Author: Uri Blass
Date: 03:24:27 09/10/05
Go up one level in this thread
On September 10, 2005 at 04:11:57, Steve Glanzfeld wrote: >On September 09, 2005 at 21:02:36, Mig Greengard wrote: > >>Convinced of what? It's a goal, not a conclusion, and there aren't any examples >>yet that I'm aware of. If a human-like training partner can be devised that, as >>a consequence, also plays a little weaker against other computers, that would be >>a worthy achievement. > >Yes, but the experience actually is: more useful, effective knowledge doesn't >make a program slower. It will lead to more cutoffs in the search tree. The >bigger eval function will need more time sometimes* but as a tradeoff, the >engine may be equally quick or even quicker. Fritz 9 has more knowledge AND is >tactically stronger than Fritz 5.32. > >*) Positional and other sophisticated evaluation functions are not used when the >material balance is already deceisive. When you're up a rook, you don't care >much if your bishop is good or bad. Since most possible continuations (nowadays >9+ plies as a typical minimum, +extension) are absolutely crazy, these are the >majority of positions which have to be evaluated, and the "knowledge burden" >will affect only a few. No crazy continuations get pruned and it is not correct that the majority of the positions that get evaluated are positions when one side is at least a rook up. 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.