Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Ply Ratings

Author: Jim Phillips

Date: 22:19:20 09/30/98

Go up one level in this thread


On September 29, 1998 at 21:02:06, Jim Phillips wrote:

>On September 29, 1998 at 14:00:40, William H Rogers wrote:
>
>>Has anyone done any research on the ratings that a program has at different
>>plys?
>>I once saw a test of chess 4.7 and they had some preliminary ratings on their
>>program at different ply levels. They could set the depth that their program
>>would search, not necessarily based on time.
>>bill
>
>Hi Bill,
>There was an article in Scientific American in approximately 1990-1991
>which had a graph showing the play ratings of various chess programs and
>how many plies deep they searched.  I don't have the article handy here;
>I would have to dig it up.  But I do remember definitely that the graph
>implied that a program's rating increased by 180-200 points for each ply
>deeper it searched.
>It was displayed as a straight-line graph.  I don't know if that's right
>or wrong; it's certainly debatable.
>-Jim Phillips

Bill,
I dug out that article when I got home and took a careful look at the
graph.  It's linear all the way from beginner up to 2000+ rating....
I assume it is for programs with a lot of knowledge, playing at 40/120
time control.  I determined that a best-fit formula to duplicate the
graph is:
  U.S. Chess Federation Rating = (200 x Ply Depth) + 550
This could be applied to your idea in your other post for a computer
program that rates its human opponent.  I've had some other ideas about
that since reading your post; if I come up with a general approach I'll
post it.  Such a program would certainly be a worthwhile project... very
interesting if nothing else....
-Jim Phillips



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.