We have devised a system to create a fitness function that will allow for more controllability as well as the ability to bring coherency back into these levels. We accomplish this in a two-pronged approach: controllability is increased by using a metric that measures how much leeway there is to perform a given action and coherence is reintroduced through the employment of a Long Short Term Memory Recurrent Neural Network (LSTM)

Installation instructions for PyTorch can be found here

We use pyjnius to wrap the Java emulator into Python. Cython is required to install pyjnius.
TQDM displays a progress bar and is helpful for seeing how long the model will take the train

pip install cython
pip install pyjnius
pip install cma
pip install tqdm

Java 1.8 is required as it is used for the Mario AI compeition 2009 A* agent made by Robin Baumgarten, A* agent.

• Luke Zeller - LukeZeller
• Justin Mott - justinmmott
• Saujas Nandi - s-nandi

This project is licensed under the MIT License - see the LICENSE.md file for details

• Volz et al. - DagstuhlGAN