scars_select — SCARS — Stability-CARS

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

Description

SCARS stability + CARS (§18 Phase 5h)

From the pls4all.sklearn.SCARSSelector docstring:

Stability-CARS hybrid (Zheng 2014).

Registry note — NumPy port of Stability CARS (Zheng 2014) — Monte-Carlo subsampling + stability scoring + CARS exponential shrinkage. Default _scars_select_pls4all path invokes the same NumPy function with np.random.default_rng(seed), giving bit-exact mask parity. The C++ splitmix64 kernel is opt-in via legacy=True.

Parameters

Name

Type

Default

Notes

n_components

int

2

Number of latent components extracted (k).

n_iterations

int

50

Number of selection iterations or Monte-Carlo passes.

min_features

`int

None`

None

sample_fraction

float

0.8

Fraction of samples drawn per Monte-Carlo replicate.

n_folds

int

3

Number of cross-validation folds used inside the selector.

seed

int

0

Random seed for reproducible sampling/initialization.

Explanations

Bibliographic source

Zheng, K., Li, Q., Wang, J., Geng, J., Cao, P., Sui, T., Wang, X. & Du, Y. (2012). Stability competitive adaptive reweighted sampling (SCARS) and its applications to multivariate calibration of NIR spectra. Chemometrics and Intelligent Laboratory Systems 112, 48–54.

Mathematical principle

Replace CARS’s coefficient-magnitude weights with coefficient-stability weights: \(w_j = |\bar{b}_j| / s(b_j)\) from the bootstrap distribution. Stability-weighted retention is more robust to spurious high-magnitude coefficients caused by particular bootstrap subsamples.

Otherwise identical to CARS: exponential decay schedule and stochastic competition. SCARS typically improves CARS on datasets with strong baseline drift or where a few high-leverage samples dominate the coefficient estimates.

Implementation

n4m_scars_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_scars_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 scars_select_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
    res = scars_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 SCARSSelector
mdl = SCARSSelector(n_components=2, n_iterations=50, min_features=None, sample_fraction=0.8, n_folds=3, seed=0)
mdl.fit(X, y)
y_hat = mdl.predict(X_test)

library(pls4all)
# Unified low-level dispatcher (May 2026 R cleanup):
res <- pls4all_method("scars_select", X, y,
                      n_components = 4L, params = list(n_iterations = 8L, min_features = 5L, sample_fraction = 0.5, seed = 11L))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.

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

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

Registry parity references 📐

  • 📐 ref.python_scars_numpy_port (python · python) — scars_numpy_port 1.0.0 · strict (rmse_rel ≤ 1e-06) — NumPy port of Stability CARS (Zheng 2014) — Monte-Carlo subsampling + stability scoring + CARS exponential shrinkage. Pinned np.random.default_rng(seed) for bit-exact reproducibility.

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 1.007.65 ms
Python · pls4all
pls4all.python✓ J 1.0010.2 ms
pls4all.sklearn⇄ J 0.334.14 ms
R · pls4all
pls4all.R⇄ J 0.334.85 ms
pls4all.R.formula⇄ J 0.335.55 ms
pls4all.R.mdatools⇄ J 0.335.92 ms
pls4all.R.pls⇄ J 0.335.72 ms
Python · external
📐ref.python_scars_numpy_portsource3.61 ms🏆
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.003.71 ms
Python · pls4all
pls4all.python✓ J 1.003.88 ms
pls4all.sklearn⇄ J 0.332.29 ms🏆
R · pls4all
pls4all.R⇄ J 0.335.06 ms
pls4all.R.formula⇄ J 0.337.16 ms
pls4all.R.mdatools⇄ J 0.336.22 ms
pls4all.R.pls⇄ J 0.336.17 ms
Python · external
📐ref.python_scars_numpy_portsource3.67 ms
BackendParity200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp✓ J 1.003.66 ms
Python · pls4all
pls4all.python✓ J 1.003.91 ms
pls4all.sklearn⇄ J 0.332.22 ms🏆
R · pls4all
pls4all.R⇄ J 0.334.89 ms
pls4all.R.formula⇄ J 0.3314.1 ms
pls4all.R.mdatools⇄ J 0.3317.6 ms
pls4all.R.pls⇄ J 0.3312.6 ms
Python · external
📐ref.python_scars_numpy_portsource9.38 ms

See also: benchmark overview · methods index · interactive dashboard