`aom_operator_pls_stack` — n4m.ensemble.aom_operator_pls_stack¶

Namespace: n4m.ensemble · Fully-qualified: n4m.ensemble.aom_operator_pls_stack · Catalog id: aom_pop.operator_pls_stack

API surface¶

C ABI (ABI 2): n4m_ensemble_aom_operator_pls_stack_fit. Use the linked public header for the exact signature, configuration, and result handles.

Python (verified public re-export): from n4m.ensemble import aom_operator_pls_stack

Signature: aom_operator_pls_stack(X, y, *, profile: str | int = 'compact', cv: int = 5, fold_ids = None, components = (2, 4, 8), alphas = (0.001, 0.01, 0.1, 1.0, 10.0, 100.0), std_penalty: float = 0.0, gap_penalty: float = 0.0, center_x: bool | None = None, scale_x: bool | None = None, center_y: bool | None = None, scale_y: bool | None = None)

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
`X`	`—`	`required`
`y`	`—`	`required`
`profile`	`str \| int`	`'compact'`
`cv`	`int`	`5`
`fold_ids`	`—`	`None`
`components`	`—`	`(2, 4, 8)`
`alphas`	`—`	`(0.001, 0.01, 0.1, 1.0, 10.0, 100.0)`
`std_penalty`	`float`	`0.0`
`gap_penalty`	`float`	`0.0`
`center_x`	`bool \| None`	`None`
`scale_x`	`bool \| None`	`None`
`center_y`	`bool \| None`	`None`
`scale_y`	`bool \| None`	`None`

Explanations¶

Bibliographic source¶

No single paper defines this ABI-2 composition. 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¶

For each declared strict AOM view, standardize its \(n\times p\) matrix, fit a PLS1 projector with \(k_{eff}\) retained components, and concatenate the resulting latent scores into an \(n\times\sum_b k_{eff,b}\) design. A Ridge head is fitted on that design. Cross-validation scores and selects the component count and Ridge penalty; it does not use base-model prediction stacking.

Appropriate uses¶

Combining complementary strict AOM views through a compact supervised latent design for a univariate response.

Limits and validation¶

The implementation supports one response and relies on each view’s PLS1 score projection. OOF predictions are used to score candidate component/penalty settings, not as the columns of the final Ridge design. Correlated views can still make the Ridge weights unstable and selection needs an outer assessment.

Implementation¶

n4m.ensemble.aom_operator_pls_stack and AOMOperatorPLSStackRegressor; C ABI n4m_ensemble_aom_operator_pls_stack_fit.

Sources and provenance¶

https://github.com/GBeurier/nirs4all-methods/blob/main/cpp/src/core/aom_operator_pls_stack.cpp; https://github.com/GBeurier/nirs4all-methods/blob/main/bindings/python/src/n4m/ensemble.py

Catalog note¶

Native strict-linear AOM operator PLS1 score stack over compact/wide operator banks with a Ridge head. Compact has 12 operators; wide has 31 strict-linear operators, including Gaussian, FCK and Whittaker variants. Single-target only. It returns fold scores, final stack features, Ridge-head coefficients for audit, and folded input_coefficients plus input_intercept for replay through NativeAOMOperatorPLSStackRegressor; native v1 builds in CUDA-enabled configurations but is not yet a fused batched GPU stack.

Timing benchmark: benchmarks/cross_binding/bench_aom_operator_pls_stack_timing.py

See also: methods index.

aom_operator_pls_stack — n4m.ensemble.aom_operator_pls_stack¶