framework for training RL agents to play (simple, finite) games
Provides gamer package, which consists of the following sub-packages:
- games: finite n-player games
- players: players of a given
gamer.gamesgame, with policy deciding moves given the game state - strategists: agents designed to optimize strategy for a given game; trains by producing players and having them play games
The main file is run from the project's root directory via the following command
python src/main.py
A incomplete list is below:
- Python 3.3 and up
- reasonably up-to-date
npm - Python packages: numpy, scipy, chess (can be installed with
npm i numpy scipy chess)