Author: Christophe Theron
Date: 22:10:56 01/28/00
Go up one level in this thread
On January 29, 2000 at 00:37:08, Dann Corbit wrote:
>On January 29, 2000 at 00:15:05, Christophe Theron wrote:
>
>>On January 28, 2000 at 22:47:02, Dann Corbit wrote:
>>
>>>You make some good points, and it also depends on what you are really after.
>>>
>>>If you have a 9-0 run, it *could* be a fluke, but *odd are* it's not. So if you
>>>don't care very much how accurate the answer is, then you might just decide to
>>>quit testing at this point.
>>>
>>>So the degree of testing is really a function of the accuracy desired.
>>>Which is the [excellent] point that you made in your post.
>>
>>
>>That's just what I wanted to say. I just try to see which program is better, I'm
>>not trying to give them relative elo ratings.
>>
>>The process of telling which program is better takes a variable amount of games.
>>Few games if the elo difference is big, more games if it is smaller.
>>
>>I'm not sure my numbers are accurate, and I would appreciate to see somebody
>>checking them.
>
>As long as you are willing to accept wrong answers once in a while, I don't see
>a problem with it. For instance (I did not check to ensure your figure of 80%
>was right, but assuming that) 1/5 of the time, your conclusion will be wrong!
>
>If you don't mind being wrong 1 out of 5 times, then that method would be OK.
So we agree. So if you want to be right 95% of the time, you just have to
compute the table for 95% confidence.
But still the methodology is OK I think.
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.