Author: Tim Foden
Date: 23:26:08 04/05/04
Go up one level in this thread
On April 05, 2004 at 17:09:29, Dieter Buerssner wrote: >On April 05, 2004 at 10:51:06, Tim Foden wrote: > >>I've used multi-variate linear regression to generate a set of evaluation >>values, but I have some bugs in the implementation (as pointed out by Dave >>Gomboc when I spoke to him in Graz), and I haven't tested the values at all. I >>still plan to pursue this some more when I have the time and the inclination. > >How are you doing the fitting technically. As I understand it, the values are >not linear independent. For example, you would get the same result, if you gave >knight material 0.5 less, and all knight mobility values 0.5 more. If you look very closely, one of the mobility values is missing. This means that the equivalence is not there. So far this is the best solution to the problem that I've found. I've got an idea for a slightly better way, but I haven't implemented it yet. >If I recall, how this was done, in one step there is an inversion (half way) of >a matrix. Yes, there is an inversion. I use the "sweepout" method that inverts and solves at the same time. I'm not necessarily doing this in the order that would be best for absolutely accurate values though. This was just a test to see if there would be any use in doing it at all. My conclusion was that there may be... but I have to think about the best way... so here I am 1.5 years later still thinking about it! :) >Using the Gauss Jordan method, this will leave you with a zero pivot element. I purposely leave out the pivot (or offset) element. I can add it back in easily if I change my mind. >Or are you using some more sophisticated methods like SVD decomposition? > >> "knight material value" 3.567 3.1712 3.1393 3.0996 >> "bishop material value" 4.1172 3.533 3.3734 3.2835 > >> "queen material value" 7.1422 9.5802 9.4663 9.1281 > ^^^^^^ > >Can this be correct? Q less than N+B in he opening. This is a good point. I never claimed that the values were correct. :) This is one of the questionable ones. I forgot to discount the result value for positions further away from the final result, and this may be because of that. Cheers, Tim.
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.