rep_select — REP — Recursive Elimination of Predictors

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

Description

REP-PLS repeated VIP selection (§18 Phase 5s)

Registry note — R plsVarSel::rep_pls repeated VIP-filtered selection. Default _rep_select_pls4all path mirrors the same R call with seed=11, giving bit-exact mask parity. The C++ step-count backward-elimination kernel is opt-in via legacy=True.

Parameters

Name	Type	Default	Notes
n_components	int	2	current public binding signature
n_steps	int	10	current public binding signature
min_features	int | None	None	current public binding signature
remove_count	int	1	current public binding signature
n_folds	int	3	current public binding signature
rep_ratio	float	0.5	registry benchmark cell value
rep_vip_threshold	float	0.5	registry benchmark cell value
rep_repeats	int	3	registry benchmark cell value

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. https://doi.org/10.1016/j.chemolab.2012.07.010 — same review as shaving_select; §3.3 Recursive elimination introduces the fixed-count variant implemented here.

Mathematical principle

REP removes a fixed count of features per recursive step (rather than a fraction as in shaving). At each step, sort features by absolute coefficient score, remove the \(m\) lowest, refit, record CV-RMSE. Return the subset with lowest CV-RMSE across all retained trajectories.

Useful when the analyst wants control over total iteration count: with \(m\) features removed per step, the process terminates in \(\lceil p / m \rceil\) iterations. Same intent as shaving but with linear instead of geometric decay.

Appropriate uses

Iterative elimination based on regression-coefficient-derived variable importance for parsimonious calibration.

Limits and validation

Coefficient rankings inherit scaling and collinearity sensitivity, and repeated validation queries can bias model choice.

Implementation

n4m_feature_selection_rep_select.

Sources and provenance

Current implementation: cpp/src/core/rep_selection.cpp.

API and bindings

C ABI (ABI 2): n4m_feature_selection_rep_select. Use the linked public header for the exact signature, configuration, and result handles.

Python (verified public re-export):

from n4m.feature_selection.wrapper import REP

Source signature: REP(*, n_components: int = 2, n_steps: int = 10, min_features: int | None = None, remove_count: int = 1, n_folds: int = 3) (n4m/_impl/selection.py).

R (source-verified): rep_select(X, Y, n_components, n_steps = 10L, min_features = 5L, remove_count = 1L).

The source signature has additional required inputs, so no example call is fabricated.

MATLAB / Octave (source-verified): rep_select(X, Y, n_components, n_steps, min_features, remove_count).

The source signature has additional required inputs, so no example call is fabricated.

Registry parity references 📐

• 📐 ref.r_plsvarsel (R · r) — plsVarSel 0.10.0 · strict (rmse_rel ≤ 1e-06) — R plsVarSel::rep_pls — repeated VIP-thresholded variable selection.

Benchmarks

Archived measurement identity. Backend labels in this table are the raw IDs recorded when the benchmark ran (including historical pls4all.* IDs). They preserve measurement provenance and do not describe a current public Python, R, or MATLAB binding; use the source-verified API and bindings section above for current entry points.

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 = archived binding-harness result 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; archived language-harness rows show binding parity against the C++ backend. Hover the icon for role and tolerance band.

1 thread

Backend	Parity	200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp	✓ J 1.00	605.7 ms
Python · archived pls4all benchmark
pls4all.python	✓ J 1.00	588.4 ms
pls4all.sklearn	⇄ J 0.21	3.35 ms🏆
R · archived pls4all benchmark
pls4all.R	⇄ J 0.21	5.16 ms
pls4all.R.formula	⇄ J 0.21	6.34 ms
pls4all.R.mdatools	⇄ J 0.21	6.53 ms
pls4all.R.pls	⇄ J 0.21	6.07 ms
R · external
📐ref.r_plsvarsel	source	241.7 ms

3 threads

Backend	Parity	200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp	✓ J 1.00	589.6 ms
Python · archived pls4all benchmark
pls4all.python	✓ J 1.00	594.0 ms
pls4all.sklearn	⇄ J 0.21	2.18 ms🏆
R · archived pls4all benchmark
pls4all.R	⇄ J 0.21	5.01 ms
pls4all.R.formula	⇄ J 0.21	5.61 ms
pls4all.R.mdatools	⇄ J 0.21	5.95 ms
pls4all.R.pls	⇄ J 0.21	5.87 ms
R · external
📐ref.r_plsvarsel	source	239.4 ms

10 threads

Backend	Parity	200×40 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp	✓ J 1.00	611.9 ms
Python · archived pls4all benchmark
pls4all.python	✓ J 1.00	630.6 ms
pls4all.sklearn	⇄ J 0.21	2.24 ms🏆
R · archived pls4all benchmark
pls4all.R	⇄ J 0.21	4.92 ms
pls4all.R.formula	⇄ J 0.21	6.09 ms
pls4all.R.mdatools	⇄ J 0.21	5.95 ms
pls4all.R.pls	⇄ J 0.21	5.79 ms
R · external
📐ref.r_plsvarsel	source	246.5 ms

---

See also: benchmark overview · methods index · interactive dashboard