NEAT (NeuroEvolution of Augmenting Topologies) is a method developed by Kenneth O. Stanley for evolving arbitrary neural networks. This project is a Python implementation of NEAT. It was forked from the excellent project by @MattKallada, and is in the process of being updated to provide more features and a (hopefully) simpler and documented API.
For further information regarding general concepts and theory, please see Selected Publications on Stanley's website.
neat-python is licensed under the 3-clause BSD license.
Im simple terms, a file built on top of neat-python to run two NEAT agents simultaneously. This is exemplified in a 2 player veriosn of PONG.
This project is a compilation of two things:
- An edit of a file from
neat-pythonbased on the actual creation of each NEAT population - An example of an implementation of this new file using PONG on the Atari 2600 with emulation via OpenAI Retro
This is basically just a folder containg the custom code built on top of neat-python. It alters population.py in such a way that two independent populations are run per instance.
This is the file that actually runs two populations against eachother in a Video Game environment. Input data are metrics derived from the game based on the current score, and rectangular coordinates (eventually converted to polar coordinates) of sprites in the game.
*NOTE: in order to run this code, you must already have OpenAI Retro installed on your computer. The Pong-Atari2600 is a folder that goes in the /data folder of retro
Ex: C:\Users\john-doe\gym-retro\retro\data\stable\Pong-Atari2600