Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Benchmarking chess algorithms

Author: Dann Corbit

Date: 11:38:49 07/20/99

Go up one level in this thread


On July 20, 1999 at 14:28:54, KarinsDad wrote:

>On July 20, 1999 at 13:58:38, Dann Corbit wrote:
>[snip]
>I think this is really a matter of doing homework on research that has already
>been done. I think the answers are already out there, it just would take someone
>some time to collate what is already there. For example:
>
>http://www.cs.vu.nl/~aske/Papers/abstr-ks.html
>
>When I started my project, I read a lot of papers. I did not completely
>understand them all, but I understood enough to know that doing things the
>standard ways was a deadend which would get me at best to the same place as
>everyone else. So, I started coming up with my own ideas and decided to
>incorporate them with techniques which have worked for others, but which did not
>conflict with what I was doing.
>
>Since ALL of these different techniques you mention have been done by a LOT of
>people in the past, one can conclude that the current best standard algorithms
>are probably already being used by Fritz, Hirarcs, Junior, Schredder, Ferret,
>Crafty, and others. So, it may be easiest to just ask the authors of the
>strongest programs which basic algorithms they are using. You already know the
>strengths and weaknesses of these various programs. Knowing EXACT statistical
>information may be more misleading then it is informative since in my opinion,
>the next iterative step in chess programming is to NOT use a bunch of standard
>algorithms, but to come up with something new (a paradigm shift).
I don't think I could possibly disagree more strongly.  I think the notion that
you invent your own stuff and assume it is better is a truly horrible idea if
you don't know exactly how good the alternatives are.  Your notion of checking
to see what has already been done is an excellent one that I strongly agree
with.  However, having a controlled framework for testing ideas is an idea whose
time has come.

If you don't know how good the competition is, you have no idea how good your
new ideas are.

Take the example of good programs.
Perhaps one program does one thing better than anyone else, and that is why it
is in the top ten.  Perhaps a different program does a different thing better
than anyone else and that is why it is in the top ten.  Yet another program may
do a bunch of things fairly well and that is why it succeeds.  Without knowing
the true behavior of the algorithms involved, it is not science.  It is not art
either.  It's just monkeying around.  Tinkering can be fun, but someone who
understands fully all the processes involved and the principles behind them will
kick the butt of a tinkerer every time.

Having a carefully prepared list that describes in detail the behavior of every
algorithm at your disposal is *essential* in order to even know if your idea is
better!



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.