Author: Odd Gunnar Malin
Date: 18:30:27 05/20/04
Go up one level in this thread
On May 20, 2004 at 16:17:54, Günther Simon wrote: >On May 20, 2004 at 15:22:07, Andrei P wrote: > >>On May 20, 2004 at 14:34:36, Matthias Gemuh wrote: >> >>>On May 20, 2004 at 13:12:50, Andrei P wrote: >>> >>>>how to get positional and tactical rating (preferably in elo points) for a given >>>>engine? is there a table with this kind of data for a large number of engines? >>> >>> >>> >>> >>>In a game without major blunders, a tactical shot is the result of a positional >>>preparation. How then do you want to judge "positional vs.tactical" from end >>>results ? Note that good positional play without some tactical exploitation >>>will at most draw games. >>> >>>/Matthias. >> >>ok, it is impossible to rate the engine based on the whole game. what about test >>suites? if one takes a positional test suite and calibrates it by checking >>againt human performance, then we have a way to assign "positional rating" to >>the engines. of course, the "positional rating" we get is somewhat arbitrary but >>should be a good measure of engine positional strength. the same approach can >>provide a "tactical rating" > >There is a fundamental problem in this. You will find dozens of tactical >test suites and no real positional test suites at least for programs. >The reason is that tactical problems can be deep but calculable. >Positional play demands a kind of 'plan' for a much deeper amount of >moves to reach a certain goal and it's not only about finding a few >(right) moves without the right idea. >I guess we need some really well played and commented games and let >the program find a lot of positional good moves in each of those. >Difficult anyway... > >Günther There don't need to be too much difference between tactical and positional test suites. Boths includes calculation of variations. In a positional test suite the end position isn't a gain of material but a positional clear advantage like bad bishop weak pawns etc. As I understand one good 'positional' player like Karpov use very basic knowledge to find his 'Karpovian' moves (knight outpost etc.). I'm not sure such a test suite would be any good for computer test, what would be nice instead was a sort of 'mate in one' like test suite with many positions. Since chess is a game there isn't a need to find the 'scientific' correct move, better is to find the move(s) that GM's etc. think give the best chances. Odd Gunnar
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.