I have been working on Reinforcement Learning for years, on and off. I decided to dedicate some time in July and August to working on it, a couple of hours a day on average. I implemented several RL algorithms, including DQN and Policy Gradient (REINFORCE), by hand across multiple Atari games, and utilized Stable Baselines for standardized benchmarking. I aim to expand the number of games and algorithms, creating a unified model to play them all, similar to previous publications. Additionally, I plan to extend this to board games, enabling the creation of customized agents. Some rely on well-known planning algorithms like Monte Carlo Tree Search, while others can clone the behavior of famous players. This requires a smart storage solution to index and serve all the games, which is a fun engineering challenge nonetheless. Stay tuned!
https://github.com/shehio/rl
Scroll down for the videos!