[computer-go] Anchor Player

Jacques Basaldúa jacques at dybot.com
Thu Dec 21 12:04:50 PST 2006


Hideki Kato wrote:
> Increasing KOMI is much easier than placing stones, right?

Stuart A. Yeates wrote:
> Increasing komi is much easier than placing stores, but a 
> much weaker representation of how go games are actually 
> played in the real world.

A very huge komi >30 points, apparently solves the database
lookup problem, but that's not really true. The josekis that
are useful for equal play are way too slow if you are 30 
points below.

Playing handicap or unbalanced komi correctly requires taking 
risks. You *have* to underestimate your opponent. If you 
don't, if he is as strong as you are, you won't catch up
30 points unless he blunders.

Handicap play is a *different* problem. Surely fascinating 
and some authors may love to deal with it, but some others 
may think that creating a balanced engine (one which expects 
its opponent to play with its own strength) is hard enough.

Pairing programs with a difference of more than 3 or 4 kyu
should be done very infrequently. In this range, winning
probabilities should not exceed 80% or so. I think the 
most interesting comparison is between similar kyu programs.
Knowing that A beats B 3 times of 4 is precise enough. 

Jacques.



More information about the computer-go mailing list