Re: Advice on evaluation

[David Fotland <fotland@xxxxxxxxxxxxxxx>]

Many Faces does exactly the same thing.  One of the group
evaluations is Unsettled (lives or dies in one move).  At terminal
nodes in the evaluation it tries kill or save moves on these groups
on the whole board to get a stable evaluation.

The issue was with groups that are not killable or savable in one
move.  Instead, there are running or surorunding moves or vague,
big eye moves.  A single ply of lookahead for either side still
leaves the groups unstable, just stonger or weaker.

If group strength 0 is dead, and group strength 100 is alive, I want
to distingush between 

a) group strength 50, my move gets to strength 100, enemy move gets to
strength 0

and

b) group strength 50, my move gets to strength 55, enemy move gets to
strength 45.

Moving to attack or defend group A should be bigger than attacking group
B.  The value of attacking or defending group A is proportional to the
size of the group.  Attacking group B is more likely worth some constant
value, no matter how big it is (and may be worth less if the group is
bigger).

David

At 11:16 AM 5/17/99 -0700, Eric Pettersen wrote:

>My program (gottaGo) already kind of does this.  One of the life/death  
>values it can assign to a group is "killable"; this value means that the  
>group can be killed if the opponent attacks it before a defending move is  
>played.  Conversely, the group is alive is a defending move is made before  
>the group is attacked.
>
>When gottaGo (a fixed-depth search program) reaches a terminal node for  
>evaluation, it notes what groups are killable.  It then "auctions" the  
>groups on the list off; it assumes that the side to move will kill/save the  
>largest value group, the other side will then kill/save the next largest  
>value group, etc. until all killable groups are taken.
>---
>			Eric Pettersen
>			pett@xxxxxxxxxxxxxxxxx (NeXTmail capable)
>			http://www.cgl.ucsf.edu/home/pett
>
>