Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: STATISTICS

Author: Michael Neish

Date: 18:43:36 01/20/00

Go up one level in this thread


On January 20, 2000 at 08:07:28, Robert Hyatt wrote:

>Yes, although the reasoning is often twisted a bit.  IE yes, there is a
>probability of almost 1.0 that if you play enough games, that one program will
>win N in a row (ie 4 in the above).  But the probability that this is the
>_first_ N games played is very low.  Somewhat like the "runs test" for random
>number generators.

Of course you are right, however we were discussing the probability of a washout
occurring if only four games are played, and that is one in eight if both
computers/players are evenly matched and there are no draws.  The figures I gave
for a 15-9 score are absolute probabilities, in other words you'd expect a 15-9
score to occur in sixteen out of a hundred 24-game matches.  I did the same
calculation including draws, but here one needs to know the proportion of games
that tend to end in draws, which is surely different for each computer and also
depends on the strength difference between them.  It's curious that if you play
a 24-game match between computers of equal strength, you can expect a 13-11 and
14-10 score to occur _more often_ than the expected 12-12, and a 15-9 score only
slightly less often.  Draws tend to even things out a little, so that extremely
lopsided results have a lower chance of occurring.  If you're interested I will
give you the figures that I came up with, although anyone with a little bit of
patience can easily work them out.

While we're talking about this, do you know a place where I could get a table of
percentage expectancy as a function of Elo difference?  And also, do you know
what the Elo difference is between White and Black?  I heard it was about 25 or
27, but this is off the top of my head.

Cheers,

Mike.




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.