Author: Chris Carson
Date: 14:57:29 07/26/00
Go up one level in this thread
On July 26, 2000 at 17:12:17, ShaktiFire wrote: >Nice analysis. Good to see someone actually >run the numbers through. Still need a few more >games to claim 95% confidence of having a program >at least 2500 level. 2500 is a key number isn't it? "It is possible to infer something quantitative from the data that Chris has accumulated. Based on some plausible assumptions I make the following claim: ******************************************************* ***Claim: Chris Carsons' data implies that some program has a rating of 2496 or more with approximately a 95% confidence.*** ******************************************************* If the results were produced by a single program, its TPR would be 2544. This gives us some idea of how many rating points we have to give up to have 19-to-1 odds in our favor. To be fair, we should also mention that we could as well infer that there is a program with a rating at or below 2602 with a 95% confidence! The result above is based solely on the data and is therefore most useful only if the situations which produced the data do not change. On the other hand, such considerations may be used just like changing ratings to track the progress of say, GMs perhaps learning successful anti-computer tactics and of programmers fending them off. Anyway, I am interested in what, if anything, CCC makes of such a result. The rest of this post just derives the claim above. It is divided into three sections for easier reading." If I read it right the claim is 2496 or more (lower bound). It also goes on to say that if a single program produced the results, then the TPR would be 2544. Also, 2602 or below is the upper bound for a 95% confidence. Did I miss something? 2544 is greater than 2500 and the upper bound is 2602. If I am mis-interperting, my mistake. :) Best Regards, Chris Carson
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.