David Fotland: <03bc01cb92b5$63f77a10$2be66e...@com>:
>There is a simple formula to estimate the confidence interval of a result.
>I use it to see if a new version is likely better than a reference version
>(but I use 95% confidence intervals, so over hundred of experiments it gives
>me the wrong answer too often).
>
>1.96 * sqrt(wr * (1 - wr) / trials)
>
>Where wr is the win rate of one version vs the reference, and trials is the
>number of test games.  For 99% confidence the 1.96 constant is different.

It's 2.58 for 99% intervals.  The formula, however, cannot be used if 
the win rate is near 0% or 100%, Markus kindly tought me several years 
ago.

Hideki

>I typically run 500 to 5000 test games, which gives a 95% confidence
>interval of 1 to 4 percent.  Then I can calculate the ELO difference at the
>upper and lower confidence bounds to see the range of ELO differences.
>
>David
>
>> -----Original Message-----
>> From: [email protected] [mailto:computer-go-
>> [email protected]] On Behalf Of Darren Cook
>> Sent: Thursday, December 02, 2010 9:25 PM
>> To: [email protected]
>> Subject: [Computer-go] Elo points, improvements and confidence
>> 
>> How many games do two programs need to play to be able to say with 95%
>> confidence that a new feature/bug fix has given a 50 ELO improvement?
>> 
>> What about 200 ELO? What about 99% confidence? I'm sure there must be a
>> straightforward equation for this, but google doesn't understand what I
>> am asking it, and my own statistics knowledge is letting me down.
>> 
>> TIA,
>> Darren
>> 
>> 
>> --
>> Darren Cook, Software Researcher/Developer
>> 
>> http://dcook.org/gobet/  (Shodan Go Bet - who will win?)
>> http://dcook.org/work/ (About me and my work)
>> http://dcook.org/blogs.html (My blogs and articles)
>> _______________________________________________
>> Computer-go mailing list
>> [email protected]
>> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
>_______________________________________________
>Computer-go mailing list
>[email protected]
>http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
-- 
Hideki Kato <mailto:[email protected]>
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to