Author: Reinhard Scharnagl
Date: 09:03:51 01/04/04
Go up one level in this thread
Hi Bob, >After reading the thread started by this bulletin I get the impression that the >methods discussed have the primary benefit of requiring little computation and >hence a very small percentage of the microprocessor's time. [also, not very much >time required to produce the code] for me the Smirf method as shown here targets only static average piece values. Its implications for dynamic detail evaluation I have not yet described at my homesite. But as far as I can see, this is really cost intensive, sorry. >There are many positions cited and discussed in the printed chess literature >which show radically different piece valuations applicable to only to the >specific position being discussed. As a trivial example, if it is mate in one >and the final checking move is to be made by a pawn, then the value of that pawn >is infinite. There have some levels of evaluating to be distinguished. Tactical combinations as e.g. matings have nothing to do with average piece values. >It is not clear to me that the simplified model being used is optimal in terms >of amount of benefit versus cost. Perhaps a more complex but more accurate >model would produce higher engine performance. I don't know, of course. I prefer better (and costly) evaluated knots. But a evaluation method should be based on verifyable assumptions. That is what I am still missing. Therefore I try to describe a method which could stand practice tests. And I have begun with average piece values. Here Capablanca's Chess is a good testing field, because the values of the two fairychess pieces are not fixed by a centuries old tradition. Tests can show, which basic method seems to fit better. Regards, Reinhard.
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.