# `variable_select_vip` — VIP (Variable Importance in Projection) _Group_: **Variable selector** · _Registry tolerance_: `1e-06` ## Description VIP top-k variable selection (§18 Phase 5a, method=0) From the `pls4all.sklearn.VIPSelector` docstring: > Variable Importance in Projection top-k selector (Favilla 2013).
Full Python sklearn-wrapper docstring ```text Variable Importance in Projection top-k selector (Favilla 2013). Parameters ---------- top_k : int Number of features to keep. n_components, solver, center_x, scale_x, tol, max_iter Underlying PLS hyperparameters used for VIP scoring. Notes ----- Exposes ``vip_scores_`` as an alias for the generic ``scores_`` attribute, for callers used to the chemometrics naming convention. ```
> **Registry note** — R `plsVarSel::VIP` on `pls::plsr(method='kernelpls', scale=FALSE)`. pls4all pins the matching solver (`Solver.KERNEL_ALGORITHM`, `scale_x=False`, `scale_y=False`) and `compute_vip_scores` implements the same column-normalised W formula, so the selected-index masks agree bit-for-bit (`max_abs=0`). ### Parameters | Name | Type | Default | Notes | |------|------|---------|-------| | `top_k` | `int` | `None` | Number of features to retain. | | `n_components` | `int` | `2` | Number of latent components extracted (k). | | `solver` | `str` | `'simpls'` | Inner algorithm: 'nipals', 'simpls', 'svd', 'kernel', 'orthogonal-scores', 'power', 'randomized-svd', 'wide-kernel'. | | `center_x` | `bool` | `True` | Subtract the column mean of X before fitting. | | `scale_x` | `bool` | `True` | Standardize X columns to unit variance before fitting. | | `tol` | `float` | `1e-06` | Convergence tolerance for iterative solvers (NIPALS / power-iteration). | | `max_iter` | `int` | `500` | Maximum iterations for iterative solvers. | ## Explanations ### Bibliographic source Wold, S., Sjöström, M. & Eriksson, L. (2001). *PLS-regression: a basic tool of chemometrics*. Chemometrics and Intelligent Laboratory Systems 58(2), 109–130. ### Mathematical principle VIP scores quantify each feature's contribution across all $k$ latent components of a PLS model, weighted by how much each component explains of $\mathbf{y}$: $\mathrm{VIP}_j = \sqrt{\frac{p}{\mathrm{SSY}} \sum_{a=1}^{k} w_{ja}^2 \, \mathrm{SSY}_a}$, where $w_{ja}$ is the loading weight of feature $j$ in component $a$ and $\mathrm{SSY}_a$ is the explained sum of squares of $\mathbf{y}$ in component $a$. The normalisation guarantees $\sum_j \mathrm{VIP}_j^2 = p$, so the heuristic $\mathrm{VIP}_j > 1$ identifies features contributing more than their fair share. VIP is the workhorse of spectroscopic variable selection — simple, deterministic, fast, and well understood. ### Implementation `n4m_variable_select_rank` with metric=VIP. Reference: R `plsVarSel 0.10.0`. MATLAB header (`bindings/matlab/+pls4all/vip_select.m`): ```text pls4all.vip_select VIP-based feature ranking. res = pls4all.vip_select(X, Y, n_components, top_k) Fits an internal SIMPLS model (store_scores=1) and ranks features by their Variable Importance in Projection (VIP) scores. ``` ### 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** ::::{tab-set} :class: pls4all-bindings :::{tab-item} C ABI · libn4m :sync: c :class-label: lang-c ```c /* C ABI — libn4m */ n4m_context_t* ctx = n4m_context_create(); n4m_config_t* cfg = n4m_config_create(); n4m_method_result_t* res = NULL; n4m_variable_select_rank(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); ``` ::: :::{tab-item} Python · pls4all (raw) :sync: python-raw :class-label: lang-python ```python import pls4all from pls4all._methods import variable_select_rank with pls4all.Context() as ctx, pls4all.Config() as cfg: res = variable_select_rank(ctx, cfg, X, y, n_components=4) # then: res.matrix("predictions"), res.matrix("coefficients"), # res.vector("mask"), res.scalar("intercept"), … ``` ::: :::{tab-item} Python · pls4all.sklearn :sync: python-sklearn :class-label: lang-python ```python from pls4all.sklearn import VIPSelector mdl = VIPSelector(top_k, n_components=2, solver='simpls', center_x=True, scale_x=True, tol=1e-06, max_iter=500) mdl.fit(X, y) y_hat = mdl.predict(X_test) ``` ::: :::{tab-item} R · pls4all_method() :sync: r-dispatcher :class-label: lang-r ```r library(pls4all) # Unified low-level dispatcher (May 2026 R cleanup): res <- pls4all_method("variable_select_vip", X, y, n_components = 4L, params = list(top_k = 10L)) # res is a named list with MethodResult arrays/scalars. # selected_indices / top_k_intervals are 1-based. ``` ::: :::{tab-item} MATLAB · pls4all (MEX) :sync: matlab-mex :class-label: lang-matlab ```matlab res = pls4all.vip_select(X, y, 4); % see header of bindings/matlab/+pls4all/vip_select.m for full % parameter surface: % res = vip_select(X, Y, n_components, top_k) yhat = predict(res, Xtest); ``` ::: :::{tab-item} MATLAB · pls4all (classdef) :sync: matlab-classdef :class-label: lang-matlab _No idiomatic classdef wrapper — invoke `pls4all.fit("variable_select_vip", X, y, …)` directly from the unified MEX factory._ ::: :::: **Registry parity references** 📐 :::{card} :class-card: external-refs - 📐 **`ref.r_plsvarsel`** (R · r) — `plsVarSel` 0.10.0 · strict (rmse_rel ≤ 1e-06) — R `plsVarSel::VIP` ranking on `pls::plsr(method='kernelpls', scale=FALSE)` — matches the pls4all kernel-PLS path used by `_variable_select_rank_pls4all(rank_method=0)`. The top-k indices are returned (1-based -> 0-based in the loader). ::: ### Benchmarks Adaptive wall-clock per cell measured against [`full_matrix.csv`](../benchmarks/overview.md). 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  ·  ✗ 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`](../benchmarks/methodology.md)). 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. ::::{tab-set} :class: parity-tabs :::{tab-item} 1 thread :sync: threads-1
BackendParity50×250 (ms)100×50 (ms)100×500 (ms)100×2500 (ms)200×40 (ms)250×50 (ms)500×50 (ms)500×500 (ms)500×2500 (ms)2500×50 (ms)2500×500 (ms)2500×2500 (ms)10000×50 (ms)10000×500 (ms)
C++ native · libn4m
pls4all.cpp.blas✗ +4e-013.02 ms🏆2.60 ms18.5 ms89.6 ms2.54 ms4.69 ms16.3 ms🏆228.1 ms1.5 s552.5 ms9.6 s🏆53.9 s15.5 s177.3 s
pls4all.cpp.blas+omp✗ +4e-013.67 ms2.05 ms17.6 ms🏆88.2 ms2.38 ms8.24 ms17.4 ms216.0 ms🏆1.4 s🏆505.4 ms🏆9.9 s54.3 s14.3 s🏆175.5 s
pls4all.cpp.omp✗ +4e-013.42 ms1.59 ms19.0 ms85.4 ms2.57 ms4.84 ms18.4 ms217.6 ms1.4 s546.4 ms9.8 s62.4 s14.6 s172.8 s
pls4all.cpp.ref✗ +4e-015.69 ms1.71 ms19.1 ms83.7 ms🏆2.45 ms6.34 ms18.9 ms235.9 ms1.5 s531.8 ms10.2 s51.8 s🏆15.9 s159.5 s🏆
Python · pls4all
pls4all.python✓ bind5.74 ms———2.57 ms5.39 ms————————
pls4all.sklearn✓ bind3.28 ms———2.16 ms🏆3.61 ms🏆————————
R · pls4all
pls4all.R✓ bind16.6 ms———6.41 ms14.6 ms————————
pls4all.R.formula✓ bind28.7 ms———7.15 ms13.6 ms————————
pls4all.R.mdatools✓ bind23.1 ms———6.88 ms12.5 ms————————
pls4all.R.pls✓ bind28.3 ms———6.33 ms12.0 ms————————
MATLAB · pls4all
pls4all.matlab✗ +1e+004.15 ms———2.64 ms4.25 ms————————
pls4all.matlab.classdef✗ +1e+005.10 ms———3.26 ms5.45 ms————————
R · external
📐ref.r_plsvarselsource18.6 ms———17.0 ms24.8 ms————————
::: :::{tab-item} 3 threads :sync: threads-3
BackendParity50×250 (ms)100×50 (ms)100×500 (ms)100×2500 (ms)200×40 (ms)250×50 (ms)500×50 (ms)500×500 (ms)500×2500 (ms)2500×50 (ms)2500×500 (ms)2500×2500 (ms)10000×50 (ms)10000×500 (ms)
C++ native · libn4m
pls4all.cpp.blas✓ ref————2.45 ms—————————
pls4all.cpp.blas+omp✓ ref————2.89 ms—————————
pls4all.cpp.omp✓ ref————2.53 ms—————————
pls4all.cpp.ref✓ ref————2.42 ms—————————
Python · pls4all
pls4all.python✓ bind————2.34 ms—————————
pls4all.sklearn✓ bind————1.95 ms🏆—————————
R · pls4all
pls4all.R✓ bind————5.80 ms—————————
pls4all.R.formula✓ bind————6.22 ms—————————
pls4all.R.mdatools✓ bind————6.22 ms—————————
pls4all.R.pls✓ bind————6.08 ms—————————
MATLAB · pls4all
pls4all.matlab✗ +1e+00————2.51 ms—————————
pls4all.matlab.classdef✗ +1e+00————3.00 ms—————————
R · external
📐ref.r_plsvarselsource————15.7 ms—————————
::: :::{tab-item} 10 threads :sync: threads-10
BackendParity50×250 (ms)100×50 (ms)100×500 (ms)100×2500 (ms)200×40 (ms)250×50 (ms)500×50 (ms)500×500 (ms)500×2500 (ms)2500×50 (ms)2500×500 (ms)2500×2500 (ms)10000×50 (ms)10000×500 (ms)
C++ native · libn4m
pls4all.cpp.blas✓ ref————2.44 ms—————————
pls4all.cpp.blas+omp✓ ref————3.27 ms—————————
pls4all.cpp.omp✓ ref————2.30 ms—————————
pls4all.cpp.ref✓ ref————2.36 ms—————————
Python · pls4all
pls4all.python✓ bind————2.92 ms—————————
pls4all.sklearn✓ bind————1.57 ms🏆—————————
R · pls4all
pls4all.R✓ bind————3.96 ms—————————
pls4all.R.formula✓ bind————4.41 ms—————————
pls4all.R.mdatools✓ bind————4.50 ms—————————
pls4all.R.pls✓ bind————4.84 ms—————————
MATLAB · pls4all
pls4all.matlab✗ +1e+00————2.33 ms—————————
pls4all.matlab.classdef✗ +1e+00————2.60 ms—————————
R · external
📐ref.r_plsvarselsource————9.47 ms—————————
::: :::: --- _See also_: [benchmark overview](../benchmarks/overview.md) · [methods index](index.md) · [interactive dashboard](../landing/dashboard.md)