Author: Robert Hyatt
Date: 12:36:11 03/07/06
Go up one level in this thread
On March 07, 2006 at 03:49:47, Richard Pijl wrote: >>If one's results do not rotate approximately as described >>for the four positions and you say the evaluation is an >>issue, what kinds of evaluation issues have you seen that >>could explain it?!? > >I had this problem long time ago in the Baron as well. To be able to detect >these issues I created a little tool that reads a large PGN file, and for every >position in the PGN file calls the static evaluation function for that position >mirror the position and call static eval again. Whenever that didn't give >exactly the same score for both positions, I print both fens, and continue. >Usually, if there is an evaluation mirroring bug, it doesn't take long for it to >find it. > >Richard. I've also had this forever it seems, in the "evtest" code in crafty. I have 'flip' (reverse board, make white pieces black and vice-versa, then copy exchange ranks 1 and 8, 2 and 7, etc. I also have a "flop" command that just exchanges files a and h, b ahd g, etc, but doesn't change the color. I then do something like: s1 = evaluate() flip s2 = evaluate() flop s3 = evaluate() flip s4 = evaluate() and the scores must match, except that after a flip the sign must change since the pieces are changing. My current code will run through several thousand FEN positions with nary a complaint since I no longer have any asymmetry whatsoever...
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.