Author: Uri Blass
Date: 11:14:13 07/09/04
Go up one level in this thread
On July 09, 2004 at 13:46:54, Fabien Letouzey wrote: >On July 09, 2004 at 13:31:43, Dan Honeycutt wrote: > >>Hi Uri: > >>It appears you made several mistakes. Just days before the tournament you were >>ordering new hardware and trying to improve the playing strength with some >>evaluation from fruit which, pardon the pun, bore no fruit. It appears you >>should have been deciding how you were going to operate movei, how to manage >>time, which interface to use, which book line to play etc. - all with engine and >>hardware in a set, stable state. > >Not to mention that Fruit's evaluation function is one of the worst parts of the >engine! I tried to warn him ... but failed. > >Fabien. I think that fruit's evaluation is good but it is possible that I simply did not implement things correctly in the last days when I tried to evaluate backward pawns or did not try to evaluate the right evaluation terms of fruit. There are still some things that worked that I took from Fruit's code in the previous days: 1)Public Movei never evaluated passed pawns as weak pawns and it seems that evaluating isolated passed pawns as weak pawns helped. 2)I believe that some better piece square table for the king(based on fruit piece square table) including calculating average between king opening table and king endgame table also helped(I did it slightly different than fruit and centralizing the king has the same value in rook endgames or bishop endgames and the transition ends when sides have slightly less than a queen and it starts only when the sides lose enough material relative to the opening position. The table for side is also dependent only on the opponent pieces. 3)What I tried in the last days before the tournament did not help but during the tournament I decided to reduce the panelty of doublepawns from 0.2 pawns to 0.15 pawns(I am not sure if the change is productive but it scored slightly better against previous version and it is also moresimiliar to other programs that even punish less double pawns) I did not try to implement candidate passers and I may try it later. 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.