shaving_select — Shaving (recursive elimination)

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

Description

Shaving iterative variable trimming

From the pls4all.sklearn.ShavingSelector docstring:

Iterative SR-shaving variable elimination.

Registry note — R plsVarSel::shaving(method='SR') iterative selectivity-ratio trimming — CV-error-minimising survivor set. Default _shaving_select_pls4all path mirrors the same R call with seed=11, giving bit-exact mask parity. The C++ step-count trajectory kernel is opt-in via legacy=True.

Parameters

Name

Type

Default

Notes

n_components

int

2

Number of latent components extracted (k).

n_steps

int

10

Number of elimination passes performed.

min_features

`int

None`

None

shave_fraction

float

0.2

Fraction of the worst-ranked variables removed at each shaving step.

n_folds

int

3

Number of cross-validation folds used inside the selector.

Explanations

Bibliographic source

Mehmood, T., Liland, K. H., Snipen, L. & Sæbø, S. (2012). A review of variable selection methods in partial least squares regression. Chemometrics and Intelligent Laboratory Systems 118, 62–69 (§3.2 Shaving).

Mathematical principle

Shaving recursively eliminates a fraction \(\rho \in (0, 1)\) of the lowest-scoring features at each step, refits PLS, and tracks the CV-RMSE of the shrinking subset. The subset with the lowest recorded CV-RMSE across the whole shaving trajectory is returned.

Compared to backward variable elimination (BVE — see next), shaving removes a batch of features per step instead of one, which is faster (\(O(\log p)\) steps vs \(O(p)\)) but more aggressive — a single bad shave removes many useful features irrecoverably. Recommended \(\rho \le 0.2\) to keep shave granularity reasonable.

Implementation

n4m_shaving_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_shaving_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 shaving_select_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
    res = shaving_select_fit(ctx, cfg, X, y, n_components=3)
# then: res.matrix("predictions"), res.matrix("coefficients"),
# res.vector("mask"), res.scalar("intercept"), …

from pls4all.sklearn import ShavingSelector
mdl = ShavingSelector(n_components=2, n_steps=10, min_features=None, shave_fraction=0.2, 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("shaving_select", X, y,
                      n_components = 3L, params = list(n_steps = 12L, min_features = 3L, shave_fraction = 0.2))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.

res = pls4all.shaving_select(X, y, 3);
% see header of bindings/matlab/+pls4all/shaving_select.m for full
% parameter surface:
%   res = shaving_select(X, Y, n_components, n_steps, min_features, shave_fraction)
yhat = predict(res, Xtest);

No idiomatic classdef wrapper — invoke pls4all.fit("shaving_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::shaving(method='SR') — iterative SR-shaving of low-importance features. Uses the same set.seed(11) as the pls4all wrapper so the CV fold assignments coincide.

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 1.001.0 s
Python · pls4all
pls4all.python✓ J 1.00440.9 ms
pls4all.sklearn⇄ J 0.502.32 ms🏆
R · pls4all
pls4all.R⇄ J 0.505.59 ms
pls4all.R.formula⇄ J 0.507.94 ms
pls4all.R.mdatools⇄ J 0.506.31 ms
pls4all.R.pls⇄ J 0.506.29 ms
R · external
📐ref.r_plsvarselsource38.2 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.00424.1 ms
Python · pls4all
pls4all.python✓ J 1.00445.2 ms
pls4all.sklearn⇄ J 0.502.24 ms🏆
R · pls4all
pls4all.R⇄ J 0.505.24 ms
pls4all.R.formula⇄ J 0.505.90 ms
pls4all.R.mdatools⇄ J 0.506.06 ms
pls4all.R.pls⇄ J 0.505.99 ms
R · external
📐ref.r_plsvarselsource94.0 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.00454.0 ms
Python · pls4all
pls4all.python✓ J 1.00443.2 ms
pls4all.sklearn⇄ J 0.502.26 ms🏆
R · pls4all
pls4all.R⇄ J 0.505.07 ms
pls4all.R.formula⇄ J 0.505.85 ms
pls4all.R.mdatools⇄ J 0.505.96 ms
pls4all.R.pls⇄ J 0.505.97 ms
R · external
📐ref.r_plsvarselsource35.1 ms

See also: benchmark overview · methods index · interactive dashboard