Author: Charles Unruh
Date: 05:48:13 10/21/99
Go up one level in this thread
On October 20, 1999 at 14:23:22, Roger D Davis wrote: >About what percentage of agreement was there between chessmaster with its >chessmaster personality and Ferret at the beginning? This is an important >baseline. > >If it was 95%, then getting it up to 98% isn't that impressive. But if it was >80%, then 98% is impressive, I would think. > >Also, are the 2% of moves that Ferret and the Ferret personality disagree >blunders by the Ferret personality? How are these moves evaluated by Crafty, for >example? It's really been 99.9% over the 70 games only 2 or 3 different moves over the 70 games, but i know CM can't replicate some things in the endgame such as B and N mate so i say 98%. I'm sure it could even be tweaked better than i have tweaked it. > >There is also a "cross-validation" issue. The parameters you have for your >Ferret personality now are optimized for the 70 Ferret games used. If you >applied your Ferret personality to another sample of Ferret games, however, the >percent of agreement can be expected to drop. How much is an empirical question. >It might be significant, and it might not be. Would be interesting to apply your >Ferret personality to another "random" sample of Ferret games, and to a sample >of Ferret games with offbeat openings, so that you can get offbeat positions in >which disagreements would arise, thus allowing you to optimize further. > >I would think there would be much interest in this, since Ferret is not a >released program. > >Roger > > > > > >On October 20, 1999 at 12:23:19, Charles Unruh wrote: > >>On October 20, 1999 at 12:14:13, Roger D Davis wrote: >> >>>Interesting. Did you have a methodology for making the adjustments, or was it by >>>hand? by hand >> >> What are the time controls at which Ferret's moves were reproduced? >> >>I used timestamped ferret games, and made sure that CM comes up with the move >>+/- 3seconds of ferrets and then replay the game at the same time control, to >>make sure the game is an exact duplicate. >> What >>>does the percentage drop to if you exclude opening moves and forced moved? >> >>I only use moves from when the progs are out of book. >> >> >>> >>> >>> >>>On October 20, 1999 at 10:57:44, Charles Unruh wrote: >>> >>>>I took over 70 games that ferret played, then i kept adjusting chessmasters >>>>settings until it plays every move that ferret plays, with 98% accuracy. I'll >>>>trade the personality with someone who creates another clone.
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.