Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: NO MORE human vs computer game again

Author: martin fierz

Date: 00:51:15 08/05/03

Go up one level in this thread


On August 04, 2003 at 19:04:39, Dieter Buerssner wrote:

>On August 04, 2003 at 18:36:27, Dann Corbit wrote:
>
>>If a player rated 500 points below a machine in real strength plays enough
>>games, he will win one of them.
>
>If we can trust the statistics behind the ELO system. I am not sure, if it is
>true for bigger rating differences (in the tails of the distributions). Also, to
>my knowledge, the statistics does not say anything about the distribution of
>wins/draws/losses. If you are expected to get 1% of the points, you could do it
>by just drawing every 50th game and don't win any game.
>
>>It may take quite a lot of games.  But this is not an opinion, it is a fact.
>
>I think, it is not a "pure" fact. It (the percentage of points, you reach, not
>necessarily the number of wins) might be a fact, when we take the statistics
>assumed behind the Elo system as a fact. But still my above point about
>distribution of wins and draws would be valid.
>
>>This assumes simply playing games based on strength.  If the human is clever at
>>learning computer weakesses or playing anticomputer chess, it may tip the
>>balance somewhat.
>>
>>Win expectency for a difference of 0 points is 0.5
>>Win expectency for a difference of 100 points is 0.359935
>>Win expectency for a difference of 200 points is 0.240253
>>Win expectency for a difference of 300 points is 0.15098
>>Win expectency for a difference of 400 points is 0.0909091
>>Win expectency for a difference of 500 points is 0.0532402
>>Win expectency for a difference of 600 points is 0.0306534
>>Win expectency for a difference of 700 points is 0.0174721
>>Win expectency for a difference of 800 points is 0.00990099
>>Win expectency for a difference of 900 points is 0.00559197
>>Win expectency for a difference of 1000 points is 0.00315231
>
>I am no native English speaker. Win expectency sounds a bit misleading to my
>ears, because it does not really mean, that you win that many games.
>
>BTW. I calculate slightly different numbers. The correct formula is:
>
> p = 0.5 - 0.5*erf(rating_difference/400.)
>
>With:
>
>                           x
>                            -
>                 2         | |          2
>   erf(x)  =  --------     |    exp( - t  ) dt.
>              sqrt(pi)   | |
>                          -
>                          0
>
>Or for the numerically interested people, for rating_difference big, the form
>
>p = 0.5*erfc((rating_difference)/400.)
>
>would be better, when a "good" erfc-function is available (erfc(x) =: 1-erf(x)).
>It would avoid to calculate the difference of numbers of equal size.
>
>
>Regards,
>Dieter

there is this jeff sonas guy who sometimes is interviewed on the chessbase
website. he's a statistician (?) and did some work on improving the rating
system. among other things, i believe he also thought about the difference of
having white/black and about how points are made - with wins or rather with
draws. i.e. . you can read more about his theories on his site,
www.chessmetrics.com, or in this chessbase article on dortmund:
http://www.chessbase.com/newsdetail.asp?newsid=1094

cheers
  martin



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.