Hex game bot powered by reinforcement learning.
The envrionment is based on FirefoxMetzger's implementation with a few tweaks:
- separated classes into multiple files
- implemented the random policy inside an agent class
- improved the performance of the random agent by caching some arrays
- customizable reward function
- RGB rendering
Observation space is described by: (board, player) = (NxN matrix, {0, 1})
Action space is described by: (x, y) where it is any position on the board.
Note: There is no support for the swap rule at the moment.