ridge_mkl_superblock — n4m.compose.aom_superblock.ridge_mkl_superblock¶
Namespace: n4m.compose.aom_superblock · Fully-qualified: n4m.compose.aom_superblock.ridge_mkl_superblock · Catalog id: aom_pop.ridge_mkl_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_mkl_superblock
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 canonical publication defines this product-specific MKL-light surface. 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 centered block \(Z_b\), compute \(K_b=Z_bZ_b^\top\) and its alignment \(A_b=\langle K_b,Y_cY_c^\top\rangle_F/(\lVert K_b\rVert_F\lVert Y_cY_c^\top\rVert_F)\). The top-\(k\) blocks receive simplex weights \(w_b=\max(A_b,0)/\sum_b\max(A_b,0)\); multiplying each retained block by \(\sqrt{w_b}\) before Ridge yields the final weighted superblock. If every retained alignment is nonpositive, the implementation uses uniform weights \(1/k\) instead.
Appropriate uses¶
Comparing a small collection of complementary strict spectral representations with a controlled linear head.
Limits and validation¶
The weighting is not equivalent to full multiple-kernel learning. The uniform fallback makes nonpositive alignments non-discriminative. Top-\(k\), weights, and Ridge penalty are selected fold-locally and refit on all rows; stateful/nonlinear branches break simple coefficient replay.
Implementation¶
n4m.compose.aom_superblock.aom_ridge_mkl_superblock and AOMRidgeMKLSuperblock; 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 MKL-light superblock over strict-linear single-operator AOM views (core fit_aom_superblock). It learns non-negative train-only kernel-target-alignment weights over the top operator blocks inside every alpha-CV fold, refits the weights on all rows, fits Ridge on the equivalent weighted superblock, and folds the final coefficients back to original-input input_coefficients plus intercept (affine N4MM state). It excludes donor branch_global, row-reference-dependent preprocessing, nonlinear kernels and nonlinear AOM Ridge modes.
Timing benchmark: benchmarks/cross_binding/bench_aom_ridge_mkl_superblock_timing.py
See also: methods index.