Author: Robert Hyatt
Date: 10:18:26 04/08/03
Go up one level in this thread
On April 08, 2003 at 11:57:22, Uri Blass wrote: >On April 08, 2003 at 11:30:55, Robert Hyatt wrote: > >>On April 08, 2003 at 11:16:24, Uri Blass wrote: >> >>>On April 08, 2003 at 10:57:21, Richard Pijl wrote: >>> >>>>On April 08, 2003 at 10:04:02, Uri Blass wrote: >>>> >>>>>[D]8/5pp1/4p1kp/1Q6/2rqP2P/8/5PP1/5RK1 w - - 0 27 >>>>> >>>>>This position happened in a game of movei. >>>>>For the game see http://f11.parsimony.net/forum16635/messages/46511.htm >>>>> >>>>>I found that latest movei also suffers from the same evaluation problem. >>>>> >>>>>Movei evaluates the position as almost equal. >>>>> >>>>>I found that Crafty and a lot of programs understand that black has a clear >>>>>advantage(more than 0.5 pawns for black from the >>>>>first iterations). >>>>> >>>>>I understand that black is better but the question is what factor in evaluation >>>>>helps a lot of programs to understand it. >>>>> >>>>>I can explain reasons to give advantage for black from program point of view: >>>>> >>>>>1)The black king is more advanced and it is known that the king should >>>>>be at the last rank except endgames. >>>> >>>>I don't agree. After h5+ the black king is locked out from the immediate action >>>>and has to retreat to h7. >>> >>>I understand it but piece square table gives h7 worse score >>>than g1 for the king except endgames. >> >>Yes, but the piece square table (in crafty anyway) is a _small_ part of the >>score for >>a pawn or piece. It is really intended to be a "tie-breaker" and nothing else, >>so that if >>all else is equal (it rarely is) then the best piece-square value will break the >>tie. > >It is also a small score for movei but it has no big scores >for black here. > > >> >> >>> >>>I did not ask for opinion of humans but how do programs know it. >>> >>> >>>> >>>>>2)The white pawns are more advanced so by piece square table they >>>>>get bigger bonus after a move like h5+. >>>> >>>>But that would increase the white score? >>> >>>Yes >>>Piece square table give bigger score for the side that >>>it's pawns are more advanced. >> >>Same comment again. Piece/square table values in Crafty are nothing but >>tie-breakers. >>The main positional scores are based on lots of other considerations, with the >>pc/sq values >>breaking a tie when two different positions seem to have the same non-pc/sq >>scores. >> >>> >>>> >>>>I think you can detect black's advantage with three eval terms: >>>>- Limited mobility of the white rook. It cannot leave square f1 because of >>>>immediate threats on f2. >>> >>>I understand it but the question is if program understand it. >>>Moei use number of moves to evaluate mobility and f1 is only slightly worse >>>than possible sqares of the black rook. >> >>Then something is wrong with the way you are counting mobility. Mobility >>along the first rank is _far_ less important than mobility along an open file, >>or >>along your opponent's 2nd rank (your 7th rank). If you are just counting >>squares you >>can move to, that is probably the problem. I don't do mobility for all pieces, >>as there >>are other terms that produce similar effects. IE rook on an open file means a >>rook has >>mobility along that file, for example. > >I do not do rook on an open file and movei knows to put rook in an open file >thanks to mobility reasons. > >I know that my mobility evaluation can be improved but I am >sure that it is clearly better than not doing mobility. Maybe for you. But as I mentioned, there are _many_ ways to end up with a mobility-component without doing real mobility. Open files and horizontal activity work well in Crafty, as well as being able to move to an open file in one move. All that translates into mobility. Mobility is cheap to do for me, and I do it for bishops, as an example since it is easier than "open diagonals" to understand and produces the same result. But open files are different, because all files are not equal and so I choose to handle that differently. > >> >> >> >> >> >>> >>>>- Black has two attacking pieces, White only one. >>> >>>Movei does not count attackers and I do not know if this is an important factor. >> >>Two pieces attacking squares around the king, being close to the king, is an >>important factor. three pieces is even more important. This is a >>"second-order" >>evaluation term that has to factor in interrelationships among the pieces. > >I have small scores for attacking squares >near the king but I do not count attackers. Then you will play moves like Qh5 or Qh6 to get your queen close to the opponent's king, without having any supporting piece. Even though it would be better to centralize if you don't have any piece coordination to make moving the queen closer more useful. I call these sorts of terms "second order" where pieces interact with each other to improve or reduce the overall score. Individual piece scores are "first order". For king safety in Crafty, I have individual piece placement, then king safety pawn structure, and then pieces close to the king (the more the merrier) and I run that through a mathematical function that folds it all together and computes a score that is somewhat exponential based on the more good features it finds, the higher the score goes, but beyond what you would call a "linear relationship". > >> >> >>> >>>I think that the important factor is weaknesses of the white pawns. >> >>The score was not a pawn structure issue, as the score output showed. Pawn >>structure is in the "pawn evaluation" line which was +.06 (good for white but >>just barely). > >I understand but when I look at the game the problem was >that movei was unable to defend it's pawns. > >I agree that it is impossible to say based only on the position of >the pawns that the pawn structure is a weakness but I think that with >different pawn structure that was demonstrated in the second diagram white has >no problems in similiar conditions. > >Uri I think that being unable to defend the pawns was not a function so much of where the pawns were located, but due to the restricted white piece activity and the superior black piece activity. Black could attack more squares simultaneously while white's rook was pretty well stuck.
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.