[Computer-go] gnugoquad

5 views
Skip to first unread message

Don Dailey

unread,
Apr 25, 2012, 10:00:50 AM4/25/12
to computer-go
There has been an issue with gnugoquad not cleaning up the games and thus distorting the ratings and quality of the matches.  

In order to prevent that I have changed the password of the gnugoquad account,  which means it can not longer be used.    That is the only way I have to disable an account and of course I cannot prevent the owner from logging on with another name - but I don't think that is an issue and I don't think it was malicious.  

Anyway,  this is a heads up to the owner of that account,   just to let you know that this was not done maliciously but only because we did not have any way to contact you (whoever you are) in order to help you fix this issue.       If you want to use those names again (after fixing the bot) just let me know and I can reset the name. 

Don



Don Dailey

unread,
Apr 25, 2012, 10:08:12 AM4/25/12
to computer-go
This is a P.S. to my previous comments.

I had to reset the server in order to force gnugoquad off,   so anyone running bots may want to check for a graceful restart as it comes  back up (within 5 minutes or so.)  
Don

Joshua Shriver

unread,
Apr 25, 2012, 10:23:47 AM4/25/12
to compu...@dvandva.org
That was me, sorry didn't know it was having problems. It's just stock gnugo was running it mostly as test machine for others to play with and to fill some slots on 9, 13, 19.

I'll take it offline.
-Josh

On Wed, Apr 25, 2012 at 10:00 AM, Don Dailey <daile...@gmail.com> wrote:
_______________________________________________
Computer-go mailing list
Compu...@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Don Dailey

unread,
Apr 25, 2012, 10:38:03 AM4/25/12
to compu...@dvandva.org
It's off-line now anyway because I changed the password.    Of course you can set it up again with the correct command line arguments for cleaning up if you want to get it back on line.       

Do you want me to change the passwords back?    If so, send me the name of the password you want by private email.  But I think it's just better to start with a new identify.  

Don

Hiroshi Yamashita

unread,
Apr 25, 2012, 3:28:23 PM4/25/12
to compu...@dvandva.org
Hi,

I think "Stop-cur" has similar problem.
Its playing style is clearly stronger than random "Brown-1.0fvh".
But Stop-cur allows all suicide moves.

So no resigning programs can easily beat him.
"Brown-1.0fvh" does not resign, so its rating is higher.
Weak resign-able MC bots lose to him.

Regards,
Hiroshi Yamashita
--------------------------------------------------------------------------------

folkert

unread,
Apr 28, 2012, 8:07:51 AM4/28/12
to compu...@dvandva.org
>> There has been an issue with gnugoquad not cleaning up the games and thus
>> distorting the ratings and quality of the matches.
...
> I think "Stop-cur" has similar problem.
> Its playing style is clearly stronger than random "Brown-1.0fvh".
> But Stop-cur allows all suicide moves.
>
> So no resigning programs can easily beat him.

So if understood this correctly: if stop "sees" that the opponent does
suicide moves, it should allow a resignation?

> "Brown-1.0fvh" does not resign, so its rating is higher.
> Weak resign-able MC bots lose to him.

For information of everybody: brown is not an application I wrote myself
(stop is), I just run it. It is brown v1.0 by un...@lysator.liu.se


Folkert van Heusden

--
----------------------------------------------------------------------
Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com

Hiroshi Yamashita

unread,
Apr 28, 2012, 10:56:01 AM4/28/12
to compu...@dvandva.org
Hi,

>> So no resigning programs can easily beat him.

Sorry, my english is terrible.
This means programs that do not resign can beat "Stop-cur".

> So if understood this correctly: if stop "sees" that the opponent does
> suicide moves, it should allow a resignation?

I think if "Stop-cur" does not play suicide move over 7 stones,
or filling eye move, his rating will be stable.

For example, Stop lost his winning game by suicide.
http://cgos.boardspace.net/9x9/SGF/2012/04/28/1933559.sgf

Regards,
Hiroshi Yamashita


----- Original Message -----

Don Dailey

unread,
Apr 28, 2012, 12:32:05 PM4/28/12
to compu...@dvandva.org
The only way to be sure you are not getting unfairly penalized is to set your bot to not resign.    Then if you are playing a strong player who is grossly under-rated because it does not clean up,  you will not be unfairly penalized.    

Don

folkert

unread,
May 9, 2012, 4:48:22 PM5/9/12
to compu...@dvandva.org
I'm a little confused: I looked at my gtp code, and I don't see any
command that looks like a resign-command:

out("= protocol_version\n");
out("name\n");
out("version\n");
out("known_command\n");
out("list_commands\n");
out("quit\n");
out("boardsize\n");
out("clear_board\n");
out("komi\n");
out("final_score\n");
out("play\n");
out("genmove\n");
out("place_free_handicap\n");
out("set_free_handicap\n");
out("kgs-game_over\n"); // KGS specific
out("\n");

Only 'play' accepts a 'pass'.


> The only way to be sure you are not getting unfairly penalized is to set
> your bot to not resign. Then if you are playing a strong player who is
> grossly under-rated because it does not clean up, you will not be unfairly
> penalized.
>
> Don
> > So no resigning programs can easily beat him.
> >>
> > Sorry, my english is terrible.
> > This means programs that do not resign can beat "Stop-cur".
> >
> > So if understood this correctly: if stop "sees" that the opponent does
> >> suicide moves, it should allow a resignation?
> >
> > I think if "Stop-cur" does not play suicide move over 7 stones,
> > or filling eye move, his rating will be stable.
> >
> > For example, Stop lost his winning game by suicide.
> > http://cgos.boardspace.net/**9x9/SGF/2012/04/28/1933559.sgf<http://cgos.boardspace.net/9x9/SGF/2012/04/28/1933559.sgf>


Folkert van Heusden

--
MultiTail na wan makriki wrokosani fu tan luku den logfile nanga san
den commando spiti puru. Piki puru spesrutu sani, wroko nanga difrenti
kroru, tya kon makandra, nanga wan lo moro.
http://www.vanheusden.com/multitail/

folkert

unread,
May 9, 2012, 4:51:31 PM5/9/12
to compu...@dvandva.org
>> So if understood this correctly: if stop "sees" that the opponent does
>> suicide moves, it should allow a resignation?
>
> I think if "Stop-cur" does not play suicide move over 7 stones,
> or filling eye move, his rating will be stable.
>
> For example, Stop lost his winning game by suicide.
> http://cgos.boardspace.net/9x9/SGF/2012/04/28/1933559.sgf

Hmmm.
So you mean that it plays suicide moves. I thought I had that covered
because afaik suicide moves are not allowed. What I know by suicide
moves are placing at a spot where you would not have any freedoms.


regards,

Folkert van Heusden

--
MultiTail ist eine flexible Applikation um Logfiles und Kommando
Eingaben zu überprüfen. Inkl. Filter, Farben, Zusammenführen,
Ansichten etc. http://www.vanheusden.com/multitail/

Don Dailey

unread,
May 9, 2012, 5:04:18 PM5/9/12
to compu...@dvandva.org
resign is not a gtp command,  it's a possible move that your engine can play.    

If your program plays an illegal move such as suicide,  it loses the games on CGOS.     

The issue with resign was incorrectly stated by me.   If you resign you lose but of course you expected to lose anyway or else you would not have resigned.    This is not about resignation. 

However if you don't "clean up" the game a program that does NOT resign or pass can take advantage and beat you unfairly.     Just remember these are not Japanese rules,   even if the game is  "played out"  you must clean up dead groups or else the opponent gets to keep them.    

Don

Hiroshi Yamashita

unread,
May 9, 2012, 8:35:27 PM5/9/12
to compu...@dvandva.org
Hi,

> So you mean that it plays suicide moves. I thought I had that covered

Sorry, it was my mistake. It is self-atari, not suicide.

> http://cgos.boardspace.net/9x9/SGF/2012/04/28/1933559.sgf

In this game, 112th move G5 was self-atari.

Regards,

----- Original Message -----
From: "folkert" <fol...@vanheusden.com>
To: <compu...@dvandva.org>
Reply all
Reply to author
Forward
0 new messages