Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: It takes math to show truth, matter how strongly you feel about it.

Author: Christophe Theron

Date: 21:34:30 01/28/00

Go up one level in this thread


On January 28, 2000 at 16:50:15, Dann Corbit wrote:

>On January 28, 2000 at 16:40:31, Ed Schröder wrote:
>[snip]
>>>It's coincidence, then.
>>>
>>>The method does not work.
>>
>>It's really simple. Play 100 blitz games with Winboard, the Rebel build-in
>>autoplayer or the one of Fritz between 2 equal engines.
>>
>>Match-1 (100 games) 60-40
>>Match-2 (100 games) 40-60
>>Match-3 (100 games) 55-45
>>
>>All very normal.
>>
>>From my experience I say play at least 200. To be absolutely sure 300. This
>>is the main important valuable argument of the SSDF, the high number of games.
>
>With a large number of games (300+) you will get very good results.
>What I object to is if someone runs ten games, gets a 7-3 split, and pronounces
>the first program king and master over the second.  It's a lunatic thing to do,
>and yet it happens all the time.


Dann, I'm surprised that you overlook the fact that I have given this for a 80%
confidence.

If you overlook this, now I understand why most of the time peoples forget about
the interval of confidence and go to the wrong conclusions...

If you follow my method you'll be wrong once in 5 matches. That's what 80%
confidence says.

7-3 does not happen "all the time". It happens less than 20% of the time.

However I would not disagree that this means it happens too often.

Then you just have to recompute the method for a 95% confidence. That's all.



    Christophe



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.