A more in-depth fork of https://github.com/minerharry/neat_game_testing, attempting to teach a neural network how to play Super Mario Bros 1.
NOTE: for the game, I am using marblexu's (https://github.com/marblexu) awesome Python Mario Bros game, at https://github.com/marblexu/PythonSuperMario/. I have torn it apart to optimize and fit it with my infrastructure :).
Streamed on https://www.twitch.tv/minerharry 24/7 when running. Some extra details about the training progress can be found in the panels below the stream.
Resources:
- https://neat-python.readthedocs.io/en/latest/neat_overview.html - the python library I am using to run these tests
- http://www.cs.ucf.edu/~kstanley/#publications - papers about neat, including the original that started the whole thing
- http://nn.cs.utexas.edu/downloads/papers/stanley.ec02.pdf - a specific paper in the above list with some helpful information about the reproduction algorithm