Evoman is a video game playing framework to be used as a testbed for optimization algorithms.

A demo can be found here: https://www.youtube.com/watch?v=ZqaMjd1E4ZI

Setup

To install requirements type in command: pip install -r requirements.txt

If you want to train NN with DEAP network run: python controller_specialist_deap.py

or

python controller_generalist_deap.py

Hyperparameters

In the file config.yaml there's a set of parameters you can tune. The config utilizes hydra library.

Inference

When you trained neural net from the above script you can see it in action with the following command: python controller_specialist_deap.py

DEAP

All files ending with "deap"

NEAT

All files in the folder "neat"

Optuna

In order to run training with automatic hyperparameter search, you can run one of the following commands:

python controller_specialist_deap.py --multirun

or

python controller_generalist_deap.py --multirun

The results of your experiments can be found in directory multirun/{timestamp}/optimization_results.yaml

Legend:

"optimization_" files are used to find the best solution

"contoller_..." these files run the soulution found by "optimization_..." files

(for Task 1 consider just the files containing "specialist" in their name

The given neural network is in the file "demo_controller.py"

All files containing "dummy" are draft files from where we can start to implement our own solutions

The neural net that was implemented by Jacob is in the folder "evolve"