Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 9 rounds will not always give you the "best" program

Author: Rolf Tueschen

Date: 12:45:56 01/21/03

Go up one level in this thread


On January 21, 2003 at 14:07:13, Dann Corbit wrote:

>On January 21, 2003 at 06:00:27, Rolf Tueschen wrote:
>
>>On January 20, 2003 at 21:20:35, Dann Corbit wrote:
>>
>>>On January 20, 2003 at 21:18:48, Dann Corbit wrote:
>>>
>>>>On January 20, 2003 at 17:47:05, Frank Phillips wrote:
>>>>>On January 20, 2003 at 16:51:00, Dann Corbit wrote:
>>>>>>No contest can truly tell us which program is strongest.  Not even a trillion
>>>>>>rounds of round-robin.
>>>>>
>>>>>So how many?  Or can we never know anything?  And what does it therefore mean to
>>>>>be strongest?
>>>>
>>>>When the programs are very close (as is the case in reality) we will never know
>>>>the answer.
>>>>
>>>>The farther apart the programs are, the more easily we can be certain.  For
>>>>instance, I am utterly convinced the deep fritz is stronger than TSCP.  I am not
>>>>convinced that it is stronger than Chess Tiger.
>>>>
>>>>If there is a 1000 ELO difference between two programs, we will find out very
>>>>quickly.
>>>>
>>>>If there is a 100 ELO difference, it becomes more difficult, but a huge number
>>>>of games will give us a very convincing result.
>>>>
>>>>If there is a 10 ELO difference, it will be impossible to tell, as the
>>>>randomness in play will be a greater difference than the difference in true
>>>>strength.
>>>
>>>P.S.
>>>There is nothing wrong with knowing we cannot find an answer with certainty.
>>>There is something wrong with pretending that we can.
>>
>>Maybe I can also be of help here in that debate. I see a general mistake in the
>>argumentation.
>>
>>You say that possibley two progs are 10 Elo point seperated. And you want to
>>argue that in such a case the "real and superior strength" could NEVER be found.
>>Whereas if the difference is 100 Elo points that should be possible to discover.
>>You even reason that never means a trillion of games.
>>
>>Now where is the fallacy?
>>
>>Very simple. And I think it proves beyond a doubt that you are NOT a
>>statistician and you never were, at least not on universitarian level, which is
>>not bad in itself after all, so please don't take this as an insult. You take
>>the 10 points - for spooky reasons - as eternal value. But that is wrong!
>>Because the points are already a result of former games and 10 points might
>>speak into the direction zero or into the direction 100. But exactly this would
>>be discovered if we had more games. So also here, as in your coin analogy, you
>>are completely misleaden. I'm sorry, Dann.
>
>I am not a statistician.  My degree is in numerical analysis.
>
>Nevertheless, we will never discover which is stronger, Deep Fritz or Chess
>Tiger.  By the time billions of games could be played to determine the answer
>(if -- indeed -- billions is enough) nobody will care about either program any
>more.  Furthermore, nobody is interested in performing that much effort to
>discover the answer to a question nobody will care about.
>
>I don't need to bother to do the math to know that I am right.

Here you are right but let's not give up so early, I bet that one of the two is
better in the longer run. And I am sure it is - - - - - wait a minute.... it's -
- - - Fritz! Chess Tiger is for French people, you know! On holidays I read
Science et Vie and there in that Journal of France Christophe had the chance to
make a lot of noise :) for his product, as if it had won anything. I planned to
write a message about it but then, what for, France is not a leading chess
nation after all, neither Guadeloupe or Africa. No need to colonialize the whole
world with Fritz. Enough to know that t's the best. So, like Czub I prefer the
NO GAME judgment. Fritz is better.

:)

Rolf Tueschen



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.