ridge_superblock — n4m.compose.aom_superblock.ridge_superblock

Namespace: n4m.compose.aom_superblock · Fully-qualified: n4m.compose.aom_superblock.ridge_superblock · Catalog id: aom_pop.ridge_superblock

API surface

C ABI: no standalone exported symbol is declared for this method.

Python (verified public re-export): from n4m.compose.aom_superblock import aom_ridge_superblock

Signature: aom_ridge_superblock(X, y, *, operators = None, alpha: float | None = None, alphas: Sequence[float] = (0.0001, 0.01, 1.0, 100.0), cv: int = 5, fold_ids = None, block_scaling: str = 'rms', center_x: bool = True, center_y: bool = True)

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

operators

—

None

alpha

float | None

None

alphas

Sequence[float]

(0.0001, 0.01, 1.0, 100.0)

cv

int

5

fold_ids

—

None

block_scaling

str

'rms'

center_x

bool

True

center_y

bool

True

Explanations

Bibliographic source

No canonical paper defines this ABI-2 superblock wrapper. 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

Create \(B\) branch outputs \(Z_b\in\mathbb{R}^{n\times p}\), concatenate \(Z=[Z_1|\cdots|Z_B]\in\mathbb{R}^{n\times Bp}\), center columns, optionally RMS-scale each block, then fit Ridge on the resulting superblock. Its penalty is selected and refit within the declared CV protocol.

Appropriate uses

Linear calibration from multiple explicitly declared preprocessing branches.

Limits and validation

Column scale and block width affect Ridge regularization; centering, RMS scaling and penalty selection must be fold-local. The method is restricted to the declared strict-linear branch family for raw-space replay.

Implementation

n4m.compose.aom_superblock.aom_ridge_superblock and AOMRidgeSuperblock; no standalone C ABI symbol.

Sources and provenance

https://github.com/GBeurier/nirs4all-methods/blob/main/bindings/python/src/n4m/_impl/native.py; https://github.com/GBeurier/nirs4all-methods/blob/main/bindings/python/src/n4m/compose/aom_superblock.py

Catalog note

Native AOM Ridge superblock over strict-linear single-operator AOM views (core fit_aom_superblock). It concatenates the operator outputs, centers the columns and optionally scales each block to unit RMS with train-fold statistics, selects the Ridge alpha by train-fold CV, and folds the final superblock coefficients back to original-input input_coefficients plus intercept, so the role state is an affine N4MM predictor. It excludes donor branch_global, MKL/kernel, row-reference-dependent preprocessing and nonlinear AOM Ridge modes.

Timing benchmark: benchmarks/cross_binding/bench_aom_ridge_superblock_timing.py

See also: methods index.