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

Re: computer-go: Computer Go Tournament Program




   Date: Mon, 2 Jul 2001 14:34:31 -0500
   From: William Harold Newman <william.newman@xxxxxxxxxxxxxxxxx>

   On Mon, Jul 02, 2001 at 02:58:48PM -0400, Don Dailey wrote:
   > 
   > What do you think of this, based on John Tromp's idea:
   > 
   >  . After a player makes a pass, it is the other players turn to move.
   > 
   >  . A pass move must include a players notion of the score.  A valid
   >    score to report is "undefined." 
   > 
   >  . The game is over when 2 passes in a row occur with the same
   >    agreed upon score OR no legal moves are available for one side.  
   > 
   >  . There is no limit to the number of pass moves a player may make.
   > 
   >  . If a game ends without an agreed upon score,  an arbiter scores
   >    according to a strict Tromp/Taylor end of game scoring.
   > 
   > 
   > Feel free to improve this, but did I  capture your suggestion properly
   > here?
   > 
   > 
   > Don  
   > 
   > 
   >    Date: Mon, 02 Jul 2001 20:09:06 +0200
   >    From: Robert Jasiek <jasiek@xxxxxxxxxxxxxxxxx>
   > 
   >    Don Dailey wrote:
   >    > then an   interface  could  use  the Tromp/Taylor agreement phase
   > 
   >    Before implementing some agreement phases their maximal
   >    number during one game and the exact meaning of passes
   >    should be specified. The maximal number could be one or
   >    infinity, depending on the wording.
   > 
   >    --
   >    robert jasiek

   The ruleset proposed above seems to allow the match to last forever
   when players disagree on the score: I pass, he passes, I pass, he
   passes, I pass, and so forth. A sufficiently smart program can get out
   of this deadlock by exhausting all the legal moves before it starts to
   pass, but that sorta defeats the purpose of the negotiation phase. 


Good observation, my protocol  is broken then.   In that case, I can't
think of  anything simpler  than the  original  extension proposed  by
Tromp/Taylor rules,  4 consecutive  passes  and it's over  with strict
scoring.


                                                                       Or
   if it passes a few times and the game doesn't terminate, it could set
   about exhausting legal moves before it passed again; but that behavior
   seems more unlike traditional human Go than anything that's been
   proposed so far.

   It seems natural for a disagreement between programs to cause play to
   continue to the point where Tromp-Taylor scoring works, but to require
   the programs to continue beyond that, to the point where there are no
   legal moves left, seems pointless. Therefore IMHO it would work better
   to have some number of consecutive pass moves (3, 4, whatever) suffice
   to end the game regardless of whether there's agreement on the score.

Agreed.

Don




   -- 
   William Harold Newman <william.newman@xxxxxxxxxxxxxxxxx>
   As usual, this being a 1.3.x release, I haven't even compiled this
   kernel yet.  So if it works, you should be doubly impressed. 
     -- Linus Torvalds, announcing kernel 1.3.3 on the linux-kernel mailing list.
   PGP key fingerprint 85 CE 1C BA 79 8D 51 8C  B9 25 FB EE E0 C3 E5 7C