[computer-go] Komi for 9x9 computers

[Don Dailey]

That still wouldn't settle the issue for sure.   I know for a fact
that some of the strongest programs on the server play for a win
based on what komi value is used.   In other words, if komi was
6.5 instead of 7.5 they would play differently,  a bit "safer" as 
black and more agressively as white.   The goal of these programs
is to win and they don't consider the win margin, only the win
probability.   

My own program often gives up large blocks of stones (that it can easily
defend) if it doesn't need them to win.    It's very common for Lazarus
to win by 1 stone against far weaker opponents when it could very easily
demolish those same opponents.

The only way to get accurate statistics is to change to 6.5 komi on the
server to judge it.  In other words,  Lazarus, Viking, AnchorMan
and probably other programs would lose a lot of games if I recalculated
the results of all the games based on 6.5 komi.    These wouldn't just
be white losses and it's hard to say if they would balance properly for
calculating the value of a different komi.

For komi oblivious programs, you could simply recalculate the results
of the games assuming different komi values and see what happens.   I
could probably do this for matches between komi oblivious programs but
I don't know which one they are.

Does GnuGo make different moves during the game based on the komi?


- Don


 
On Tue, 2006-05-02 at 21:47 +0200, Chaslot G (CS) wrote:
> > I would be curious to see what 6.5 komi does too.
> 
> If you have all the results in your logs, it should be possible to make
> the same statistics on komi of 7 and 6.5.
> 
> Why not having draws sometimes?
> 
> Guillaume
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/