Author: Roger D Davis
Date: 14:48:52 04/11/99
Go up one level in this thread
On April 11, 1999 at 12:22:25, blass uri wrote: > >On April 11, 1999 at 09:53:15, Hans Christian Lykke wrote: > >>On April 11, 1999 at 05:03:12, blass uri wrote: >> >>> >>>On April 10, 1999 at 10:41:03, Harald Faber wrote: >>> >>> >>>>Who knows...what I suspect is that, opposite to what many people DENY, SSDF has >>>>MUCH influence on the sales. >>> >>>I hope that they do not have much influence. >>> >>>The main problem with the ssdf list is that most of the games are not public and >>>I cannot know if the testers played the not public games or only reported >>>results without playing games. >> >>Oh, Uri, please - if you continue this way, there will be NO SSDF list in the >>future. >>I see no alternative to the SSDF-list > >I did not say to have no list but only to tell the public that most of the games >are not public. > >You do an important job when you give games but the problem is that many testers >do not do it. > >everyone can decide if he can trust the results. > >I am interested in the following additional lists: > >1)a list based on only public games >2)a list based only on public games without using knowledge of the opponent >(without Nimzo99 games against Junior,Rebel,Fritz) > >My words (I hope that they do not have much influence) were mainly because of >the fact that Junior5's rating is going down and I cannot find the reason. > >It lost against Hiarcs7,Rebel9,Mchess6 in the ssdf games and I could not find >the games to see the reason that it lost. > >I do not say that there were no games but I have no way to know if the games >were played or only part of them were played. > >Uri I think the request to actually SEE the games is very legitimate. Otherwise, Uri is right...you have to decide for yourself whether to trust the results. Take scientific research, for instance. Deliberate falsification of data is rare, but it does occur. You do sometimes get "one bad apple," even though the system as a whole works well. Assuming the samples are fairly large, one way to test the results might be to calculate the program rankings separately based on public and non-public games. Next, some nonparametric statistical test could be applied to the rankings of the various programs. If there's a significant difference, then the public/non-public variable is distorting the results in some systematic way that deserves experimental control. Roger
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.