On Thu, 2007-04-05 at 09:15 +0900, Darren Cook wrote: > This is an interesting result, on one of my favourite topics. I've > spent > a lot of time analyzing 9x9 games by very strong/pro players in detail > and decided correct komi is probably either 6 or 7 points. Give me a > few > more years and I may be able to narrow that down :-).
It seems likely that it's 7 when all tests show 6.5 wins for black and 7.5 win for white. But it's not a proof of course. With weak programs the "correct" komi must be lower or white wins too much. But the fact that as you test stronger programs, the komi needs to be higher is an indication that 7 might be a lower bound, perhaps the correct value. 6 is a value that works with weaker programs, not stronger ones. With random play I believe 3.5 komi works out to be closest to fair. So my vote is that 7 is a better guess than 6. It's possible that the value is neither 6 or 7 but 5 or 8 but it's extremely difficult to defend for one side or the other. It's kind of like that in chess. Many people believe the opening position is a draw, but it's certainly quite a bit more difficult for black. So the correct komi in GO may still give one side a more difficult position to defend, even if it's correct. There is some evidence from these experiments that 7 is correct, but it's more difficult for white that black (because 6.5 is easier to win for black than 7.5 is to win for white.) - Don _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
