Reinforcement learning project on Tombs
Tombs: https://github.com/Waznop/Tombs
RL Actor critic implementation inspired by: https://github.com/JamesonWeng/go-lite
TombsRL is a reinforcement learning project on an original board game called Tombs. The Actor Critic model achieved a winrate of ~65% against a random opponent after around 3 days of training. The current project was implemented with little knowledge of neural networks and minimal hyperparameters tuning. During a future revamp, I will fine-tune the model to better fit the learning environment, use CNNs, take partial observability into consideration and explore more state-of-the-art algorithms such as PPO.