Author: Miguel A. Ballicora
Date: 12:16:33 01/20/02
Go up one level in this thread
On January 20, 2002 at 14:36:44, Uri Blass wrote: >On January 20, 2002 at 10:25:50, David Rasmussen wrote: > >>On January 20, 2002 at 10:12:28, Uri Blass wrote: >> >>>On January 20, 2002 at 09:41:48, David Rasmussen wrote: >>> >>>>On January 20, 2002 at 09:29:24, Severi Salminen wrote: >>>> >>>>>On January 20, 2002 at 09:01:50, David Rasmussen wrote: >>>>> >>>>>>There must be a value system of material that takes care of all special cases. >>>>>> >>>>>>1,3,3,5,9: >>>>>> >>>>>>Has the following problems: >>>>>>3 pawns for bishop or knight is almost always a bad idea. >>>>>>2 knights/bishops for rook and pawn is almost always a bad idea. >>>>>>2 rooks for queen is often not a good idea. >>>>>>3 knights/bishops for a queens is often not a good idea. Then again, often it is >>>>>>:) >>>>>> >>>>>>What is your best bet? >>>>> >>>>>And sometimes a Bishop is better than a knight. So: >>>>> >>>>>P=1, B>N>3*P, R+P>2*B, Q>2*R and Q>3*B. >>>>> >>>>>So maybe P=1, N=3.2, B=3.4, R=6, Q=13? >>>>> >>>>>Severi >>>> >>>>I don't want to score bishop higher than knight. It depends on dynamic factors >>>>that should be in evaluation anyway. I just want to avoid extra code to evaluate >>>>special cases, as Crafty does. I think it is possible. >>>> >>>>/David >>> >>>There is no answer for your questions. >>> >>>It is even not possible to know what is the meaning of P=1 >>>because the value of the pawn is dependent in the square of the pawn. >>> >>>If you do not use piece square table for pawns then your program may be weak >>> >>>If you use piece square table for pawns than P=1 is eqvivalent to P=1.2 when you >>>only change the piece square table for the pawns. >>> >>>Uri >> >>Look, I am not talking about perfect evaluation here. I am talking about a >>mature evaluation function, but instead of evaluating material special cases >>(three pawns for a bishope etc.) as crafty does, I think all of these _material_ >>special cases can be done with material values alone. All the other evaluation >>stuff will still be there. >> >>/David > >but it is still not clear because the other evaluation stuff is important. > >If you have pawn value of 0.8 and >piece square table that says: >0 for 2th rank >0.05 for 3th rank >0.10 for 4th rank >0.20 for 5th rank >0.40 for 6th rank >0.70 for 7th rank > >It is the same as having pawn value of 1 and piece square table of >-0.2 for the 2th rank >-0.15 for the 3th rank >-0.1 for the 4th rank >0 for the 5th rank >0.2 for the 6th rank >0.5 for the 7th rank > >Uri Exactly. In fact, a bishop could have a value that is the same as the knight but in the middle of the tree could have a "higher" value because it gets bonus for mobility. You can have a bishop at 3.5 and mobility bonus can go from -0.5 to 0.5 and there is no difference with a scheme where the bishop is 3.0 and mobility goes from 0.0 to 1.0. It all depends on whether you add penalties and bonuses. It only affects the quick "material balance" that you use for lazy evaluation, pruning in quick search etc. Otherwise, it would be possible to write a program without any value for the pieces. However, for those quick material balances is not that important if a bishop is 3.0 or 3.5. They key of the matter, IMHO, is to determine the right value of the piece for each critical moment of the game. That is, its "average value" +- the positional value. One is not independent of the other. 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.