Author: Dan Andersson
Date: 04:04:46 05/06/02
Go up one level in this thread
>>Naturally there is also a good deal of chess "knowledge" in a good search, but >>it is really different from what most people generally consider to be >>"knowledge". > >These statements somewhat amaze me. I would have thought the most important >thing for a chess engine (given an already decent search) is to have a good >positional evaluation. > Given that piece-square programs can play excellent chess. You would do better to improve that program by giving the search a better shape than increasing the evaluator. F.ex. a well tuned extension scheme giving a 50% larger tree would in most cases beat an evaluation giving the sane slowdown. I believe that a program 'style' and 'chess knowledge' come from the interplay between search and evaluation and is reinforced by the feedback gained by the programmer when he evaluates the resulting play. >Not good in terms of quantity, but good in the terms of being seldom wrong >by a large amount. > All evaluators are far from perfect. And in many cases litte better than material+random. A lot of noise in there. Heck you can play chesslike by adding a random number for each leaf node and propagating that to the eavluation. That would be similar to mobility. And if you have an advance search you might steer the search by assigning large heuristic values. MvH Dan Andersson
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.