Author: José de Jesús García Ruvalcaba
Date: 16:00:12 10/09/99
I know I can also contact SSDF by e-mail; but some SSDF members read this forum and my post can be of interest to the computer-chess community, and it might be adopted by other testers, too. My suggestion is to play the matches with learning programs without restoring the original book weights for each match. I.e. if some tester has played one match with program X on computer Z, leave whatever X has learnt for the next match. And for the third match leave whatever X has learnt in the first two matches, and so on. This can be taken to the point that all the testers who are playing X on the same hardware level share their learning files on a regular basis, if the X has this feature. However, I recommend to use a fresh book for the first match of X on the next hardware level Z+1, as it is a different entry in the rating list. The purpose is to balance the positive effects of book learning on X's score with some negative effects produced by the same book learning which are only apparent when playing many opponents (they do not show up when playing against a single opponent). José.
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.