emcuve_select — EMCUVE — Ensemble MC-UVE

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

Description

EMCUVE ensemble MC-UVE (§18 Phase 5n)

From the pls4all.sklearn.EMCUVESelector docstring:

Ensemble Monte-Carlo UVE selector.

Registry note — R plsVarSel::mcuve_pls called n_ensembles times with deterministic seeds (11 + e) and vote-aggregated. Default _emcuve_select_pls4all path mirrors the same R loop, giving bit-exact mask parity. The C++ splitmix64 EMCUVE 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.

n_ensembles

int

10

Number of UVE replicates aggregated by majority vote.

vote_threshold

float

0.5

Minimum vote fraction required to retain a variable in EMCUVE.

n_folds

int

3

Number of cross-validation folds used inside the selector.

Explanations

Bibliographic source

Han, Q.-J., Wu, H.-L., Cai, C.-B., Xu, L. & Yu, R.-Q. (2008). An ensemble of Monte Carlo uninformative variable elimination for wavelength selection. Analytica Chimica Acta 612(2), 121–125. https://doi.org/10.1016/j.aca.2008.02.032 — extends the MC-UVE procedure of Cai et al. (2008) (stability_select) by aggregating independent MC-UVE rounds through a vote rule.

Mathematical principle

Run multiple independent MC-UVE rounds with different seeds, threshold each independently, then vote across rounds: a feature is selected if it survives thresholding in a majority of rounds. Robust against single-round instability caused by particular bootstrap samples.

The voting rule has a free parameter (majority threshold); the default of \(\lceil R/2 \rceil\) is the median-style majority. Stricter thresholds give smaller but more reliable subsets.

Implementation

n4m_emcuve_select.

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_emcuve_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 emcuve_select_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
    res = emcuve_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 EMCUVESelector
mdl = EMCUVESelector(n_components=2, noise_features=50, noise_seed=0, n_ensembles=10, vote_threshold=0.5, n_folds=3)
mdl.fit(X, y)
y_hat = mdl.predict(X_test)

library(pls4all)
# Unified low-level dispatcher (May 2026 R cleanup):
res <- pls4all_method("emcuve_select", X, y,
                      n_components = 4L, params = list(noise_features = 5L, n_ensembles = 5L, vote_threshold = 0.5, noise_seed = 11L))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.

res  = pls4all.fit("emcuve_select", X, y, "NumComponents", 4);
yhat = predict(res, Xtest);

No idiomatic classdef wrapper — invoke pls4all.fit("emcuve_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 repeated N times with different seeds, then vote-aggregated. Same algorithm family as pls4all’s EMCUVE. RNGs differ; mask metric ~0=perfect.

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 1.001.3 s
Python · pls4all
pls4all.python✓ J 1.001.3 s
pls4all.sklearn✓ J 1.001.98 ms🏆
R · pls4all
pls4all.R✓ J 1.004.51 ms
pls4all.R.formula✓ J 1.005.44 ms
pls4all.R.mdatools✓ J 1.005.25 ms
pls4all.R.pls✓ J 1.005.57 ms
R · external
📐ref.r_plsvarselsource922.4 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.001.3 s
Python · pls4all
pls4all.python✓ J 1.001.3 s
pls4all.sklearn✓ J 1.002.00 ms🏆
R · pls4all
pls4all.R✓ J 1.004.69 ms
pls4all.R.formula✓ J 1.005.25 ms
pls4all.R.mdatools✓ J 1.005.99 ms
pls4all.R.pls✓ J 1.006.05 ms
R · external
📐ref.r_plsvarselsource918.1 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.001.3 s
Python · pls4all
pls4all.python✓ J 1.001.3 s
pls4all.sklearn✓ J 1.001.90 ms🏆
R · pls4all
pls4all.R✓ J 1.004.22 ms
pls4all.R.formula✓ J 1.005.03 ms
pls4all.R.mdatools✓ J 1.004.95 ms
pls4all.R.pls✓ J 1.005.01 ms
R · external
📐ref.r_plsvarselsource907.4 ms

See also: benchmark overview · methods index · interactive dashboard