Author: Martin Giepmans
Date: 17:29:24 03/31/03
Go up one level in this thread
On March 31, 2003 at 17:46:40, Uri Blass wrote: >On March 31, 2003 at 15:20:35, Martin Giepmans wrote: > >>On March 31, 2003 at 14:35:48, Uri Blass wrote: >> >>>Today latest movei evaluates passed pawns and protected passed pawns based on >>>the ranks of the pawn(I will later change it to evaluate them >>>based on other factors like square of the kings). >>> >>>The problem is that the order of evaluation is not correct. >>> >>>If I have a pawn at b5 and another passed pawn at c5 then >>>I want to evaluate them as better than pawn at b4 and >>>protected passed pawn at c5. >>> >>>I thought to give a bonus for passed pawn that has pawn in the left side of it >>>but the problem is that in that case the evaluation is not >>>symmetric and in the folowing diagram if the black pawn is in c6 and not f6 >>>I get different result. >> >>I think giving bonus/2 if there is a pawn on the right OR on the left side >>would do the trick. It's symmetrical. >> >>> >>> >>>[D]8/7k/5p2/8/3PP3/8/8/K7 w - - 0 1 >>> >>> >>>Another problem is what to do when there are 3 passed pawns. >>> >>>[D]8/7k/8/8/3PPP2/8/8/K7 w - - 0 1 >>> >>>In that case even if I decide to give only one bonus for pair >>>I may get here 2 bonuses and after the central pawn progress >>>one square forward I will have only one bonus. >>> >>>I wonder how other programs solve that problem. >>> >>>protected passed pawns are good but other things are also good and the problem >>>is how to order the things in the right order. >>> >>>Uri >> >>In my engine I give a bonus for a passed pawn that is >>(a) protected by a pawn (left or right) >>or >>(b) has a pawn on the left or on the right >> >>If both are not true (pawn is not protected and has no neighbor) I give a >>small penalty. The penalty is so small that it will not hinder the advance >>of the passed pawns. But it helps the engine to keep it's pawns "together". >> >>Cheers, >>Martin > >Today I have a small bonus for passed pawns in all cases based on the rank of >the pawn. > >I do not think that it is a good idea to have a panelty for passed pawns >even if they have no neighbour becuase there are cases when they are dangerous. True, but the danger can be detected by other means (distance to opponent king, attack/defense situation of promotion square etc) > >2 passed pawns at a and h file may be very dangerous and beat easily 3 connected >pawns c,d,e in a pawn endgame. I agree. In my engine I use extra evaluations for those cases. There are many complicating factors and it's probably impossible to do it right. I think the fundamental question is: what to evaluate and what to leave to search? And: is it possible to speed up the search? > >I know that there are cases when they are weak so I decided to have no bonus and >no panelty for them in my previous version but I found based on games that a >small bonus with no panelty seems to be slightly better than nothing > >I may continue later by adding a panelty to cancel the bonus in part of the >cases. > >I do not do big changes but one change and after it testing and the question >is how to evaluate passed pawns in a way that better will evaluated as bigger. > >giving 1/2 bonus for every pawn that is in a pair is not a good idea because 3 >connected pawns on the same rank may get 4.5 bonuses (1 for every one of them >for the fact that it is a passed pawn and 3/2 bonuses for the fact that they are >in pair). > >The problem is that if the central pawn advance you have only 4 bonuses(1 for >everyone and 1 for protected passed pawn). > >You can claim that 3 connected passed pawns are not very common but I still do >not like the evaluation. > >Uri
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.