On 4/4/07, Sylvain Gelly <[EMAIL PROTECTED]> wrote:
> However, it would be accurate to play a few thousand games at 7.5 komi,
> then a few thousand and 6.5 and compare the white/black win percentage. The
> one closest to 50% would be the one to use.
If some are interested by the results, I just did a test.
9x9 games, MoGo with 3k simulations per move against itself. Of course
this player is much weaker than the one playing on KGS, but at least
it enables me to make thousands games in some hours.
Here are the results (wins for black):
komi 6.5:
1943/3640 (53%)
komi 7.5
1754/3640 (48%)
Traditionally the weaker player plays black and it seems reasonable
that starting the game should not be a disadvantage. Consequently, if
these statistics are at all admissible, my choice would be for 6.5
komi.
Sylvain, could you run the same test on 7x7 to verify that there the
'correct' komi would be 9 (try 8.5 vs 9.5)? If MoGo wouldn't converge
to 9 we probably shouldn't have much confidence in the generalisation
of the above results for higher levels of play on 9x9 (or you could be
on your way to discover an error in the human solution for 7x7 :-).
Either way I think it would be interesting to know what comes out...
Erik
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
