[computer-go] Experiments with UCT

[steve uurtamo]

> That's why I believe it would be really
> useful to recognize
> and ignore these moves.   I just haven't figured out
> how.

i have a question for everyone that is tangentially
related.

in yose, it seems like the following set of steps
is mandatory to maintain a lead or to make up a
small difference in lead:

i)   calculate the longest sequence of sente moves
     that improve your score, summing them along
     the way
ii)  calculate the largest gote move that improves
     your score
iii) do i) and ii) from your opponent's point of
     view.

look at these sums:

my sente + my gote - opponent's sente - opponent's
gote
my gote - opponent's sente - opponent's gote + my
sente
my gote - opponent's gote + my sente - opponent's
sente
... etc.

pick the sequence that's maximal.

where the "+/-" sign is overloaded to mean that all
calculations after it are taken from the final move
of the argument before the "+/-" sign.

i guess the short question is whether or not there
is a good concept of sente and gote in anyone's
code once yose is reached?  sometime after joseki
sequences are exhausted, it seems that opponents
will be much more likely to understand tennuki than
your code, but toward yose*, people's code should
be much, much better at evaluating these sequences
(since there are many fewer and the combinations
aren't brutal).

* here i'm sloppily defining yose as being when
  all life and death problems have effectively been
  completed.

s.

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com