> -----Original Message----- > From: Don Dailey <[EMAIL PROTECTED]> > To: computer-go <computer-go@computer-go.org> > Sent: Mon, 11 Aug 2008 11:09 pm > Subject: Re: [computer-go] Re: What's happening at the European Go Congress?
> On Tue, 2008-08-12 at 11:50 +0900, Darren Cook wrote: > > > Also, if you are down 8 or 9 stones, maximizing your winning chances is > > > still the right strategy, right? > > > > With MCTS algorithms the error margin is high at the start of the game, > > and low in the endgame. In a handicap game against a stronger opponent > > the assumption is that the weaker player will make more mistakes (i.e. > > has a higher error margin overall). But MCTS programs don't see it that > > way - their opponent model is the same strength as they are. So they > > choose a move that gives them 95% (+/- 20%) win (against themselves) > > instead of the better move that they only gives them a 90% (+/- 20%) win > > (against themselves). (I.e. I'm saying their error margin in the opening > > is much greater than the difference in their estimate of move values.) > There could be something to that. > Do you believe that they will play the 90% move if they are told they > are not really down 9 stones? > I did a bunch of experiments and ALWAYS got a reduced wins when I faked > the komi. But there are a million ways to do this and I may not have > stumbled on the right way. If my engine plays in a high handicap game (and it has to be a pretty high handicap), for the first moves, it can't see a difference for any moves and plays randomly. I can fix this by making the playout asymetrical. I make the playout moves for black lighter (higher probability of being random). With this adjustment, it makes reasonable looking moves. I haven't tested this extensively because I don't have any need for an engine that plays better in high handicap games. - Dave Hillis
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/