Particle Swarm Optimization — Hidden Intelligence in Bird Flight

Minimal, hackable PSO playground built with pygame, numpy, and matplotlib. Tweak the fitness function, watch the swarm react, and keep iterating.

Quickstart

# using uv (recommended)
uv sync
uv run python main.py

# or plain pip
pip install -e .
python main.py

You need Python 3.13+ because this repo leans on the latest pygame wheels.

What You Get

• Vectorized PSO core with inertia/cognitive/social parameters exposed at the top of main.py.
• A 3D surface preview (preview_3d()) so you can sanity-check the objective before the swarm launches.
• A grab bag of alternate functions (Gaussian, Rastrigin, Ackley, Griewank, etc.) ready to uncomment.
• Color-mapped heat field rendered via pygame for instant feedback at 60 FPS.

Customize the Experiment

• Objective: swap the lambda assigned to f(x, y) or plug in your own function. Keep it vectorizable for fast previews.
• Swarm behavior: tune a, b, b_hat, and c to explore constriction vs. exploration regimes.
• Grid resolution: adapt WINDOW_SIZE and PARTICLE_SIZE to trade off fidelity vs. speed.
• Visualization: comment out preview_3d() if you want to skip the Matplotlib surface step.

Read More

• Full write-up on the underlying ideas
• _reading6 1995 particle swarming.pdf for the original Kennedy & Eberhart paper.