Author: Sandro Necchi
Date: 07:00:54 09/14/04
Go up one level in this thread
On September 13, 2004 at 17:52:33, Sune Fischer wrote: >On September 13, 2004 at 16:38:56, Sandro Necchi wrote: > >>OK, I have a question: >> >>No suppose you test program A with book A against program X with book X and you >>get a very good score. >> >>Do you call this a proof that both the program A and the book A are good? > >I'd say the total "A" combination is better than "X" combination. >I can't say anymore without further experiments (or studying the games). This is in contrast with what you stated. I only wanted to point out that results may be meaningless if not supported by game analysis. Now, you say the same. > >>If yes, then you test program A with book A against program B with book B and >>the score is still very good! >> >>Then this is a further proof...correct? > >I guess my answer was "no", so... > >>Now you test against program C with book C and program A get killed! >> >>Then you show the games to a very strong player which is friend of yours and he >>immediately tells you that book A and program A are weak because they play weak >>and the reason it won against program X and B is because they did not reply >>correctly to weak moves... > >Sure the objective truth is hard to find, it's just a relative measurement. OK, so you agree with me that the score itself is not necessary a "proof". > >Just because an 1800 player beats a 1400 player doesn't mean that the 1800 >player is perfect. OK!!!! Finally! > >>At this point you find out that the only TRUE finding is that program A with >>book A are weak and can score good only against WEAK opponents... > >"Weak" is again relative, it was stronger than X and B. OK, but statistically it was scoring well...so now you agree... > >>Do you still call this "only sure proof"? > >The strong player did not really tell me anything the data hadn't already shown >me. Wrong. The player explain why that happened. > >At best a strong player and his analysis can speed up development, but >ultimately I will rely on testing it anyway. >Strong players might know a lot about chess, but they do not always understand >what goes on inside of a search tree. Well, it depends on what we mean strong player...to me it is above 2600 Elo...do you really think the same? > >>>Every change however logical it may seem must be tested, but in looking for >>>explicit features to add to the evaluation I think you are right that games >>>should be analyzed for inspiration. >> >>Analyse games seriously is the only way to draw some logic conclusions...believe >>me! > >I don't think it is the only way, I know some programmers who aren't very strong >players have also made good programs. OK, this is true, but if a very strong chess player (>= 2700) would be a very good programmer too (a top one), than I believe it could create a stronger program... Only because this did not happen yet, it does not mean that it will never happen... > >-S. Sandro
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.