Author: Steve Maughan
Date: 05:30:48 11/26/01
Go up one level in this thread
Ed, >One recent example: Century 4 eval is more speculative than previous versions. >Why? Because I believe (confirmed by my test results) that the program can >handle it because of the mixture of smart search and fast hardware. A deep >search simply filters out most of the "too speculative errors" of eval and >the net result is a better, even more attractive chess engine. Interesting. One similar thing I've got on my to-do list is to look at chnaging the size of the positional terms based on search depth. As an example, suppose you have a routine for trapped pieces. If you detect a trapped or semi trapped piece at ply=2 then it would seem silly to give it as big a positional penalty as if you found it at ply=15. In other words if you find it at ply=15 and it is significant enough to make it into the PV then it probably is really trapped, however if you are at ply=2 there is a real chance that the peice may not be trapped at all - hence the evaulation is *really* speculative. It's easy to see this with something like a trapped piece but can one extrapolate it to *all* positional terms? I don't know I haven't tinkered yet. One other question - at blitz or bullet chess was the 'Old' Rebel with the extra horizon knowledge still the same strength as the 'New' Rebel without the horizon knowledge? Regards, Steve Maughan
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.