Author: Russell Reagan
Date: 04:49:07 05/21/04
Go up one level in this thread
On May 21, 2004 at 06:10:29, Vasik Rajlich wrote: >Computers also can't calculate all of the variations. > >How about this: > >"tactical rating" == performance on tactical testsuite (this means clear >tactical problems - not ones that can be "solved" by king safety evaluation or >passed pawn evaluation) The whole idea of seperating the recognition of tactical patterns from the recognition of positional patterns doesn't make sense to me. I understand what Andrei is wanting to do, but I don't know why anyone would want to do that (no offense). All of these methods presented for finding a tactical and positional rating involve test suites, and test suites are never going to give a clear picture of ELO rating. Any amateur programmer here could tune their engine to solve every test suite perfectly, but it still isn't going to beat Shredder. Finding the best move involves both tactical knowledge and positional knowledge. Consider a simple example where you have to execute a combination that allows you to trade queens and rooks, leaving you in a king and pawn endgame where you have a pawn majority that will give you a passed pawn and allow you to win the game. The actual promotion of the pawn may be 30 moves after the combination, but it is enough to know that you can promote it eventually because of your pawn majority. Is that position a tactical position or a positional position? It's both, and it really doesn't matter how we classify it. You have to know two patterns to find the right move: the particular combination pattern used that allows you to trade off the pieces, and the pattern that allowed you to promote your pawn much later in the game. >"positional rating" == actual rating - tactical rating > >Of course it's true that the positional level of an engine does not come only >from the evaluation function. A deeper search will appear to understand more, >while wrong pruning can give the appearance that the engine understands less. > >Vas > >>Also, an ELO rating is only a relative indication of who is better. Only ELO >>differences matter. If I came up with a test that gave you a tactical and >>positional rating in ELO points, it wouldn't mean anything to people or engines >>that didn't take the test. I could have a 5000 ELO rating in some rating pool, >>but if Kasparov entered that rating pool, he would have around a 6800 ELO >>rating.
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.