Particle Swarm Optimization (PSO) – Python Implementation

This project provides a clean and extensible Particle Swarm Optimization (PSO) implementation in pure Python and NumPy. It is designed for research, teaching, and experimentation with continuous, possibly multi-modal optimization problems.

---

Overview

Particle Swarm Optimization is a population-based metaheuristic inspired by the collective motion of birds and fish. A swarm of candidate solutions (particles) moves through the search space, combining:

• Inertia – keeps the particle moving in its current direction.
• Cognitive component – pulls the particle toward its own best position.
• Social component – pulls it toward the best global position found by the swarm.

This implementation supports arbitrary search-space dimension and user-defined objective (fitness) functions.

---

Key Features

• Object-oriented, modular design (Particle, ParticleSwarmOptimizer).
• Fully configurable hyperparameters:
  • inertia weight w
  • cognitive and social coefficients c1 and c2
  • maximum velocity as fraction of domain size.
• Optional linear scheduling of the inertia weight and early stopping based on convergence.
• Built-in handling of box constraints (variable-wise lower and upper bounds).
• Reproducible runs via random seed.

---

Quick Example

import numpy as np
from pso import ParticleSwarmOptimizer

# Objective: minimum at x = 10
def fitness(x: np.ndarray) -> float:
	return float((x[0] - 10.0)**2)

def main() -> None:
	bounds = np.array([[0.0, 20.0]])  # search domain
	optimizer = ParticleSwarmOptimizer(
		fitness_f=fitness,
		domain_dim=1,
		particle_num=30,
		bounds=bounds,
		seed=0
	)

	best_pos = optimizer.optimize(num_iterations=200, verbose=True)
	print("Best position:", best_pos, "fitness:", optimizer.gbest_val)

if __name__ == "__main__":
	main()