Author: Dieter Buerssner
Date: 12:00:02 10/04/03
Go up one level in this thread
On October 03, 2003 at 16:20:21, Sune Fischer wrote: >What you are doing is to find the probability of the result using an already >given probability distribution. I wouldn't call the probabilities of win/draw/loss (which can be experimentally sampled) a probability distribution ... It is more like the question: You have a dice with 6 face. Each number has the same probability. You do 100 throws and add together the number of each throw. Now you can calculate the probabilities of each result, or ask yourself, how likely is it that the sum will be 380 or bigger. Or the similar question: I got 400 as sum, is it lkely that my assumption about the dice was correct? >The questions is, with what level of confidence can we claim that the program >scoring higher is also better? I agree, that there is a better way (the way Remi has done it, I mentioned his article). But I think my table gives already a good overview. >We know that the confidence level grows as the number of games increases, so it >has to be part of the formula somehow. Not only with the number of games, but of course also with the result. The interesting thing, that Remi had found out (and that is understandable after a while) was, that for the question draws play no role. So 1010-1005 with 2000 draws gives the same confidence as 10-5 with no draws. My manner of calculating the table will show this also (assumed draw probability will be different of course). Regards, Dieter
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.