Author: Miguel A. Ballicora
Date: 19:58:57 08/05/02
Go up one level in this thread
On August 05, 2002 at 17:09:23, Vincent Diepeveen wrote: >On August 05, 2002 at 16:56:41, Vincent Diepeveen wrote: > >I hope you realize that DIEP played hundreds of games versus knightcap >at the internet, and thousands in total, and that i was a lot of these >games watching. > >knightcap (the pc version) was a very popular opponent to beat when it >was at the end of its learning curve, giving away pieces for just 2 pawns >and one check. > >I have not seen a single testrun which managed to get to a better local >maximum. *always* it was playing towards a parameter set where some >parameters were at least 10 times the value it had to be. giving away >pawns and pieces for just a few checks, it's real incredible. You are assuming that your hand tuned parameters are in a local maximum (or minimum I prefer to call, it depends what you use as axis). Probably you are close to the minimum. So close that it is very difficult to realize that you could do better. In that situation, I believe that it could be possible to find a procedure that takes you to the minimum (or at least closer). If you are already in a minimum because your tuning was perfect, of course, there will be not procedure that will make you do better! Even if you are in the perfect minimum, once you introduce a new parameter you automatically are shifted. To come back, I believe, automatic tuning could be useful in this case. [snip] >>Obviously automatic tuning is only useful if it manages to create better >>local maxima when it runs longer. >> >>Does it with your protein research? I don't do this computational research (yet), but I am aware of it. Actually, if you make a simple change and iterate, you find the minimum of the valley you are already in. You might find few bumps, but those are solved increasing the "virtual" temperature (it helps you jump the little bumps). If you find that the iterations go crazy, that means that the change you did is disruptive. That is also useful information. As I said, starting from scratch would be like trying to find the lowest valley in the Andes but your starting point is the middle of the pacific ocean. :-) Regards, Miguel
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.