Author: Rémi Coulom
Date: 05:12:33 11/24/02
Go up one level in this thread
I have just updated my web page with a new improved version. I found a simple trick to make the computation much cleaner and faster, and the program does not require the multiple precision library anymore. The trick is that, if you note P(n0,n1) the probability to be the best after n0 losses and n1 wins, the following relation holds: P(n0,n1) = 0.5 * (P(n0 - 1, n1) + P(n0, n1 - 1)) I have yet to find an intuitive explanation for this equation. I suppose it is tandard statistical stuff, but, again, I am not an expert. Rémi On November 23, 2002 at 10:51:35, Rémi Coulom wrote: >Hi, > >I have just added a small utility to my web page. You give it the number of wins >and losses of a match, and it calculates the likelyhood that one player is best. >It comes with a short paper explaining how the calculation works. You can get it >at: >http://remi.coulom.free.fr/ > >I know someone else (sorry, I do not remember who) posted something similar some >time ago. I wanted to make my own and share it with everyone. The way it works >may sound a little naive to an expert in statistics. The technique I use >probably already has a scientific name, and there may be better ways to >implement it, but I am confident it is correct. I believe it is much better than >the Student's t-test that I have seen used in some scientific papers. > >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.