[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go:life and death



At 01:16 PM 5/29/2001 -0700, you wrote:
>>>>> "DF" == David Fotland <fotland@xxxxxxxxxxxxxxxxx> writes:

    DF> If you want results in a full playing program, speed is
    DF> important.  And the hard-coded life/death
    DF> reading algorithms get better results than the rule based
    DF> systems.  One problem with rule-based sytems
    DF> is that they are brittle.  Their assertions are either true or
    DF> false.  I've found that to play well, the system
    DF> has to use confidence estimates, since very few of the static
    DF> evaluations are completely certain.

Isn't this a good argument for something like B* search, which was
built to take advantage of these kinds of estimates? How well do you
think this avoid the brittleness problem you speak of, David?
Yes B* search might do better. I don't have a strong opinion on B* search
since it's been a long time since I studied it, and I never coded or used it.
I try to model the way I think about go in my program. When I look at groups,
sometimes I've very confident of my static evaluation, and sometimes I'm
not sure so I do some more reading. I think a good static life/death evaluation
must include some kind of confidence measure.


-Patrick Bridges

--
*** Patrick G. Bridges                  bridges@xxxxxxxxxxxxxxxxx ***
***                #include <std/disclaimer.h>                 ***
David Fotland