Hi!
On Tue, Jul 05, 2011 at 07:48:14AM +0200, Leon Matoh wrote:
> So to show the concept I played two games on 13x13 between
>
> 1) standard pachi
>
> 2) and modified pachi with random komi from fixed window
> [-17.5,32.5] for all game
> board.c 1404 > return (32.5 - rand() % 51) + board->handicap +
> scores[S_WHITE] - scores[S_BLACK];
> meaning that for EACH quick count I aplliyed komi as a random
> number from window
FYI, I have done this (plus adding a komi, dynamic komi turned off)
and tested over a sample of 5000 games on 9x9. Taking white with komi
-7.5, the original version beats GNUGo 59.6%+-1.4%. With your
modifications, it beats GNUGo in 36.3%+-1.2% of cases.
So the performance is significantly reduced. Maybe it is possible to
implement some useful evaluation based on your approach, but I'm afraid
that you probably haven't presented very convincing case so far and you
will have to do most of the work yourself to prove it's a viable
direction - noone of us has shortage of interesting ideas to try
on their own time. :-)
I'll be happy to help you out with coding issues if you choose to
modify Pachi and offer you some computing power for play-testing, but
you will have to get your hands dirty too.
--
Petr "Pasky" Baudis
UNIX is user friendly, it's just picky about who its friends are.
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
