Author: Robert Hyatt
Date: 22:11:20 06/05/02
Go up one level in this thread
On June 05, 2002 at 21:59:19, Jon Dart wrote: >I used Crafty 18.10: > > >Crafty v18.10 > >White(1): Black(1): end-game phase > clearing hash tables >note: scores are for the white side >material evaluation................. -2.00 >development......................... 0.00 >pawn evaluation..................... 1.78 >passed pawn evaluation.............. 0.00 >passed pawn race evaluation......... -5.90 >king safety evaluation.............. 0.00 >interactive piece evaluation........ 5.30 >total evaluation.................... -0.82 > >Note that the overall score is -0.82 (advantage to Black). > >I guess I scanned the eval terms a little too fast and came to the wrong >conclusion. > >Weirdly, the "passed pawn evaluation" is way biased toward Black, but the >"interactive piece evaluation" (which basically means miscellaneous eval terms) >largely cancels it out. I don't know where the +5.3 comes from. > >Still, this version seems to be doing better than later versions, and the >overall evaluation at least has White close to even. > >--Jon There is a bug. The "real evaluation" gets -.82.. the "score" command is not quite right. The PassedPawnRace() code incorrectly thinks that white can stop the black pawn because the black pawn needs 6 moves to promote (5 pawn moves plus one for the king to get out of the way). This is wrong. I am working on a fix so that it will _really_ know this is won, but there are so many exceptions with the two pawns vs the bishop... ie move the white king up a couple of squares and the black f-pawn can block the diagonal so that the h-pawn can promote first... ugly stuff to compute statically...
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.