>> I think you are too much of "chess player" :-) >> The fact that he is 0.5 point in the lead does not imply he is >> (much) stronger. >> > I didn't say that. My point is that if he is beating you then he is > not likely to be a lot weaker than you, and so he is probably just as > aware as you are about the score and thus you can't assume he is likely > to make an error. Therefore this algorithm would have it's best > chance of working with an opponent that really is much weaker, however > it's not likely you would be in this situation against a much weaker > opponent. So you are designing an algorithm that might help you in a > case that is not very likely to happen. > > I base that logic on my observations that once the score goes below 10% for Lazarus, it is losing. It's extremely rare that it salvages a game once the score goes below even 20%.
I have seen it happen and I had to gather my composure only to see that the opponent lost on time or by illegal move. On a VERY RARE occasion it has happened with seki, where the other program didn't understand seki and allowed Lazarus to save the day unexpectedly. But Lazarus didn't play stupid because there was still a few percent chance of winning. Your fear is unjustified that it will start giving up prematurely because if there is a little hope left in the position, a monte carlo program is not going to give away huge parcels of territory just for laughs, it's going to do what it needs to do to maintain that hope. I would remind you that often this will involve not letting go of territory foolishly. It only begins to let go "foolishly" when even the random lines of play all lead to lost games. - Don >> Any player, in particular a human player, is capable >> of the making a mistake. So it is important to stay on the 'small' >> losing line. That might a difference to chess, where there is no >> 'small' loss. >> > I guess we disagree on how important this is. To me it's only important > as a tie-break, if all else is equal. My feeling is that it more > important to find the line that gives you the best winning chances and I > would only resort to some other method if the score was zero (or so > close to it that it was statistically insignificant.) The alternative > you are proposing is to systematically distrust the programs calculation. > > Near the part of the game where the score is going to zero as the > program loses, often there is are 1 or 2 lines of play that show an > occasional win, even though the vast majority of play-outs show a > loss. But most lines of play lose in EVERY SINGLE play-out. I > could care less that it doesn't lose big because if a random player can > beat me with it then why should I expect that my opponent cannot handle > it? Wouldn't it be better to choose a line that was at least good > enough that random play was not enough to guarantee the win? > > - Don > > >>> So at best you hope your opponent will make a stupid mistake in an >>> obviously lost position for you. >>> >> No, the opposite. Not a stupid mistake; I am hoping for the subtle >> mistake. But you throw that opportunity away If you play "desparate" >> moves just because you think you will lose the game by 0.5 points. >> >> >>> There is nothing wrong with this, if it's what you want to lose sleep >>> over, but how much do you expect to gain from it? I see people >>> getting excited about this idea as if it's the holy grail of computer go >>> and will add 50 ELO or more. >>> >> Nobody called this the "holy grail" ... but I agree with you that >> there are bigger problems in computer Go ... >> >> Christoph >> _______________________________________________ >> computer-go mailing list >> computer-go@computer-go.org >> http://www.computer-go.org/mailman/listinfo/computer-go/ >> >> > _______________________________________________ > computer-go mailing list > computer-go@computer-go.org > http://www.computer-go.org/mailman/listinfo/computer-go/ > > _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
