Author: Bas Hamstra
Date: 16:52:53 08/17/04
Go up one level in this thread
>>How about dumping that futility stuff? Just cut away all captures with SEE < 0 >>that's good enough. If it doesn't solve more tactics, than why use it? If it >>*really* searched 0.20 ply deeper it oughta solve more tactics... I don't like >>it and I have never liked it and IMO it doesn't combine well with checks in the >>qsearch. >> > >I agree. But your program sounds more impressive when you can list all >those neat features. :-) Just kidding but it's not far from the truth. >I hear about XYZ program with every whiz-bang feature and I really start >to wonder about the methodology used to verify each one! I think most amateurs are working for years to get it right. Critical stuff needs carefull verifying. For instance I checked my SEE code with a second recursive approach (using a sort of mini-qsearch with all captures to one square and MakeMove()) and only when I find no differences I trust the code. Same for hashing: compute a hashcode from scratch at every node and compare with the incremental value. Same for out-of-check routine, check it against the regular (pseudo legal) movegenerator. etc etc. You can only trust verified code, at least in my case there's ALWAYS something I overlooked and the doublecheck will show it... Bas.
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.