pso_select — PSO-PLS — Particle Swarm Optimisation

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

Description

PSO-PLS — Binary Particle Swarm variable selection (§48)

Registry note — Python pyswarms 1.3.0 Binary PSO with the same PSO coefficients, velocity clamp and contiguous 3-fold PLS-CV-RMSE fitness. Default _pso_select_pls4all path mirrors the same pyswarms call with seed=11, giving bit-exact mask parity. The C++ splitmix64 PSO kernel is opt-in via legacy=True.

Parameters

Name	Type	Default	Notes
n_components	int	2	current public binding signature
n_swarm	int	30	current public binding signature
n_iterations	int	50	current public binding signature
w	float	0.729	current public binding signature
c1	float	1.494	current public binding signature
c2	float	1.494	current public binding signature
v_max	float	4.0	current public binding signature
n_folds	int	3	current public binding signature
seed	int	0	current public binding signature

Explanations

Bibliographic source

Kennedy, J. & Eberhart, R. (1995). Particle swarm optimization. IEEE ICNN 1995, vol. 4, 1942–1948. — binary PSO variant used for variable selection. Verified primary link: https://doi.org/10.1109/ICNN.1995.488968.

Mathematical principle

Binary PSO maintains a swarm of particles where each particle’s position is a \(p\)-bit feature mask. Velocity updates blend the particle’s personal best, the swarm’s global best, and an inertia term; positions are stochastically rounded to 0/1 via a sigmoid.

Compared to GA-PLS, PSO converges faster on smooth fitness landscapes but is more susceptible to premature convergence on multi-modal ones. The two methods are complementary: PSO for quick reconnaissance, GA for final polish.

Recommended swarm size: 20–50. The fitness is again PLS CV-RMSE on the masked subset.

Appropriate uses

Population-based exploration of binary wavelength masks under a predictive fitness function.

Limits and validation

Binary PSO is a heuristic without global-optimum guarantees and can overfit a repeatedly queried validation set.

Implementation

n4m_feature_selection_pso_select. Reference: Python pyswarms for the PSO core, wrapped against PLS CV-RMSE.

Sources and provenance

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

API and bindings

C ABI (ABI 2): n4m_feature_selection_pso_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 PSO

Source signature: PSO(*, n_components: int = 2, n_swarm: int = 30, n_iterations: int = 50, w: float = 0.729, c1: float = 1.494, c2: float = 1.494, v_max: float = 4.0, n_folds: int = 3, seed: int = 0) (n4m/_impl/selection.py).

R (source-verified): pso_select(X, Y, n_components, n_swarm = 30L, n_iterations = 50L, w = 0.729, c1 = 1.494, c2 = 1.494, v_max = 4.0, seed = 0L).

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

MATLAB / Octave (source-verified): pso_select(X, Y, n_components, n_swarm, n_iterations, ... w, c1, c2, v_max, seed).

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

Registry parity references 📐

• 📐 ref.python_pyswarms (python · python) — pyswarms 1.3.0 · strict (rmse_rel ≤ 1e-06) — Python pyswarms.discrete.BinaryPSO 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 PSO kernel is opt-in via legacy=True.

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	80×25 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp	✓ J 1.00	219.8 ms
Python · archived pls4all benchmark
pls4all.python	✓ J 1.00	219.3 ms
pls4all.sklearn	⇄ J 0.46	4.05 ms🏆
R · archived pls4all benchmark
pls4all.R	⇄ J 0.46	6.11 ms
pls4all.R.formula	⇄ J 0.46	6.29 ms
pls4all.R.mdatools	⇄ J 0.46	7.33 ms
pls4all.R.pls	⇄ J 0.46	7.12 ms
Python · external
📐ref.python_pyswarms	source	163.8 ms

3 threads

Backend	Parity	80×25 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp	✓ J 1.00	169.0 ms
Python · archived pls4all benchmark
pls4all.python	✓ J 1.00	298.6 ms
pls4all.sklearn	⇄ J 0.46	3.17 ms🏆
R · archived pls4all benchmark
pls4all.R	⇄ J 0.46	20.9 ms
pls4all.R.formula	⇄ J 0.46	20.3 ms
pls4all.R.mdatools	⇄ J 0.46	16.0 ms
pls4all.R.pls	⇄ J 0.46	24.6 ms
Python · external
📐ref.python_pyswarms	source	217.8 ms

10 threads

Backend	Parity	80×25 (ms)
C++ native · libn4m
pls4all.cpp.blas+omp	✓ J 1.00	165.6 ms
Python · archived pls4all benchmark
pls4all.python	✓ J 1.00	181.1 ms
pls4all.sklearn	⇄ J 0.46	2.80 ms🏆
R · archived pls4all benchmark
pls4all.R	⇄ J 0.46	4.19 ms
pls4all.R.formula	⇄ J 0.46	4.78 ms
pls4all.R.mdatools	⇄ J 0.46	4.76 ms
pls4all.R.pls	⇄ J 0.46	4.74 ms
Python · external
📐ref.python_pyswarms	source	152.3 ms

---

See also: benchmark overview · methods index · interactive dashboard