Closed tindiz closed 2 weeks ago
Thanks for the kind words!
There's a self-play command (documented in the README) that that can be customized for # of players and conventions. The performance fluctutates quite a bit since there are still a lot of bugs, but the bot is currently best at playing with H-Group conventions at level 7 or 11 with 3-4 players, usually scoring an average of 23-24 pts. It's noticeably worse at other player counts due to lack of tools (2p) and poor navigation (5p, 6p) but can still get > 20 pts on average. The Playful Sieve convention set is designed for only 2p, but I haven't worked on it as much and it currently performs slightly worse than H-Group.
Since this is a rule-based bot with no learning, I can't imagine cross-play going well. I have a basic mechanism in place for letting others make "mistakes", but the bot pretty much assumes everyone is following the same conventions it is.
That's cool! Thanks for that!
Very interesting and cool project!
I was wondering if you have done evaluation for self play and/or cross play? Thought it would be interesting to see scores over
n
games for different conventions in self/cross play.