Author: Marc van Hal
Date: 13:07:47 08/15/02
Go up one level in this thread
On August 14, 2002 at 23:53:57, Robert Hyatt wrote: >On August 14, 2002 at 06:23:31, Marc van Hal wrote: > >>On August 13, 2002 at 16:09:53, Uri Blass wrote: >> >>>On August 13, 2002 at 15:18:04, Vincent Diepeveen wrote: >>> >>>>On August 13, 2002 at 07:23:38, Uri Blass wrote: >>>> >>>>>My definition for a sacrifice or blunder >>>>>is a move that lose material based on >>>>>the depth that programs can see. >>>>> >>>>>The definition of losing material is based on >>>>>the material values 1,3,3,5,9. >>>> >>>>with all respect but your table is outdated in advance. >>>> >>>> a) 2 rooks are weaker than a queen in 99.9% of all cases >>>> the computer sees 2 rooks for a queen >>> >>>It is not the case based on experience with movei but I still evaluate >>>queen as almost 2 rooks(difference of less than 0.5 pawn). >>> >>>I evaluate queen as about 10 and rook as slightly more than 5. >>> >>>> b) this table indicates that giving away 2 pieces for a rook+pawn >>>> is great >>> >>>I know and I evaluate bishop or knight as about 3.5 in movei(I cannot give exact >>>value because the evaluation is based on the square of the piece and I have 64 >>>values for every piece based on the square but I still call it a sacrifice if a >>>program to give 2 pieces for rook and 2 pawns). >>> >>> >>> >>>For m >>>> c) giving away a piece for 3 pawns is great according to this >>>> table, especially if we know that one of the 3 pawns gets >>>> a passer >>> >>>I evaluate 3 pawns as less than a piece in movei. >>> >>>You also assume here that 1 is for a pawn that is not a passed pawn. >>>It is possible to assume that 1 is the average value of the pawn and in that >>>case pawns that are not passed pawns are evaluated as less than 1. >>>> >>>>So the definition is not correct! >>>> >>>>In diep queen = 11.5 pawns >>>> rook = 5.5 pawns >>>> piece = 3.6 pawns >>>> >>>>and i have special cases to catch weird cases which sometimes happen. >>>> >>>>>For a clear definition we need to compare >>>>>the evaluation of a program with mainly material evaluation before and after the >>>>>move when we give it an hour to search. >>>>> >>>>>A program with mainly material evaluation is defined to be >>>>>a program with positional scores >>>>>that are always smaller than 0.5 pawns. >>>> >>>>again a wrong assumption. Some very old >>>>programs (like some at dedicated machine) have like 2 extra >>>>rules in evaluation which always is like "if passer on 2nd row >>>>then score += 3.0" >>> >>>In that case they have big positional scores so I can say that they sacrificed. >>>I think that this bonus is not justified. >>> >>>Movei still does not evaluate passed pawns but inspite of this fact it is at >>>similiar level to some programs that evaluate passed pawns like amateur. >>> >>>I can see cases when movei wins when it seems to me that the evaluation of the >>>opponent for passed pawns is too high. >>> >>>> >>>>In short especially a few very materialistic beancounters are tuned >>>>to very high scores, whereas others with loads of chess knowledge >>>>usually are more relaxed in this respect. >>> >>>I did not say that programs with more knowledge sacrifice more material. >>> >>>Uri >> >> >>By increasing the valeu of a pawn it is more likely a piece will be sacreficed! >>Marc > > >This is the reason I gave up on trying to diddle with the piece values to >solve a known problem... such as two pieces for rook and pawn. I simply >added code to detect the known problems (two pieces better than rook and pawn, >three pieces better than queen, etc) and therefore don't worry about the bad >trades happening any longer... I also did increase the valeu of the queen for 1 personelety because otherwise it was happy with a trade of Queen for a pawn a Bischop and a Rook. and after 2 moves it lost the pawn of the trade! So 9 was after 2 moves instead of 9, 8. And strangely enough some chessprograms search deeper with an increased queen valeu. But in some cases this might be dangerous. But like Vincent says if you have an increased valeu for mobilety this problem is solved it is not likely the engine will exchange a queen against two rooks anyways. the longer time control the lesser the dangerous. But then there is an other valeu increasing the valeu of the bischop is never something I choose for For programs it is kind of useless because if you look at games where a bischop realy has a greater valeu it most of the time is in positions of the Kings Indian where a pawn is sacreficed to open the diagonal. Because All Whites pawns are on the White squares! (This is the kind of simple info a program can't find by search only by implentation of knowledge!) or in other positions Play Nc5 while a bischop is on e3 just for the power of the Black bishop ( programs never care about ugly positional positions except in this case. Because after dxc5 d5 becomes a passed pawn.) Bxc5 dxc5 looks positional ugly but Black has reached one of it's goals. the exchange of Whites queens bishop Which again only is usefull if you indeed can increase the mobilety of Black's Kings bischop (Maybe toying with mobilety valeus for a piece might help) Then again there are many lines of the Kingsindian it where the Black bischop doesn't get powerfull at all. And only it,s precence is of valeu. I also can give positions where two knights are stronger then two bishops. Again it all depends on the pawnstructure. how to evaluate the valeu of a piece. In short very complicated to program. Regards Marc
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.