This was my final year project, where I trained various models to play the card game Coup, using Proximal Policy Optimisation (PPO). When evaluated against random these, models exceeded the current best. Overall, this project received 80/100. A more comprhensive description of my findings, and running the game will be added soon.
Install requirements.txt through pip install -r requirements.txt
Train models by running Coup/coup/train_ppo.py
Games againt a trained model is avaliable in Coup/coup/model_play_human.py