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

Re: computer-go: gnu-go agreement protocol



Yes, the  Tromp/Taylor   rules are  very  much   based on  all    your
considerations.  It turns  out that after much  discussion  of all the
early  agreement ideas, something very simple  is  possible that is in
the same spirit as Tromp/Taylor  as an extension.  More importantly to
me,  it is purely optional,  no program is  required to use it or even
know about it.   

You make a good point  about the motivation for  playing it out to the
end, it's not a big deal with  computer vs computer.  If this protocol
wasn't trivial to implement, I would  definitely not even consider it.
I am going to implement this protocol in my  program and see if anyone
else does too.

I'm also trying  to look a  little farther ahead than just considering
this a dumb  "anal" commputer/computer mechanized interface.   I think
this ruleset  has real value   for  human players  too, especially  Go
beginners.

So I envision the early agreement protocol as implementable in a human
GUI  type of interface too.  How  it might work  is that the human (or
computer) can  mark dead stones (with a  mouse  for instance), and the
interface will  convert  this to a  score  using the  exact  agreement
protocol we have been trying to define.  

There is no reason that  something like this can't  be done in  purely
human games on a  physical board too, with  agreed upon shortcuts that
are compatible with Tromp/Taylor scoring.

Don









   Date: Wed, 4 Jul 2001 23:34:04 -0700 (PDT)
   From: Sai To Wang <saitowang_go@xxxxxxxxxxxxxxxxx>

   Hi everyone,

   IMO, in a computer vs computer game, if there is no
   human involve in the scoring, the Japanese scoring
   rule is not adequate because of the current computer
   program strength.
   Is the same for the “two consecutive passes” kind of
   end game protocol.

   What we are doing is designing a protocol that need to
   meet some requirement like:
   low complexity
   play through network like Internet and LAN
   accurately determine the winner or draw
   force both player to follow the rules that they
   agreed. (not the end game part)
   ...

   Other art kind of thing are mostly not necessary.

   For me, the major reason to make a development like
   this is for program to easily play to other program
   with minimum human supervision, gathering as much
   feedback as possible, so that the program’s strengths
   and weaknesses can be verify much easier and with more
   accuracy. That’s one way to help to make a stronger
   program.

   So, the scoring part of the protocol can be like this:
   if one player resign, he losses.
   always play until no legal move possible OR the only
   legal move is to fill in one of the last two eyes of
   any group except seki. (the arbiter may give the stop
   signal ?)
   use Chinese scoring.

   We all know that the use of Japanese scoring rules
   require much more skill. And the possibility of
   disagreement in winning or losing judgment between
   Japanese and Chinese scoring is negligible for the
   current program strength.

   Is true that make the game end some 100 play before
   the everything is fill up make the game prettier and
   cleaner for human to read, but only the person want to
   read to the end.

   It is too early to say that we are developing a
   program to play “Like” human, is more likely that we
   need a program to play decent GO first.

   Sai To

   __________________________________________________
   Do You Yahoo!?
   Get personalized email addresses from Yahoo! Mail
   http://personal.mail.yahoo.com/