Author: Uri Blass
Date: 22:42:25 02/19/04
Go up one level in this thread
On February 19, 2004 at 18:37:36, Thomas Mayer wrote: >Hi Uri, > >> It seems faster than what I thought to do and the only question is if it is >> faster than what I have. > >you will never know without trying it out. Believe me, I did try a LOT of things >which were senseless later. At the moment I have two 32 bit variables for the >hashkey -> seems to be faster for me then one 64 bit... for whatever reason. >Also I have very complicate incremental stuff for some pawnstructures. Nowadays >I am quite unsure if it really pays off. Once I did try to transfer the >pawnstructure to Bitboard... well, it was working, but dead slow. Anyway I might >give it another try in the future. You know I have some new ideas. >Also I have incremental pawn-attacktables. Nowadays I am also unsure if this the >stone of wisdom. Still I think Quark has some lack of speed and I work >permanently to increase it. There are two fields to speed up: a) increasing the >speed of the code b) shrinking the tree... Both help a lot and b) is not >measureable in nps. I think that b is more important. E.g. for the future I plan incremental attacktables and >maybe also incremental moves. (So no move generator anymore at all - it will >have the moves always handy because it updates them incremental) If this is >worth something I have no idea -> and I will never know as long I do not try it >out... E.g. when I understood Eds pages correctly his attacktables for move >ordering speeds him up about a factor of 3.2 when comparing nodes to depth. I have some attack tables that are updated incrementally but I decided that it is possible that attack tables are a bad idea and it is better to calculate everything by function. I do not plan to do a rewrite to the complicated stuff that I have because I do not like to do a rewrite and there are other things that probably can help more. Movei has a lot of lack of knowledge in the evaluation(for example it still does not evaluate knight outposts) I also believe that move ordering can be significantly improved and one of the things that I need to do and still did not do is rewriting my alphabeta. 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.