gnugo-devel
[Top][All Lists]
Advanced

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

Re[2]: [gnugo-devel] GnuGo playing on WING (fwd)


From: Arend Bayer
Subject: Re[2]: [gnugo-devel] GnuGo playing on WING (fwd)
Date: Sun, 10 Feb 2002 16:52:53 +0100 (CET)

I got this reply from Shigeru Mabuchi. Has it been noticed before that
GNU Go is stronger on smaller board sizes? (I never tried to compare.)

I asked him in reply if they could sent us the sgf-file if they observe
a wrong status decision by GNU Go at the end of the game.

Arend


---------- Forwarded message ----------
Date: Sun, 10 Feb 2002 21:05:30 +0900
From: Shigeru Mabuchi <address@hidden>
To: Arend Bayer <address@hidden>
Subject: Re[2]: [gnugo-devel] GnuGo playing on WING

Dear Arend,

>I recently visited your server and noticed that GNU Go 3.1.24 playing as
>robot3 has considerably lower ranking (about two and a half grade)
>than GNU Go 3.1.19 playing as robot1 and robot2. 
>
>Of course we would like to know whether this is an indication of steps
>backward in our GNU Go development. Are robot1/2 and robot3 running
>on similar hardware, so that we can assume this difference in strength 
>is not due to time problems?

  They are not running on the same hardware. However, their rating
varies from time to time, between 14k* and 16k*. I don't think
there is a remarkable difference between their strength.
There is a large difference when they in 19x19 and in 9x9.
Apparently GnuGo is at least 2kyu or 3kyu stronger when playing
on a small board.

  The current rating of robot3 was due to the 'cheating' of a
user who has played many games with robot3. He cheated at end-game
by changing the status of the computer stones to dead stones.
I have adjusted the rating of robot3 and meanwhile warned the
player who has cheated (lowered his rating as well).

  I plan to add a self-check mechanism to the server, so that any
end-game correction by the opponent of gnugo would be checked
by our support staff automatically. I believe that this will
act as a brake to the cheating and might pick up real examples
of end-game that gnugo cannot judge correctly (I guess this
should be a very rare case).

  Thank you for your remark and we all appreciate greatly your
contribution to the world of computer-go.

Best regards,
Shigeru Mabuchi

======= W ======= I ======= N ======= G ======
       World-wide  InterNet  Gokaisho        
==============================================
Shigeru Mabuchi  (mab on WING)
http://www.wing.gr.jp/
WING Server Address:  wing.gr.jp 1515




reply via email to

[Prev in Thread] Current Thread [Next in Thread]