[Top][All Lists]

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

[Bug-gnubg] RE: Computer Olympiad

From: motiv4u
Subject: [Bug-gnubg] RE: Computer Olympiad
Date: Tue, 8 May 2007 09:30:00 +0200

Hello François,

For GNU Backgammon, at the time being, I only participate in person, thus
Adrian Wright (and the gnubg development team) should have the last word in
this one.

Yet I'd like to point out:
Longer matches and more matches played tend to favorite the strongest
player, as luck becomes less and less envolved.

Playing best out of 3 1-pointers shouldn't be the scenario for designating
Olympiade's Gold Medal Winner.
The previous scenarios had best out of 5 15-pointers.

I think best out of 3 11-pointers is a reasonable draw-back, considering the
expected playing time of your bot will exceed the playing time of gnubg and
BgBlitz by a factor X. And X might be?

Even playing at 4ply on an old Pentium Celeron IV 2GHz, gnubg moves within 5
minutes. Extremo or WorldClass settings - and even GrandMaster setting -
make decisions within 1 minute, allowing 1,000 moves withing the time window
of the 16 hours you forsee at the Amsterdam Olympiade.
Thus time isn't the issue for gnubg.

And talking about your bot: do I understand correctly it will play on-line
matches to define it's next move?

Nardy Pillards

> -----Original Message-----
> From: François Van Lishout [mailto:address@hidden 
> Sent: Monday, May 07, 2007 5:59 PM
> To: 'Frank Berger'; 'motiv4u'; 'Adrian Wright'
> Cc: 'Chaslot G (MICC)'
> Subject: RE: Computer Olympiad
> Hello,
> I have talked with Mark Winands and he says that he allows 
> two instances of the programs to run simultaneously. Of 
> course two machines are needed and two operators. The spirit 
> of the competition is to play as many games as possible. 2 
> days of competition are enough for us too. In the morning, we 
> will participate to the workshop, so Mark proposes to start 
> around 1 pm each day. We would have a maximum of 16 hours of 
> playing time (8 hours per day).
> I'm currently writing a Backgammon program with Guillaume 
> Chaslot. He works at the University of Maastricht and is the 
> author of one of the best Go program of the world which is 
> based on a new Monte Carlo approach. We decided for about one 
> month to try this approach also on the Backgammon and see 
> what happens. We are still developing the program focusing on 
> ways to compute the random games as fast as possible. We thus 
> do not now yet if it can lead to a strong program or not. We 
> will use the championship as a test bed of the strength of a 
> Monte Carlo approach vs a learning approach in games. In 
> future works, it can of course be envisaged to use a 
> combination of both.
> We should also discuss about the thinking time. As far as I 
> know, for you it is not important as the learning has been 
> done offline and your program plays very fast. Our program 
> however will play better if it plays lots of online random 
> games... We are thus favourable to shorten to best of 3 so 
> that we have more time for each single game.
> What do you think about all this ?
> Guillaume Chaslot
> François Van Lishout
> -----Message d'origine-----
> De : Frank Berger [mailto:address@hidden Envoyé : 
> vendredi 4 mai 2007 18:13 À : 
> address@hidden; motiv4u; Adrian Wright Objet 
: Computer Olympiad
> Hello Mr. van Lishout, hello Adrina, hello Nardy,
> @Mr. van Lishout: Mark Winands from the ICGA mailed me that 
> you want to participate with your programm in the Backgammon 
> event at the Computer Olympiad. That's great news. After just 
> two programms in 2003, 2003 and 2006 another competitor!
> In the past we always settled on the mode of the competition. 
> We always played at two days and 15.6/16.6 seems a good date 
> for it (just one holyday needed).
> In 2006 due to faster algo  / computers we needed for 4 
> 15-point matches IIRC 8-9 hours. Having 3 sequential matches 
> looks hard to put in 2 days. Maybe if Nardy and Adrian are 
> both there, we could have two instances of GnuBG (or BGBlitz) 
> running so we need only two days (if the organizer allows 
> this)? Or we might shorten from best of 5 to best of three. 
> Or best of 5 with a shorter length. Or play 3 days (15-17?).
> What is your opinion?
> @Mr. van Lishout: And finally: naturally I'm curious. Could 
> you reveal some informations about your program?
> Tot ziens
> Frank Berger

No virus found in this outgoing message.
Checked by AVG Free Edition. 
Version: 7.5.467 / Virus Database: 269.6.5/793 - Release Date: 7/5/2007

reply via email to

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