vip_spa_select — VIP-seeded SPA

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

Description

VIP_SPA — VIP-mask then SPA greedy (Phase 53)

From the pls4all.sklearn.VIPSPASelector docstring:

VIP_SPA — VIP-mask + SPA greedy (Phase 53).

Registry note — Python auswahl.VIP_SPA (LSX-UniWue) — VIP > 0.3 mask then greedy SPA pick. Default _vip_spa_select_pls4all path now invokes the same auswahl.VIP_SPA call, giving bit-exact mask parity. The C++ argmax-VIP SPA-start kernel is opt-in via legacy=True.

Parameters

Name

Type

Default

Notes

top_k

int

None

Number of features to retain.

n_components

int

2

Number of latent components extracted (k).

vip_threshold

float

0.3

Minimum VIP score required to enter the SPA candidate pool.

seed

int

7

registry benchmark cell value

Explanations

Bibliographic source

Hybrid heuristic combining VIP ranking and the Successive Projections Algorithm. See registry notes; no single canonical paper.

Mathematical principle

Use VIP scores to seed SPA’s projection-orthogonal forward selection. SPA starts with the highest-VIP feature rather than the highest-coefficient one, then proceeds with the standard projection step. This biases SPA toward \(y\)-correlated seed features while preserving SPA’s collinearity-minimising selection of subsequent features.

In practice this tends to outperform plain SPA on datasets where the first SPA seed is well-known to be noise-dominated (some real-world NIR datasets) but VIP correctly flags a different region as predictive.

Implementation

n4m_vip_spa_select.

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

pls4all.vip_spa_select  VIP-then-SPA hybrid (Phase 53).

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_vip_spa_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 vip_spa_select_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
    res = vip_spa_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 VIPSPASelector
mdl = VIPSPASelector(top_k, n_components=2, vip_threshold=0.3)
mdl.fit(X, y)
y_hat = mdl.predict(X_test)

library(pls4all)
# Unified low-level dispatcher (May 2026 R cleanup):
res <- pls4all_method("vip_spa_select", X, y,
                      n_components = 4L, params = list(vip_threshold = 0.3, top_k = 6L, seed = 7L))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.

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

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

Registry parity references 📐

  • 📐 ref.python_auswahl (python · python) — auswahl 0.9.0 · strict (rmse_rel ≤ 1e-06) — Python auswahl.VIP_SPA from LSX-UniWue. Same VIP scoring and 0.3 threshold as pls4all; auswahl enumerates every candidate SPA start and picks the CV-best, pls4all takes argmax-VIP within the mask. Mask metric.

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.

BackendParity80×40 (ms)
Python · pls4all
pls4all.sklearn✓ J 1.001.34 ms🏆
R · pls4all
pls4all.R✓ J 1.002.19 ms
pls4all.R.formula✓ J 1.002.80 ms
pls4all.R.mdatools✓ J 1.002.87 ms
pls4all.R.pls✓ J 1.002.83 ms
BackendParity80×40 (ms)
Python · pls4all
pls4all.sklearn✓ J 1.000.86 ms🏆
R · pls4all
pls4all.R✓ J 1.002.00 ms
pls4all.R.formula✓ J 1.003.05 ms
pls4all.R.mdatools✓ J 1.002.84 ms
pls4all.R.pls✓ J 1.003.05 ms
BackendParity80×40 (ms)
Python · pls4all
pls4all.sklearn✓ J 1.000.85 ms🏆
R · pls4all
pls4all.R✓ J 1.001.92 ms
pls4all.R.formula✓ J 1.002.79 ms
pls4all.R.mdatools✓ J 1.002.95 ms
pls4all.R.pls✓ J 1.003.11 ms

See also: benchmark overview · methods index · interactive dashboard