Author: Thorsten Czub
Date: 18:06:08 09/26/99
Go up one level in this thread
On September 26, 1999 at 19:40:38, Georg v. Zimmermann wrote: > >>>But SSDF is only counting the _results_. And for that you need at least 200 >>>games against every opponent, probably more. >> >>thats knowing the price of the milk and trying to find out the quality by >>watching the price of the milk. >> > ><snip> > >ok, lets agree to disagree on the other topic.- but don't you agree that the >current ssdf list where the 1. program has played less than 200 games overall >(I think you need >200 against 1 to show which is stronger!) is completely >useless. Or can I get any direct comment on my experiment ? imo the ssdf is a good organisation, they are a private organisation. one has to understand the sense of the data they produce and the problems such a list produces. also one has to know about the system-immanent-errors that are produced by the programs. how do they handle doubles. how do they behave concerning different versions of one programmer? how do they treat older versions? and one has to understand the error margin. i think i have often seen wrong order of the list- and most often i have seen right order. in the old days the order was nearly senseless because almost any program they tested was a lang program. roma68000, roma6820, almeria68000,almeria68020 and and and. so often roma lost against almeria lost against ... but it is clear that an older version loses against a newer version of a programmer. not because it is stronger. but often because it is newer. later they often had different machines. and they gave the "weak" programs weak machines, meanwhile the "good" programs got good machines. hiarcs e.g. always got the weak machine because they decided that hiarcs is a weaker program than genius is. this was nonsense. but they did it - no matter what you said about it. they did not changed this even when hiarcs was world-champion. it took a year until they had "enough resources to" test hiarcs on an as fast hardware they tested genius the whole time. one day they decided to throw out some machines. they said these machines are not commercially available. but these machines WERE commercially available, although expensive (1500-2000 DM when i remember it right). this again was a help to mephisto. because TurboKit with Rebel program was as strong as richard langs machines. always there was a kind of strange influence in the list. i guess today the ranking looks ok. i don't understand why tiger is a weak program :-)))) but if they would test it on 450 mhz machines they would see HOW strong it is. why do these guys always INFLUENCE the statistics ? i mean: i trust the numbers or not. saying: tiger is weak, i don't give it the 450 machines and instead test it on p90 !!!!!! (5 times slower !!!!) is not really interesting to us, or ? we want to know how strong tiger is in relation to the 450 Mhz machines. these are some critics. despite these few points (maybe i have forgotten some) the list looks very ok. i do often use it myself. of course they should have never allowed cb this special arrangement they did. this was not ok. it was against the common sense. but - this is a private especially swedish club (foreigners not allowed!!!) and it is their affair. my main critic is that the they avoke a kind of scientific outlook they cannot fullfill in real. so people misuse their list. use the gaps within the system. this is not fair in my opinion.
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.