Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Statistical significance of a match result

Author: Rémi Coulom

Date: 11:28:57 11/23/02

Go up one level in this thread


On November 23, 2002 at 11:06:20, Gian-Carlo Pascutto wrote:

>Remi, Peter Fendrich also wrote a utility that (I think?) works similarly
>to yours. I just emailed you the source.
>
>--
>GCP

Thanks for the mail. I have taken a look at it, and the principle is the same.
The major difference is that I obtain the result analytically, whereas he uses a
Monte-Carlo method to estimate the integrals. The consequence is that my program
is much more faster, more accurate, and can deal with much larger numbers
because I use a multiprecision library. Also, I proved that the number of draws
does not count.

Steve Maughan's tool seems extremely fast and accurate. It seems like it
consistently gives a result that is one more than my program, though. For
instance, for 1000 losses and 95% confidence, his programs says 1076 wins are
required, while it should be 1075 according to my calculations. Maybe he has a
small bug. I'll email him to ask how he does, because he very probably
understood something I have not. I do not understand how he can obtain the
result that fast.

Rémi



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.