Author: Larry Proffer
Date: 07:58:01 05/08/01
Go up one level in this thread
Thank you for your reply. I concur. What we do know is that Fritz and Junior are, to all intents and purposes 'equal', or very nearly 'equal'. If we need to find a winner, it makes *no* difference how many games we play. If we play one game, Fritz has a 50% chance, Junior a 50% chance. If we play one thousand games, Fritz has a 50% chance and Junior has a 50% chance. Number of games is not relevant when they are so closely matched. All we do know is that under the BGN-Enrique match conditions, Chessbase have 100% winning chance. Now, if we introduce Tiger to the equation, we can assume it is very nearly equal or equal to Fritz and Junior. Again, if we play games, 24, 100, 3, whatever ... Tiger will win 33% of the time. Fritz will win 33% of the time. Junior will win 33% of the time. Chessbase's win chances, however, fall to 66%, because they only exclusively publish two of the above programs. So our "independant computer chess experts" certainly aren't expert in statistics or logical reasoning when they say: They added that they unfortunately didn't >>have time for a tournament with 10 programs which would have taken too long to >>run 1. They can't add up - it wouldn't have been 10 programs. 2. They can't do statistics, because the 24 game number was quite arbitrary and makes no difference anyway to what is effectively a coin-flip. 3. They can't reason logically because they apply non-senseful arguments to 'make' their case. On May 08, 2001 at 10:43:41, Martin Schubert wrote: >On May 08, 2001 at 10:19:52, Larry Proffer wrote: > >>"Braingames explain their reasoning. "We made a simple decision. We wanted >>programs which could play on multi-processor platforms as they are obviously >>stronger candidates for the Kramnik match. There are really only three >>candidates: Fritz, Junior and Shredder. We made great efforts to persuade >>Shredder to play but they declined." They added that they unfortunately didn't >>have time for a tournament with 10 programs which would have taken too long to >>run. One of the main complainants was the company REBEL. Their TIGER program is >>a single processor prgram yet still finished second in the Cadaques event run by >>Prof. Irazoqui earlier in the year. They actually have a multi-processor version >>called DEEP TIGER but that wasn't announced until after the invitations were >>made." >> >> >>Shortage of time is now the weak(?) excuse. >> >>We know they had time for a 24 game match between two programs. This produces a >>winner with a degree of confidence that this winner is 'objectively' the best >>comp-comp program. Can anyone calculate this degree of confidence? >> >How should this be possible? First you need a zero hypothesis, e.g. Fritz is as >good as Junior. Okay, that's not the problem. But statistics is only possible >when results are independent. When you're using booklearning, they're not >independent. So you can't calculate a degree of confidence. > >>If Tiger was included, then they would have been in the situation of needing to >>play a match between three programs. Since they were intending to play a three >>program match anyway (with Shredder as the third), one can assume that they'ld >>have had time to include Tiger anyway. However ..... >> >>But, and my question is this: suppose they played a 24 game match with three >>programs; my weak maths suggests that instead of each program playing 24 games >>(as in 2-player match), each program would play 16 games for a 24 game match >>with three players; >> >>then: what is the degree of confidence that the 'winner' is 'objectively' the >>best? >> >>Is it actually much different to the degree of confidence for the 2-player >>match? > >If you've the hypothesis "every program is as good as the other ones, every game >is independent from all other games,..." (which is not true), then (under this >hypothesis) it doesn't matter if one program plays only against one or if it >plays against two opponent. So the level of confidence depends only from the >number of games and the score. So the level of confidence would be the same >(under a lot of assumptions, which do not hold). >> >>How many more games are needed to reach the same degree of confidence as the >>2-player match? >> >12 more games because then every program would play 24 games. >No more games. >>Is the 'time available' argument strong or weak? >One more program: 50% more time needed. > >Martin
This page took 0.01 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.