX release parity.

Parameters:

validation_summary (dict[str, Any])

Return type:

str

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.run_original_onnx_release_parity_validation(exact_config_path, track2_config_path, onnx_release_root, repo_archive_root, source_dataframe_path, output_suffix)[source]

Run the original ONNX release parity validation.

Parameters:

  • exact_config_path (Path)

  • track2_config_path (Path)

  • onnx_release_root (Path)

  • repo_archive_root (Path)

  • source_dataframe_path (Path)

  • output_suffix (str)

Return type:

tuple[Path, Path]

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.parse_arguments()[source]

Parse command-line arguments.

Return type:

Namespace

scripts.paper_reimplementation.rcim_ml_compensation.reference_family_vs_feedforward.run_original_onnx_release_parity_validation.main()[source]

Run the command-line entry point.

Return type:

None