Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: On a formula for computer rating with different time controls

Author: Soren Riis

Date: 09:30:36 02/13/99

Go up one level in this thread


On February 13, 1999 at 05:04:58, Soren Riis wrote:

>I would like to know the relationship is between
>computers playing strength (relative to humans)
>under different time controls.
>
>When I for example play Rebel 8 in 5 min or 10 min
>(on a 366Mhz machine) it vipes of my butt.
>Already Genius 4 had (it was claimed) around 2800 in
>"blitz" rating on a pentium. This does not seems right
>and I usually beats Genius (and Rebel 8) on a 90mhz
>in 10 min games.
>
>We know that the rating roughly goes up 70 point when the
>clockfrequence is doubled. If we assume this is also the
>case for blitz rating it should be possible to come up with a
>(crude?!) formula of calculating a machines rating (against
>humans) on different machines and with different time controls.
>
>It seems to me Rebel 8 (on a 366Mhz machine) have around 3000
>in blitz rating, but this is not a scientific claim and it
>could be off some hundreds in either direction.
>
>Based on the huge experince of machine-human games (on many different
>machines and time controls) can someone suggest a reasonable
>formula?
>
>Soren Riis


Let me try to give a rough approximation: When we dobble the clock
frequence most machines increase their playing strength by 70 rating
points. As for as I know there are NO known serious derivations from
this principle. Is this right?

What happens to a human when we so-to-speak dobble the clock frequence?

I suggest the playing strength goes up by roughly 170 rating points!!

It is a bit difficult to dobble the speed of a given human brain, so
we can do the equivalent by simply dobbling the time (avaleble for the
task) for the human. If my suggestion is right we will find that
if a computer and a human are equal in say a 10 min game, then the
computer will have roughly 100 rating points more in a five minuttes game
(relative to the human) while the human will have 100 points more than
the computer in a 20 minuttes game.

Consider for example a program which have 2400 on a 90mHz. A normal
turnement game is played with say 160 minuttes for each player.
In a 40 minuttes game both the human as well as the computer has 4 times
less time. This will benefit the program which now plays with a rating
of 2600 (relative to the human). [Because the programs strength goes down
2 times 70, while the humans strength goes down 2 times 170]
In a 20 min game it would have rating 2700, while at 5 min it would have
2900.

If the program instead was running on 366mHz its playing strength would
be roughly 130 higher and so it would have 3030 in "blitz"-rating.

On the other hand in correspondance chess (say with 2 days for a move)
a program with 2400 would now (if my suggestion is correct) only have
around 2000. In other words a determined human with around 2000 in rating
(and tremendous energy and determination) would have a fifty-fifty chance
against a 2400 rating program [both human and program are allowed 48 hours
(+sleep!) for each move].

Perhaps the 170 rating points (for humans) is a bit to high? Are there any
strong practical evidence that my suggestion is wrong? Is 130 rating point
more realistic? Any suggestion?

Soren Riis










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.