I have been a strong advocate of dynamic komi in the past, but am starting to 
wonder if it is something of a kludge to fix some deeper problems. I think 
these 
are the main issues nowadays:

1) winrate does not differentiate (much) between "I totally own you" versus "I 
can win by 0.5 points" 
2) winrate is not reliable in certain kinds of positions, especially those 
involving large semeai. When I watch strong humans, they quickly reply to the 
removal of their liberties by taking the opponent's liberty, in order to 
maintain the status of a semeai. On a complicated board, programs seem to not 
partition the board in the same way that humans do. 
3) programs do not yet know much about invasions of large areas; they often 
play 
as if sketching a very loose fence around the center or moyo were enough to own 
that area.

When the question of semeais is resolved, it may then make sense to set the 
dynamic komi to the level of "how far ahead I am" - so that, as one capitalizes 
on mistakes, the "dynamic komi" ( which perhaps might be re-named "advantage" 
should ratchet upward; and if one makes mistakes, the "advantage" should slide 
back down. If a program is given nine stones, it should know that its advantage 
is, at the moment, 90 points - and should attempt to maintain that; if the 
advantage dwindles to only 10 points, the program may be in deeper trouble than 
it knows. 

At the moment, multiple semeais seem to be a big issue. 
Terry McIntyre <[email protected]>


Unix/Linux Systems Administration
Taking time to do it right saves having to do it twice.


      
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to