Author: scott farrell
Date: 19:57:14 01/17/03
Go up one level in this thread
On January 17, 2003 at 16:16:50, Uri Blass wrote: >On January 17, 2003 at 12:47:39, scott farrell wrote: > >>doing a perft 5 from the starting position it takes me 26 seconds, about >>175Knps. >> >>This is just genmoves/escapeCheck/make/unmake/check for legal. >> >>I do incremental piece squares and isCheck() >> >>and I only get 175Knps !!! with no eval !!! >> >>Scott > >I am faster in perft mainly thanks to the fact that I do not need to do make and >unmake the last ply. > >I generate all the moves after every make moves and it is one of the reason for >the fact that I do not have a fast engine in nps. > >Another reason is my slow function to generate all the possible checks in the >first plies of the qsearch. > >I simply check for every legal move that I generated if it is a check and there >are faster ways to do it. > >At this moment I try to fix another bad design of my program that is not about >king safety but fixing this design may help me to evaluate king safety. > >The problem is that Movei does not evaluate correctly if it is in endgame. >I have a varaible that gives me the value of all pieces(not pawns) and movei of >today decides that it is in endgame only if this number is not bigger than 24. > >I can add about your second post about pawn storms that I do not hash nothing >except the position and I do not plan to hash pawn structure in the near future >because Movei is not significantly slower without hash tables for pawn >structure. > Pawn hashing is a win win. It is safe and fast. You can add any amount of code for pawns in your eval function, and the speed does not change ever. The pawn hash gets a 99% hit rate mostly with a very small hash table, so you can have the most stupendous pawn eval, and it doesnt slow down at all. It is pretty simple, and not really prone to bugs. >I can evaluate protected passed pawns with price of maybe 5% in speed and 5% is >not very much. > >I may add also some evaluation for passed pawns that are not protected in part >of the cases for CCT5 but not much more than it and isolated pawns are not going >to be evaluated in CCT5. > >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.