Author: Matthew Hull
Date: 13:58:27 12/09/04
Go up one level in this thread
On December 09, 2004 at 16:44:24, Peter Skinner wrote: >On December 09, 2004 at 16:02:10, Matthew Hull wrote: > >>These were all 30 0 games played on a single Duron 900mhz machine, so 450mhz >>effectively for each program. These would be blitz games on more up-to-date >>hardware >> >>Learning on. >>Pondering on. >>hash 96m >>hashp 24m >>book width 99 >> >> >> >>xboard: Match Crafty-19.16 vs. Crafty-19.17: final score 24-23-53 (equal) >>xboard: Match Crafty-19.13 vs. Crafty-19.17: final score 43-13-44 (wipe out) >>xboard: Match Crafty-19.13 vs. Crafty-19.15: final score 22-20-58 (equal) >> >> >>Crafty's current ICC rating drop in blitz coincides with the advent of 19.16/17. >> This seems to agree with other results that put pre-1916 versions as stronger. >> >>I think something must be seriously out of whack with 19.17, at least against >>computers and at short time controls. > >Others state the same and while I suspect a bug in the search of Crafty, I can >not pin point it. I have asked Robert though. > >Despite all that, I have the best results with 19.17. I think the only difference is in eval wieghts, rather than a coding bug. You should try 19.17 against 19.15 or 19.13. That bout was really one-sided. Better yet, do a 30 0 match between 19.17 and a commercial prog. Then run 19.15 against the same prog. Start each match with learning cleared, but on, and pondering on. They can run on the same machine. See which versions makes the most progress against the commercial. I think I might do that with Crafty and Phalanx. Both will beat phalanx, but one will beat it worse than the other, I'm thinking, just to verify the result. > >Peter
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.