Author: Ed Schröder
Date: 10:30:47 04/17/00
Go up one level in this thread
On April 17, 2000 at 10:30:10, Robert Hyatt wrote: >On April 17, 2000 at 00:31:36, Ed Schröder wrote: > >>On April 16, 2000 at 13:37:24, Robert Hyatt wrote: >> >>>On April 16, 2000 at 12:23:08, Ed Schröder wrote: >>> >>>>On April 16, 2000 at 12:10:41, Fernando Villegas wrote: >>>> >>>>>I think Rebel performed a really great show. After examination the game appears >>>>>fullo of tricks and traps -positionally kind- that very easily could put in the >>>>>wrong path more than a kown top chess program in SSDF list. My congrats for Ed, >>>>>that clearly seems to have "pasted" some of the holes in the tactical arena of >>>>>Rebel. >>>>>Fernando >>>> >>>>Thank you Fernando. I was quite nervous during the ending when I saw >>>>the white king moving to the queen side to hunt for the two weak >>>>black pawns on c6 and a6 leaving its own king undevelopped. But the >>>>wall Rebel created with Nb7 did the trick and according Smyslov himself >>>>there was no win. Well.. during the game I was not convinced of that as >>>>the Smyslov position looked real promising. But these monsters are tough >>>>defenders (most of the time). When I saw Rebel showing senseless moves >>>>in its main-variation (Bg7 / Bh8 / Bg7) I relaxed as this was a sign >>>>for me there was no progress for white. I am very happy with the draw. >>>> >>>>Not to speak about the Israel League, so far 4½ out of 5. What is there >>>>left to wish? Nothing... >>>> >>>>Ed >>> >>> >>>I thought you defended as well as anyone/anything could. The only thing I >>>didn't like was to see Rebel exchange into what was almost certainly a lost >>>ending (white with a queenside majority, black king on the other side of the >>>board.) Hard to say if it was really lost or not, as Smyslov made more than >>>one move that Crafty thought was wrong. Of course Crafty could have been >>>wrong instead, so that is not conclusive. But clearly there was no way for >>>white to lose, and no way for black to win, so why go for such a pawn >>>structure??? >>> >>>IE defending great is nice, but I would rather be working hard for a win, >>>with a possible draw option, rather than working hard for a draw, with a >>>very possible loss option... >>> >>>What caused Rebel to go for that setup in the first place? IE this is why I >>>have been fiddling with the pawn majority / candidate passed pawn code for so >>>long now... If you get that position vs a GM in blitz, you almost certainly >>>lose it. And if you let them get it, they will, over and over. There are >>>way too many different examples of that 3 vs 2 queenside where white wins >>>trivially, to take a chance on reaching it, IMHO. And if you get tricked into >>>reaching it, keeping heavy pieces on the board is the right answer. Trading >>>helps the majority side... >> >>And what did you have in mind as the right value of the bonus (penalty) for >>the pawn majority on the queen-side for the positions in question? The queen >>and rook exchanges comes to mind. > >I simply make passed pawns (or majorities or candidates) more valuable as >pieces come off. I didn't say (or hopefully imply) that what I do now is >perfect, because it _still_ needs tuning. But before I did it, I got rooked >into many such endings as the one in this game. Since I did it, I may still >get suckered by the opponent offering a pawn to reach such a position, but it >doesn't happen very often. > > > >> >>It's tricky tricky IMO as evaluations for center control, pawn structure >>and mobility are more dominating most of the time in a game. More: such a >>bonus for majority must be lower than the value of a passed pawn otherwise >>your program will never force the passer and even will avoid to make a >>passer by all means resulting in stupid moves. >> > >I agree. But if you tie _all_ of those scores (passers, candidates, majorities) >to material left, you get the right 'idea' even without decent tuning. IE A >simple off-side majority is worth over 1/2 pawn with no pieces left on the board >An outside passer is worth over a pawn with no pieces left, since that is >winning (there are other things to include here like an outside passer doesn't >win if the opponent has a protected passer anywhere.) > > >>I am not saying you are not right I am just curious how you evaluate such >>difficult cases in the right way. Another side effect of pawn majorities: >>if the bonus is too high your program could make stupid defense moves to >>avoid trading material where it really should trade. > > >That is an area of difficulty. Right now I have my king-safety term over- >whelm the majority term to avoid this. But if both kings are adequately safe, >the majority term can control the game, although with queens and rooks on the >board, the term is not huge. > > > > >> >>Strangely enough it was Rebel who got a passer on the board as first one. >>Even better, Smyslov never had one :) >> >>Ed > > >Right. Several weren't sure white playing g3 was the best move... But a >passer in the middle isn't nearly the problem as a passer on the far wing. I >caught outside passers several years ago. I finally did candidates/majorities >last year after a GM friend insisted that I do something. Now he wants me to >do something else, because his favorite tactic against computers is not working >and suddenly he is losing 90% of the games. :) > >But tuning is the hard part for sure, and right now it takes way more time than >I have... Thanks for your explanations, I guess your majority code is better than mine. To me evaluating majorities makes no sense in the middle-game, some sense in the early end-game and is often decisive in pawn endings and that's why Rebel (for instance) allowed the queen exchange as other aspects of the evaluation are dominant. But it surely makes sense to reconsider the whole majority topic from scratch especially in pawn endings. Not that I am looking forward to it as it is complicated stuff with high chances of overseeing the numerous exceptional cases. But Rebel surely can be improved here. Ed PS, it seems to me that Harald is right about 31.Nb6! / Nd7 and Nb8! I will ask Mr. Smyslov for his opinion.
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.