Particle Swarm Optimization (PSO)

Implementation of Particle Swarm Optimization (PSO) consistent with the softpy library.

---

Architecture & Design

Library Integration

• Extends Softpy's FloatVectorCandidate and MetaHeuristicsAlgorithm abstract base classes
• Maintains API consistency with existing Softpy optimization algorithms
• Implements required interface methods (mutate(), recombine(), generate())

Design Patterns

• Factory Method: ParticleCandidate.generate() creates instances with proper initialization
• Template Method: Inherits optimization workflow from MetaHeuristicsAlgorithm
• State preservation through deep copy mechanics for best position tracking

Implementation Highlights

• Implemented from scratch: particle dynamics, velocity updates, and swarm evolution.
• OOP & Design Patterns: subclassed FloatVectorCandidate and applied the Factory Method.
• Metaheuristics: applied swarm intelligence concepts (personal, local, global bests).
• Numerical Computing: vectorized with NumPy, handled boundaries, and ensured reproducibility.
• Algorithm Engineering: managed in-place state updates using deep copies to preserve best positions.
• Testing: verified correctness on benchmark functions (Sphere, Rosenbrock).

---

Optimization Example

Visualization of PSO optimization on the Rosenbrock function:

Particles (red) converge toward the global optimum (star) as iterations progress.

---

Quick Start

Option A — Minimal demo

pip install -r requirements.txt
python run_demo.py

Option B — Full test harness + visualizations

Use the harness to run tests and generate optimization plots:

python test.py my_impl.py --pop 50 --nei 10 --iters 100

•	--pop: population size (number of particles)
•	--nei: number of neighbors considered for each particle
•	--iters: number of iterations (steps)

Example: sweep over multiple configurations in one run:

python test.py my_impl.py --pop 30,50 --nei 5,10 --iters 100,200

This will run PSO with:

• 2 different population sizes (30, 50)

• 2 different neighborhood sizes (5, 10)

• 2 iteration counts (100, 200)

→ creating a total of 8 runs and saving plots in the pictures/ folder.

Project Structure

├── docs/                   # Assignment specification (PDF)
├── pictures/               # Optimization plots generated by test harness
├── particle_candidate.py   # Particle representation (ParticleCandidate class)
├── optimizer.py            # Particle Swarm Optimizer class
├── my_impl.py              # Wrapper to expose both classes to the test harness
├── run_demo.py             # Minimal demo script for quick execution
├── test.py                 # Test harness with visualizations
├── requirements.txt 
├── README.md
└── .gitignore

Done by me, Stella Andorno (ialwayslikedgrime)