Pyman AI

Using darwinism to play Pacman (because it wasn’t already hard enought)

Overview

This project is a simple program that uses a neural network to play Pacman. To do the training we use NEAT, a kind of genetic algorithm that generates a population of NNs (in this case, NNs that will play Pacman) and tests them against an enviroment. The ones that perform the best get to pass down their genes

Requirements

• This custom implementation of Pacman that comes without ghosts (for the time being) and connects to the server that main.py creates.
• Python 3.6
  • Neat: pip install neat

How to run

1. Install Neat
2. Clone the game
3. run main.py to start the training from the beginning
   1. Additionally, you can run with it a file from checkpoints to see different NNs partially trained
4. Run the game

Notes

• This program was only tested on Ubuntu 18.04