Newsgroups: rec.games.abstract,comp.ai.games,comp.ai
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!godot.cc.duq.edu!newsgate.duke.edu!news-server.ncren.net!hearst.acc.Virginia.EDU!murdoch!usenet
From: David Coppit <coppit@virginia.edu>
Subject: Re: Announcing a Java-Based Othello Server with AI players.
X-Nntp-Posting-Host: mamba-fo.cs.virginia.edu
Content-Type: text/plain; charset=us-ascii
Message-ID: <319FDDC5.65D2@virginia.edu>
Sender: usenet@murdoch.acc.Virginia.EDU
Content-Transfer-Encoding: 7bit
Cc: David Engler <dae4e@cobra.cs.virginia.edu>,
	Sean McCulloch <stm4e@uvacs.cs.virginia.edu>
Organization: Department of Computer Science, University of Virginia
References: <319E1DC7.3113@virginia.edu>
Mime-Version: 1.0
Date: Mon, 20 May 1996 02:49:41 GMT
X-Mailer: Mozilla 3.0b2 (X11; I; SunOS 5.4 sun4m)
Lines: 46
Xref: glinda.oz.cs.cmu.edu comp.ai.games:4975 comp.ai:38904

Let me try to clear up some of the inaccuracies that I stated in the former
post. The server we designed is the result of a semester project for a
graduate level class. It is by no means meant to compete or replace the
server in Germany.

> Three of the computer players are running in serial, and the fourth is
> running in parallel over 20 computers ranging from Sparc 2's to
> SparcServer 10's.

One objective was to study the ease at which alpha-beta pruning can be
implemented using a mostly transparent parallel programming language
(Mentat). We also hoped to show the benefits of parallelizing the
computations.

> Our computer players, which use alpha beta pruning, are able to look 4
> to 6 levels deep. Our parallel version beats the serial version twice as
> much, and appears to be able to go one level deeper on average. The
> serial version can beat the world's best player (a computer) when it is
> only searching three moves ahead.

Four to six levels isn't that impressive by the standards of "production"
Othello playing programs. These are the results of a 5 second search running
on a SparcServer 10, without any optimizations or an opening book. If this
were a project for an AI class instead of OS, we would have spent more time
on this :). Hopefully we'll have time this summer to streamline it and make
it very sharp. The important point we made was that the overhead of
distributing the computations over 20 computers did not kill performance too
much.

Perhaps I misunderstood the relative abilities of computers and humans at
Othello. I thought that the best computers regularly beat the best humans.
Oh well, so much for posting on a subject I know little about...

One of my partners, Sean McCulloch, actually played the computers against
each other and was able to fill in the details for me. Our very cursory
tests were against Logistello when it was looking 3 ply ahead - namely
Patzer. Our player was running in serial for 30 seconds a move.

Hope that clears things up a bit!
David Coppit

-- 
-------------------------------------------------------------------------
David Coppit - Graduate Student        coppit@virginia.edu
The University of Virginia             http://www.cs.virginia.edu/~dwc3q
"Shine on you crazy diamond." - Pink Floyd
