> You have basically 2 cases when losing. One case is that the program > really is busted and is in a dead lost position. The other case is > that the program THINKS it's lost but really is winning (or at least has > excellent chances.) In the first case, we are simply playing for a > miracle, hoping that if we lower our expectations the opponent will > falter. This is at best a speculative and extremely minor > enhancement if we can even make it work and I argue that it also imposes > risks of it's own. Unless the score is actually zero in every line, > the program is going to play for any outside chance of a win, no matter > how small. Telling it that a big loss is good enough will cause it to > play safer, not more aggressively like you think. (Yes, you can > hand pick specific positions where you can "trick" it into doing the > right thing, but that is "cooking the books", it's not a real solution.)
There's a subcase of the first case, and that's either handicap games, or 9*9 with really strong and not the right komi. Then the program really IS busted, but practically can win against its weaker opponent if he plays for having best score in worst case. In fact, I first thought of handicap games when I saw these komi adjustments. One reason for it to be maybe more efficient in practice is the following: the whole aim of the evaluation function is discriminating between moves. If you really are in the tails of the outcomes of the MC, all moves may look too similar. Slightly moving your point 0 may make more easy to distinguish one move from the other. Problem: the move that gets the highest boost must also be the one that was the strongest originally (I do not take into account nakade and such, where it's definitely worth having programs fight for points). It's reasonable that the most heavy tail comes with the move whose last peak is nearer zero, but that's certainy not always the case, and that I cannot imagine how to tune. Jonas _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
