In a project for college I created a simple turn based game, with up to 4 players that can either move or attack the opponents. The players are playing over the network, meaning the clients are supposed to be programmed AIs. The client itself is fully functional, meaning it has all the game logic and can simulate complete games.
Now my task is to create a RL-Agent with a Deeq-Q-Network that learns to play the game. However, I don't really find any source to how that should be done. I was able to create an Agent with a DQN for the CartPole environment of OpenAI gym with PyTorch. Now my guess would be to create my own environment with the gym framework, but since the game itself is already implemented I was thinking if it was possible to feed data in the DQN without having to create the gym environment. As a state it would get the gamestate (which is a 2d grid, with information about the players and their remaining hitpoints) and all the possible moves in the current state as the action space. And since the game is network based, it would save the network after each game and reload it when the next starts during the training. For the training I would start the games over and over with a script and let it train for a while.
As I'm quite new to Machine Learning it seems really blurry as of how to tackle this problem and was hoping to get led in some direction on how to start.
