Super Mario World playing agent developed using Retro Learning Environment library as a final project for the Machine Learning course at Universidade Federal do ABC.
This agent uses Neuroevolution of Augmenting Topologies (NEAT) in order to evolve a Neural Network capable of playing the stage.
For install requirements:
pip install -r requirements.txt
-
Run
evolve.py
. When it completes, it will have createdwinner
, a pickled version of the most fit genome. -
Run
play.py
. It will load the most fit genome fromwinner
and run it in a new simulation to show the behavior. (See a sample movie here.)