During the last few days I have been meditating a lot about the questiion whether taking into account the margin of win into MCTS (UCT) may help or hurt.
I do not have a go program by my own, so for the moment I have to believe what programmers are saying, namely that "MCTS with margin of win as a criterion" gives worse performance in play against other computer programs. So, a negative answer... but there may be other points of view from which this phenomenon is not bad but helpful. To give an example: I am active in the scene of "computer-aided game inventing". When a human designer of board games has invented a new game he typically needs many rounds of human test players who find strengths and weaknesses of the game. As it is really a hard job to find enough test players our idea is to let the early phases of test play be done by computer in autoplay mode. (The list of basic criteria for good games include appropriate game length, balance of chances, low drawing rates.) When you now think of a game where the overall goal is to win and to avoid losing, but where also a margin of win/loss can be counted at the end (Go belongs to this class) you may pit two versions of your MCTS algorithm against each other: * MCTS(WinLoss), playing only for a win, independently of the margin * MCTS (margin), playing for win/loss, but also considering the margin. with the philosophy: The game is "interesting", when MCTS(WinLoss) performs better than MCTS(margin), and "boring" when it is the other way. Ingo. PS: Here is an early report on computer-aided game inventing. http://www.minet.uni-jena.de/preprints/althoefer_03/CAGI.pdf -- GMX startet ShortView.de. Hier findest Du Leute mit Deinen Interessen! Jetzt dabei sein: http://www.shortview.de/[EMAIL PROTECTED] _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
