Welcome

[Rémi Coulom]

Welcome to the Game AI Forum.

Some specialized forums and mailing lists have played an important role in the development of artificial intelligence for games such as chess or Go. This forum aims to offer a similar service, but with a more general scope. The discussion of any game is welcome here.

If the game you are interested in already has a forum or a mailing list, you might prefer to use it instead:

• Chess: http://www.talkchess.com/forum/viewforum.php?f=7
• Arimaa: http://arimaa.com/arimaa/forum/cgi/YaBB.cgi
• TORCS: http://www.berniw.org/trb/forum/threadlist.php

A collection of other essential links:

• Ludii: https://ludii.games/
• Chess Programming Wiki: https://www.chessprogramming.org/Main_Page
• ICGA: http://www.icga.org/ (International Computer Games Association)
• Computer Olympiad: http://www.game-ai-forum.org/icga-tourn ... n.php?id=3
• SGF: http://www.red-bean.com/sgf/ (Smart Game Format)
• GGSA: http://www.cs.ualberta.ca/%7Emburo/ggsa/ (Generic Game Server)
• Zillions of Games: http://www.zillions-of-games.com/
• Boardspace: http://www.boardspace.net/ (Play plenty of board games online)
• Little Golem: http://littlegolem.net/jsp/main/

Please reply to this message if you would like to add to these lists.

Rémi

[vintermann]

Thanks for setting up this forum. I am myself trying to write a Hex program based on UCT/MC/Bandit-techniques, and it will be great to have someone to exchange experiences with. I never felt it was quite appropriate in the computer go mailing list...

I'm only implementing the game logic yet, but I have thought about it for a while: Since pieces never move in Hex, I think you can make some assumptions in playouts which speeds things up considerably. For instance, once you have a sequence of moves which is a win for vertical, all reorderings of these moves also constitute winning sequences for vertical. Using the "all moves as first" heuristic therefore seems to be on much more solid ground than in Go.

I have wondered if it might not even be possible to calculate what the playouts would converge at, without doing the playouts at all.