vissa_select — VISSA — Variable Iterative Space-Shrinkage

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

Description

VISSA-PLS — Variable Iterative Space Shrinkage (§49)

From the pls4all.sklearn.VISSASelector docstring:

Variable Iterative Subspace Shrinkage Approach (Deng 2014).

Registry note — Python auswahl.VISSA 0.9.0 (LSX-UniWue) — canonical Deng 2014 implementation via weighted binary matrix sampling. Default _vissa_select_pls4all path mirrors the same auswahl call with seed=11, giving bit-exact mask parity. The C++ splitmix64 VISSA 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

10

Number of selection iterations or Monte-Carlo passes.

n_submodels

int

60

Number of bootstrap sub-models drawn per VISSA iteration.

ratio_kept

float

0.1

Fraction of top-scoring features retained at each VISSA shrinkage step.

threshold

float

0.5

Inclusion-probability cut-off below which features are dropped.

floor_probability

float

0.05

Lower bound applied to per-feature inclusion probabilities to avoid premature pruning.

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

Deng, B. C., Yun, Y. H., Liang, Y. Z. & Yi, L. Z. (2014). A new strategy to prevent over-fitting in partial least squares models based on model population analysis. Analytica Chimica Acta 880, 32–41.

Mathematical principle

VISSA evaluates a population of random subsets of the same size, refines the population by selecting the best by CV-RMSE, and iteratively shrinks the search space toward features that survive in many high-performing subsets. Features that appear in many top subsets are deemed important; the search converges to a consensus subset.

Different from CARS in that the search space is shrunken by consensus over a population rather than by exponential decay over iterations. This gives smoother convergence and less sensitivity to single high-leverage subsets.

Implementation

n4m_vissa_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_vissa_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 vissa_select_fit
with pls4all.Context() as ctx, pls4all.Config() as cfg:
    res = vissa_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 VISSASelector
mdl = VISSASelector(n_components=2, n_iterations=10, n_submodels=60, ratio_kept=0.1, threshold=0.5, floor_probability=0.05, 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("vissa_select", X, y,
                      n_components = 3L, params = list(n_iterations = 10L, n_submodels = 60L, ratio_kept = 0.1, threshold = 0.5, floor_probability = 0.05, seed = 42L))
# res is a named list with MethodResult arrays/scalars.
# selected_indices / top_k_intervals are 1-based.

res  = pls4all.fit("vissa_select", X, y, "NumComponents", 3);
yhat = predict(res, Xtest);

No idiomatic classdef wrapper — invoke pls4all.fit("vissa_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.VISSA from LSX-UniWue with deterministic seed=11; the pls4all default path calls the same helper with the same seed, so masks coincide bit-for-bit. The C++ splitmix64 VISSA kernel is opt-in via legacy=True.

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.

BackendParity80×25 (ms)
Python · pls4all
pls4all.sklearn✓ J 0.6010.8 ms🏆
R · pls4all
pls4all.R✓ J 0.6011.2 ms
pls4all.R.formula✓ J 0.6011.1 ms
pls4all.R.mdatools✓ J 0.6049.1 ms
pls4all.R.pls✓ J 0.6021.4 ms
BackendParity80×25 (ms)
Python · pls4all
pls4all.sklearn✓ J 0.6013.0 ms
R · pls4all
pls4all.R✓ J 0.6013.9 ms
pls4all.R.formula✓ J 0.6014.6 ms
pls4all.R.mdatools✓ J 0.6011.4 ms🏆
pls4all.R.pls✓ J 0.6011.4 ms
BackendParity80×25 (ms)
Python · pls4all
pls4all.sklearn✓ J 0.608.85 ms🏆
R · pls4all
pls4all.R✓ J 0.6011.8 ms
pls4all.R.formula✓ J 0.6012.5 ms
pls4all.R.mdatools✓ J 0.6013.1 ms
pls4all.R.pls✓ J 0.6011.2 ms

See also: benchmark overview · methods index · interactive dashboard