Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The Scrappy - Crafty Experiment

Author: Robert Hyatt

Date: 05:47:34 06/12/01

Go up one level in this thread


On June 12, 2001 at 06:38:17, Mark Young wrote:

>
>Bob, are you able to draw any conclsions yet on this experiment? It seems as
>predicted by many, players are able to spike ratings by creating sub pools of
>players on any given account. Meaning if you don't allow all players to play all
>players you corrupt the ratings pool.


I don't think it corrupts the rating pool, since this "pool" is already badly
fractured anyway.  The rating system really isn't meant to handle the various
"abuses" that go on.  IE I watched a very strong player beat a commercial
program pretty easily in a particular opening.  He then went around and
challenged all the "clones" and beat them in _exactly_ the same way.  I doubt
Elo ever thought about that kind of "thing".  :)

As far as the experiment, yes, it was interesting.  I had specifically been
working on anti=human stuff for quite a while, and (to me) it had become pretty
obvious that what was good to do against humans is not particularly good to do
against programs.  We _know_ that in open positions, the computers are at least
as good (and generally better) than human players.  Therefore, trying to reach
open positions (even if positionally it is not so good) is a good thing to do
against them... make them compete against the machine where it is best. But
against other computers, opening the position is likely not a good or bad idea
since both players are reasonably tactically accurate, but if you do it by
weakening your position, it is dangerous.

I have left scrappy running at the request of several strong players.  They
complained that "crafty" played so many computers, and computers are so bad to
jump into the middle of a match and interrupt it, they had trouble playing.
With Scrappy, they could play whenever they want without any mechanical
interruption.

So long as I have a machine that is basically idle most of the time, I will
keep Scrappy online... but this could end at any time.



>
>This would make most if not all online ratings meaningless if true.
>
>
>
>Information about scrappy(C) (Last disconnected Tue Jun 12 2001 04:19):
>
>          rating [need] win  loss  draw total   best
>Wild        2108  [6]    24     0     2    26   2108 (01-May-2001)
>Bullet      3271        853    81   158  1092   3314 (15-May-2001)
>Blitz       3373       6650   982  1743  9375   3544 (20-Oct-2000)
>Standard    2698  [3]   191    69    73   333   2700 (10-Jun-2001)
>
> 1: Crafty v18.10 (4 cpus)
> 2: Intel SC450NX quad xeon/550mhz, 512mb RAM, 5X9 10K LVDS disks
> 3: Linux 2.2.14 (SMP)
> 4: message hyatt if you have questions/problems...  this account is always
>unattended
> 5:
> 6:
> 7:
> 8: scrappy is only playing humans.  crafty plays anything.  This is an
>experiment to see how avoiding computers affects the overall rating.  If you
>operate a computer and don't like this policy, tough.
> 9: crafty and scrappy use the same identical hardware and software
>configurations.  Feel free to play crafty so long as you follow its finger
>notes.
>
>Statistics for crafty(C)         On for: 1:47     Idle:   19
>
>          rating [need] win  loss  draw total   best
>Wild        2000  [6]   332   100    11   443   2204 (03-Oct-2000)
>Bullet      3255       6664  1467  1087  9218   3255 (11-Jun-2001)
>Blitz       2995      54355 14553 10834 79742   3388 (09-Jun-2000)
>Standard    2748       4511  2011  1698  8220   2792 (25-Oct-2000)
>
> 1: Crafty v18.10 (4 cpus)
> 2: crafty now uses all 3/4/5 piece endgame databases, over 20gb.
> 3: Intel SC450NX quad xeon/700mhz, 512mb RAM, 56 gigs 10K scsi
> 4: hours online: 39,940;  percent of life online: 78%
> 5: Crafty is freely available, both source and a windows executable, for those
>that don't have unix machines they can compile on.
> 6: anonymous ftp to ftp.cis.uab.edu/pub/hyatt
> 7:  note that I +noplay computers without notes giving their hardware/software,
>or computers that don't play an even number of games.  This also includes anyone
>matching me while having !computer in their formula.
> 8: computer policy:  no more than 4 blitz or 2 standard games by any one
>program in Crafty's 20-game history here.
> 9: I do not expect to see games vs 6 different chess tiger clones.  It is your
>responsibility to know/find out the name of the program that others use to avoid
>falling into this trap.  The last opponent gets +noplayed.
>10: (2) do not interrupt human games under any circumstances.  I am mainly
>interested in playing strong humans, but will allow computers to play so long as
>they don't monopolize crafty's time.  Beware.



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.