Author: Dan Honeycutt
Date: 00:05:59 06/19/04
Go up one level in this thread
On June 19, 2004 at 02:38:14, Sandro Necchi wrote: >On June 18, 2004 at 14:31:12, Steve Glanzfeld wrote: > >>On June 18, 2004 at 13:39:29, Rolf Tueschen wrote: >> >>>On June 18, 2004 at 12:59:43, Steve Glanzfeld wrote: >>> >>>>On June 18, 2004 at 09:47:55, Rolf Tueschen wrote: >>>> >>>>[...] >>>>:-))) I can imagine how a blackout must have suddenly hit you. Has someone >>>>turned the lights off while you were writing? Why in the world is it "HYPNOSIS" >>>>??? when people believe the truth to be true? >>>> >>>>Steve >>> >>> >>>As I told you - with your insults you can't expect to get answers. You showed >>>very well that you have a reading difficulty because above I didn't write that >>>_I_ believed that the ranking lists were "similar". That was a quote from >>>Gurevich. Understood? >> >>But in fact he's right, they ARE similar!! Understood? Compare any rankings you >>like... >> >>>To all the other problems I am certain that to your reading difficulty you have >>>even worse handicaps because you don't seem to be fit to get what is being >>>discussed here. This test can't bring effective news, this is the main point. >> >>Why is this "the main point" suddenly?? You find new "main points" every day. >>Don't you know that new engine versions are released every week? Testing them >>DOES bring news, because there is no other estimation of their strength, yet. A >>good test like the WM test can tell if it's a patzer or a potential top engine, >>or what's different from the previous version of that engine... >> >>>_All_ the programmers I could read say more or less frankly that they can't work >>>with _that_ test (100 positions). Because, surprise, to know a ranking place in >>>that test or in other position tests, has no importance for their programming. >> >>Surprise: Computerchess testsuites aren't intended only for the use by chess >>programmers. Acutally they are intended mainly to be used by fans, chess >>players, common program users, to be able to investigate the strength profile >>(strengths and weaknesses) of chess programs, find estimated rankings when they >>want to... > >Hi Steve, > >do not get me wrong; I am not against you at all. > >I will try to let people understand why the programmers are not interested in >these test suite. > >> >>Since some chess programmers have said that they aren't interested much in such >>tests, this seems to be your main argument against it. But this argument is not >>valid, because tests are made for thousands of users and fans (who do not use >>tests to develope, but to TEST), and not as a developing tool for programmers. > >OK, you may have made the best test set and I think chess funs will find it >quite interesting to see if their latest chess program does perform well in this >test set. > >This is very nice tool and we all must thank you for this, but it is not >reliable (unfortunately) to estimate a program strenght. > >We have seen quite often; nearly all the time, that to modify a chess engine to >play better in those tests set a drawback. I mean that most of the time a >version of program X is better than another version of the same program >performing better in that tests set. >This means that in order to make a program stronger other things are more >important. > >In reality this is explained if you consider the following: > >1. To find the best move which allowes you to win in 30 moves instead of 60 >moves does not bring you any Elo rating at all. >2. To be able to play some !! moves and many ? moves does make the program >weaker as with 2 ? moves one quite probably will lose the game while with some >!! it may not be able to win. > >This means that the a chess program should be made overall stronger and not be >able to solve some specific positions. > >So summarizing if one program is performing better in the test set could be >stronger, but not necessarely; most of the time it is not. > >This is why the chess programmers do not rely on these test sets. > >I am not saying that it is not possible to make a test set that can help to >reach what you are looking at, but probably this must be quite different and >with a huge no. of positions covering other issues as well. > >> >>I'm sure it gives you BIG TROUBLE that the usually top-listed engines from >>gamebased rankings (Shredder, Fritz...) are also top in the WM test's results, >>while engines which are playing weak compared to these, are also ranking bad >>there :-))) It just works! Do you have sleepless nights now? >> >>Steve > >Sandro Good try Sandro, but I fear you're wasting your breath. Shredder and Fritz do better on these tests than do PatzerChess. Q.E.D. Dan H.
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.