robust_hpo — n4m.model_selection.aom_search.robust_hpo¶
Namespace: n4m.model_selection.aom_search · Fully-qualified: n4m.model_selection.aom_search.robust_hpo · Catalog id: aom_pop.robust_hpo
API surface¶
C ABI (ABI 2): n4m_model_selection_robust_hpo_fit. Use the linked public header for the exact signature, configuration, and result handles.
Python (verified public re-export): from n4m.model_selection.aom_search import aom_robust_hpo
Signature: aom_robust_hpo(X, y, *, profile: str | int = 'compact', cv: int = 5, heads = ('ridge', 'pls'))
R: no current source-verified entry point was found for this catalog method.
MATLAB / Octave: no current source-verified entry point was found for this catalog method.
Parameters¶
Name |
Type |
Default |
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Explanations¶
Bibliographic source¶
No single canonical paper defines this ABI-2 compact/wide screen. Beurier, G. et al. (2026). AOM-PLS / POP-PLS paper companion, arXiv:2605.13587, https://arxiv.org/abs/2605.13587. The product variants below are implementation-specific extensions; the paper does not by itself specify their ABI-2 orchestration.
Mathematical principle¶
Evaluate a finite bank of strict-linear spectral chains with Ridge and/or PLS heads by contiguous-fold CV RMSE, select the minimum-score tuple, refit on all calibration rows, and fold its linear coefficients back to the original feature space.
Appropriate uses¶
Fast, reproducible comparison of a declared preprocessing bank when the final model must remain a replayable linear predictor.
Limits and validation¶
Candidate selection must be nested inside external validation. Native v1 excludes stateful, sample-fitted, nonlinear, and source-routed transformations; a CUDA build does not make the complete candidate bank a fused GPU search.
Implementation¶
n4m.model_selection.aom_search.aom_robust_hpo and AOMRobustHPOSweepRegressor; C ABI n4m_model_selection_robust_hpo_fit.
Sources and provenance¶
https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/aom_robust_hpo.cpp; https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/c_api/c_api_method_result.cpp
Catalog note¶
Native strict-linear AOM robust-HPO screen over compact/wide preprocessing banks and Ridge/PLS heads. Compact has 12 chains; wide has 31 strict-linear chains, including Gaussian, FCK and Whittaker variants. Exposes input_coefficients plus intercept for replay in original feature space and backs NativeAOMRobustHPORegressor. Builds in CPU and CUDA-enabled libn4m configurations; native v1 is not the fused batched GPU grinder.
Timing benchmark: benchmarks/cross_binding/bench_aom_robust_hpo_timing.py
See also: methods index.