Author: Gian-Carlo Pascutto
Date: 10:08:16 04/15/02
Go up one level in this thread
On April 14, 2002 at 20:00:55, Christophe Theron wrote: >1) I do not know how to "tune" my program to make it better against a given >opponent. I do not know how to "tune" for a given time control either. "Tuning" >is something like an urban legend in computer chess. People have no idea of the >difficulty to do this, so when they need to explain something strange they will >say that some "tuning" has been done. People cannot admit that they do not know, >so they'll use whatever obscure explanation they have at their disposal. >"tuning" and "will be better at longer time controls" are great classics in this >area. I'll give an example. My program has some areas of the evaluation that have a certain weight and are somewhat independent. This could be kingsafety, pawn structure, etc... I have to find a balance between those. I did so by choosing a few candidates on intuition that I thought would probably be the best, and let them play >2000 games against a wide variety of amateur and professional programs. I pick the settings with the best performance. I could do the same again, but only play one opponent at a time, and remember which settings have the best score. I have no idea whether the different settings would cause singnificantly different performances vs certain opponents, though. But I will certainly try this if I play another tournament. >If I had more computers at my disposal I would not use them to play against my >top competitors. I find this an interesting statement. Since one only fixes weaknesses when they are repeatedly exposed, one would want an opponent that is able to pound them often. I would expect that your top competitors are best at this? -- GCP
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.