Re: computer-go: Computer Go Tournament Program

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

   From: Daniel Hallmark <dgh@xxxxxxxxxxxxxxxxx>
   Date: Thu, 05 Jul 2001 10:53:53 -0500

   On 4 July 2001, Don Dailey wrote:

   > My slight preference is to include the pair score simply because it is
   > a more complete report of how the game ended.  Another side of me says
   > a single score seems more in the mathmatical spirit of simplicity that
   > Tromp/Taylor  is based  upon.   What do you think?    I could go  with
   > either.
   > 
   > Having just  said that, I think I  like the single score report better
   > now!  There is  no reason a  more complete  breakdown cannot be  added
   > after  the game by  the  arbiter  software  but  need not involve  the
   > protocol.


   If the arbiter software is going to maintain the "official" game record
   and score report I like the idea of it being able to provide additional
   information after the game.  In this spirit, would there be any interest
   in adding an optional feature of the protocol to allow programs to divulge
   their own evaluation of group/string/area status (black/white-owned, live,
   dead, seki, etc.).  A single point within each group paired with the 
   status would be sufficient.

   I understand this is completely unnecessary for the purpose of resolving
   tournament game results, but it could be offered as a completely optional
   portion of the protocol to provide additional information in the final
   game record.

   Daniel Hallmark



Yes, that's an excellent  idea but is more an  interface issue, not  a
rules issue.   I know  we  are mixing  the two  up  a little in  these
discussions,  probably my fault  because I am  looking at this from an
implementation point of view, not a purely theoretical point of view.

To go along with what you are suggesting, I have just  taken a look at
the GTP communication  protocol  that  gnugo  has  implemented.   They
actually have some commands that support this kind of thing.

I also took a little time to add this protocol to my Go program.  It's
quite easy to use  and implement.  I  tested my implementation using a
perl script in  the gnugo distribution  called "twogtp",  it let's two
GTP programs play against each  other.  When both  sides pass, it asks
each program to report their notion of the final score.

I am going to make a version of this for Tromp/Taylor and add the code
to  the perl script to  actually do the scoring.  I   may also add the
agreement protocol later.  I'll pass the script back to the gnugo team
(if they want it) when I am done.

I will probably also do a GUI for go using GTP and perl/TK.  I already
have a GUI in C for  xwindows (using xforms)  but this is probably not
very compatible with windows stuff, and I want it to be.

The GTP communication protocol is  still not standardized according to
the  documentation I  read, they  are  apparently  waiting to see  how
things progress.   


Don