This is a terminal-based AI test program in Python3 for an extended tic-tac-toe game. Check this link for details and other implementations.
A total of 2 rule sets are supported in this program, according to this statement in the original post.
What if one of the small boards results in a tie? I recommend that the board counts for neither X nor O. But, if you feel like a crazy variant, you could agree before the game to count a tied board for both X and O.
In this program, the rule set recommended in the post is refered as normal, while the other is refered as bizarre.
To enhance the unsatisfactory performance (e.g. competance, computational cost) of the program for now, the following tasks are still in progress
- Train proper models with better performance for both rule sets.
- Optimize the algorithm for MCTS.
- Clean up the code.
- Implement state-evaluation function/network.
- Utilize dynamic computational cost.