Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: statistics, 10 events tell us what ?

Author: Peter Fendrich

Date: 06:04:51 08/13/98

Go up one level in this thread



With some background within statistics, as opposed to what you might think, I
allmost completely agree with you...
It all comes down to capturing information and we should use both "brain halves"
instead of only one side... :)

If you try to predict the next 10 games with only 10 previous games known there
are a bunch of statistical problems to deal with. Suppose we wan't to apply the
"Normal distribution" (the bell curve) it can be done. There are lot of good
algorithms to use and that's what prof. ELO did. It's even possible to know how
sure we are within an error interval. The problem here is that we are only
simulating the Normal distribution and we need more games before it's even
possible to apply the formulas. At least, say 30-50 games. So with 10 games we
shouldn't even try. (It would be like *manually* taking the time for running 100
m races and then compute the mean value with 10 decimals. Useless
information...)

There are other distributions to apply in the 10 games case, but the problem
with prediction remains.
The numeric information only from the previous results gives us very little
information.

What you are proposing is to look at the games themself, which will give us much
more information but of a kind which is very hard to use in formulas.
It's possible to find out patterns and compare them with the actual result.
Suppose we have a 10-0 result and observes that PgmA completely crushed PgmB in
every game, then we would predict about the same result in the next 10 games.
Another scenario is of course contradicting patterns, telling us that the result
was more or less random which will not at all support the 10-0 result.

So by using all the information available and using the statistics as one of
several inputs it's easier to make good predictions.
Common sense is not a bad thing after all, even if it probably (in the 10 game
match) will give us different predictions, but that's life... :)

//Peter





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.