When I've looked at these losses they were due either to bugs, or to bias in
the playouts, for example when there is a semeai.  The program will think it
has 80% win rate when it is actually already behind.

David

-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Petr Baudis
Sent: Monday, December 21, 2009 6:44 AM
To: computer-go
Subject: Re: [computer-go] one more look at the scoring function

On Mon, Dec 21, 2009 at 01:57:16PM +0200, Petri Pitkanen wrote:
> They also all lose games on endgame same manner. Having won a game by 30
pts
> they start giving away those points for - sometimes - imaginary safety,
> allowing other player to come within striking distance. Some sort dynamic
> komi would be nice  in endgame, but would probably not work.

  If this leads to a loss, I think that should happen only when there is
some consistent bias in the tree that misevaluates some part of the
position, and that imposing the komi *might* in some specific cases
patch up the deficiency, but probably won't have big impact in general
case. This is good experimentation subject though.

                                Petr "Pasky" Baudis
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to