uve_select — UVE — Uninformative Variable Elimination

Group: Variable selector · Registry tolerance: 1e-06

Description

UVE noise-thresholded selection (§18 Phase 5d)

From the pls4all.sklearn.UVESelector docstring:

Uninformative Variable Elimination (Centner 1996) with artificial noise variables for the threshold.

Registry note — R plsVarSel::mcuve_pls Centner et al. 1996 Monte-Carlo UVE — augment X with ncol(X) uniform-noise columns, threshold real |mean/sd| stability scores against the noise max (fallback to top-ncomp |RI| when the survivor set is too small). Default _uve_select_pls4all path mirrors the same R call with seed=11, giving bit-exact mask parity. The C++ fixed-noise-count kernel is opt-in via legacy=True.

Parameters

Name

Type

Default

Notes

n_components

int

2

Number of latent components extracted (k).

noise_features

int

50

Number of artificial noise variables appended to X for the UVE threshold.

noise_seed

int

0

Seed for the appended noise variables.

min_features

`int

None`

None

Explanations

Bibliographic source

Centner, V., Massart, D. L., de Noord, O. E., de Jong, S., Vandeginste, B. M. & Sterna, C. (1996). Elimination of uninformative variables for multivariate calibration. Analytical Chemistry 68(21), 3851–3858.

Mathematical principle

UVE introduces a clever threshold: augment \(\mathbf{X}\) with \(p\) columns of deterministic artificial noise (unit-variance Gaussian, fixed seed), fit a PLS on the augmented \([\mathbf{X}, \mathbf{X}_{\mathrm{noise}}]\), then compute MC-UVE stability ratios for all columns. The noise columns establish a baseline distribution of stabilities under the null hypothesis ‘this column contributes nothing’. Any real feature whose stability falls below the maximum stability of the noise columns is eliminated.

The threshold is therefore data-adaptive — it tightens automatically as \(n\) grows (noise stabilities concentrate) and relaxes when the SNR is low. UVE is the canonical starting point for any chemometrics paper proposing a new selector; everything since is benchmarked against it.

Implementation

n4m_uve_select. Reference: R plsVarSel.

MATLAB header (bindings/matlab/+pls4all/uve_select.m):

pls4all.uve_select  Uninformative Variable Elimination (artificial-noise threshold).

Usage

Every pls4all binding tab dispatches into the same C kernel; the external libraries listed at the bottom of the page are the parity references registered in benchmarks.parity_timing.registry. Switch tabs to read the same fit in your language. The R package now ships drop-in-compatible facades for the CRAN pls package (plsr, pcr, mvr) and for the mdatools::pls(x, y, ...) matrix idiom — those tabs appear only on the methods that have a meaningful equivalence.

pls4all bindings

/* C ABI — libn4m */
n4m_context_t* ctx = n4m_context_create();
n4m_config_t*  cfg = n4m_config_create();
n4m_method_result_t* res = NULL;
n4m_uve_select_fit(ctx, cfg, &x_view, &y_view, /* hyperparams */, &res);
/* … read coefficients / mask / scores via */
/* n4m_method_result_get_double_matrix / vector / scalar … */
n4m_method_result_destroy(res);
n4m_config_destroy(cfg);
n4m_context_destroy(ctx);

import pls4all
from pls4all._methods import uve_select_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
    res = uve_select_fit(ctx, cfg, X, y, n_components=4)
# then: res.matrix("predictions"), res.matrix("coefficients"),
# res.vector("mask"), res.scalar("intercept"), …

from pls4all.sklearn import UVESelector
mdl = UVESelector(n_components=2, noise_features=50, noise_seed=0, min_features=None)
mdl.fit(X, y)
y_hat = mdl.predict(X_test)

library(pls4all)
# Unified low-level dispatcher (May 2026 R cleanup):
res <- pls4all_method("uve_select", X, y,
                      n_components = 4L, params = list(noise_features = 5L, noise_seed = 11L))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.

res = pls4all.uve_select(X, y, 4);
% see header of bindings/matlab/+pls4all/uve_select.m for full
% parameter surface:
%   res = uve_select(X, Y, n_components, noise_features, noise_seed)
yhat = predict(res, Xtest);

No idiomatic classdef wrapper — invoke pls4all.fit("uve_select", X, y, …) directly from the unified MEX factory.

Registry parity references 📐

  • 📐 ref.r_plsvarsel (R · r) — plsVarSel 0.10.0 · strict (rmse_rel ≤ 1e-06) — R plsVarSel::mcuve_pls with a noise threshold — UVE elimination of low-stability features.

Benchmarks

Adaptive wall-clock per cell measured against full_matrix.csv. Only backends that implement this method are listed; libraries without the method are omitted.

Verdict  ·  ✓ ref / ≈ ref / ~ shape mark a reference-gate pass at strict / relaxed / qualitative tolerance  ·  ✓ bind = pls4all binding agrees with the C++ baseline  ·  ⇄ cross-check = documented by-design selector/RNG/model, noncanonical API/facade convention, or secondary oracle  ·  ✗ divergent  ·  ⚠ error  ·  — not run. The fastest backend per column is marked 🏆.

Reference gate: strict — numeric equivalence (rmse_rel_tol ≤ 1e-06).

Rows tagged with 📐 are the canonical parity references for this method (declared in parity_timing.registry). C++ and external rows show reference parity; pls4all language bindings show binding parity against the C++ backend. Hover the icon for role and tolerance band.

BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.001.5 s
Python · pls4all
pls4all.python✓ J 1.001.7 s
pls4all.sklearn⇄ J 0.753.28 ms🏆
R · pls4all
pls4all.R⇄ J 0.7510.3 ms
pls4all.R.formula⇄ J 0.757.89 ms
pls4all.R.mdatools⇄ J 0.756.20 ms
pls4all.R.pls⇄ J 0.756.77 ms
R · external
📐ref.r_plsvarselsource222.0 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.00607.0 ms
Python · pls4all
pls4all.python✓ J 1.00640.4 ms
pls4all.sklearn⇄ J 0.751.81 ms🏆
R · pls4all
pls4all.R⇄ J 0.754.61 ms
pls4all.R.formula⇄ J 0.755.51 ms
pls4all.R.mdatools⇄ J 0.755.45 ms
pls4all.R.pls⇄ J 0.755.70 ms
R · external
📐ref.r_plsvarselsource397.3 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.00737.2 ms
Python · pls4all
pls4all.python✓ J 1.001.5 s
pls4all.sklearn⇄ J 0.752.82 ms🏆
R · pls4all
pls4all.R⇄ J 0.759.76 ms
pls4all.R.formula⇄ J 0.759.70 ms
pls4all.R.mdatools⇄ J 0.755.84 ms
pls4all.R.pls⇄ J 0.756.21 ms
R · external
📐ref.r_plsvarselsource230.2 ms

See also: benchmark overview · methods index · interactive dashboard