Author: Frank Phillips
Date: 04:59:43 04/30/04
Go up one level in this thread
On April 30, 2004 at 07:11:02, rasjid chan wrote: > > But still as others posted, and from theory of alpha-beta that it relies > only on one signal, ie eval(), evaluation determines how alpha-beta >searches and the returned best-move. Sure, chess is a zero sum game with only three outcomes: 1, 0 or ½. So, unless the search sees to the end of the game then an evaluation is needed to chose an end point. I suppose, in fact, to differentiate between all endpoints that are not won, lost or drawn. I view the evaluation (and chess theory) as a predictor of the likely outcome of a position for which the end result is unknown. Sometimes the prediction will be wrong, even with best play. Presumably due to tactics or wrong or inapplicable theory or the evaluation function being a poor predictor in the specific case. The question then is how much of an evaluation do you need to do well against the opponents you are likely to play eg to take you to a particular level. There will always be occasions where the presence of specific knowledge hurts (over-valuing some positional factors that are not as important as others in specific positions or vice versa) as will the abscene of specific knowledge in particular circumstances. In theory search or egtbs (the same as search but pre-computed backwards) must dominate all else. Fortunately chess is still interesting because from a practical point of view it cannot. I doubt there is single set of killer positional elements and their weighting factors. Nor a single way to arrive at a decent set. And I would repeat that on a modern cpu a standard search plus any sensible evaluation function should take you way above 1900.
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.