Author: blass uri
Date: 20:26:08 03/16/00
Go up one level in this thread
On March 16, 2000 at 20:25:02, G. R. Morton wrote: >On March 16, 2000 at 17:20:19, blass uri wrote: > >>On March 16, 2000 at 16:54:08, José de Jesús García Ruvalcaba wrote: >> >>>On March 16, 2000 at 15:59:24, G. R. Morton wrote: >>> >>>>On March 16, 2000 at 14:59:48, Dann Corbit wrote: >>>> >>>>>On March 16, 2000 at 13:49:28, G. R. Morton wrote: >>>>> >>>>>>If really true, can someone try to explain how can this can be with its highest >>>>>>rating against computers? Enrique Irazoqui at >>>>>>http://www.computerschach.de/tourn/cad00.htm calinmed that Junior6 showed >>>>>>superior positional understanding in actual play. Can someone enlighten? >>>>> >>>>>Test suites are just one measure of a program's ability to play. >>>>> >>>>>Little Goliath is a killer in test suites, but does not play as well as the very >>>>>best chess engines (but it's no slouch either). Rebel can be tuned to play >>>>>strongly or to solve chess test suites strongly, and the settings are very >>>>>different. >>>> >>>>You are suggesting that a program can be very good at both tactical & positional >>>>test suites but mediocre at play (& vice-versa) but not saying how this can be. >>> >>> Playing chess and solving testsuites are different tasks, and different >>>approaches give best results for each of them. >>> >>>>But if this is true it is very puzzling since every move of a computer game can >>>>be thought of the software s solution to a tactical or positional test. >>> >>> Not quite. Most positions seen in a game do not have a key move that solves it. >>> >>>>There >>>>should be a very strong correlation one would think. >>> >>> There is no correlation between playing strenght and performance at testsuites. >>> >>>>As a comparison, people >>>>who score very high on I.Q. and SAT tests are nor just good on such tests >>>>which is, of course, why these tests are given. Their strong correlation with >>>>high performance or success in many other activities are well noted (see The >>>>Bell Curve book for instance). >>> >>> I think this comparison is not valid. Human beings are very different from >>>chess engines. >>> >>>>Did not Larry Kauffman once claim that his test >>>>suites could be used to fairly accurately the software s rating? Anyway the >>>>interesting question of how remins. >>> >>> There are some known trick to improve performance at testsuites which do not >>>increase playing strenght. Reducing the pawn value is one of them, I think. >>>José. >> >>Reducing the piece value in general is one of them but reducing the pawn value >>is not one of them. >> >>One of the changes in Junior6 relative to Junior5 is reducing the pawn value and >>Junior6 has problems in finding right sacrifices of a piece for pawns because of >>this reason. >> >>Uri > >Thanks, I think you & Mr Ruvalcaba have made some perceptive points. I >especially find interesting your theory (posted in reply to me elsewhere)that >these tests are not really very good because we really don’t have all that good >an understanding of chess, especially, I would think, for devising good >positional test suites. Would you say, then, that the top rated Junior 6a is >probably the best positional player (as Enrique might claim) but the positional >tests, say, on which Rebel 10B outperformed Junior6 were not adequate to >indicate this? It is possible. It is also possible that there are better positional programs releative to Junior that are weaker in tactics relative to Junior. I do not know Uri
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.