Author: Robert Hyatt
Date: 09:06:54 10/02/01
Go up one level in this thread
On October 02, 2001 at 09:26:59, Uri Blass wrote: >On October 02, 2001 at 09:14:24, José Carlos wrote: > >>On October 02, 2001 at 01:38:36, Uri Blass wrote: >> >>>On October 02, 2001 at 01:30:09, Dave Gomboc wrote: >>> >>>>On October 02, 2001 at 00:49:54, Uri Blass wrote: >>>> >>>>>On October 01, 2001 at 23:50:46, Robert Hyatt wrote: >>>>> >>>>>>On October 01, 2001 at 18:43:59, Uri Blass wrote: >>>>>> >>>>>>>On October 01, 2001 at 17:38:23, Robert Hyatt wrote: >>>>>>> >>>>>>>>On October 01, 2001 at 17:34:19, Uri Blass wrote: >>>>>>>> >>>>>>>>>On October 01, 2001 at 16:50:47, Robert Hyatt wrote: >>>>>>>>> >>>>>>>>>>On October 01, 2001 at 16:20:44, Uri Blass wrote: >>>>>>>>>> >>>>>>>>>>>On October 01, 2001 at 15:06:21, Robert Hyatt wrote: >>>>>>>>>>> >>>>>>>>>>>>On October 01, 2001 at 14:28:35, Uri Blass wrote: >>>>>>>>>>>> >>>>>>>>>>>>>>No but most of the endgames are not pawn endgames. >>>>>>>>>>>>> >>>>>>>>>>>>>I do not say that I never saw Junior play endgames. >>>>>>>>>>>>>I say that I never saw it plays pawn endgames against humans. >>>>>>>>>>>>> >>>>>>>>>>>>>It is not enough to get an endgame in order to take advantage of Junior's >>>>>>>>>>>>>weaknesses in some kind of pawn endgames and GM's need also to get into >>>>>>>>>>>>>positions when pawn endgames that Junior does not understand are relevant in the >>>>>>>>>>>>>search. >>>>>>>>>>>>> >>>>>>>>>>>>>Amir also did not say that endgames are not important and that he has nothing to >>>>>>>>>>>>>fix in endgames and he talked about pawn endgames. >>>>>>>>>>>>> >>>>>>>>>>>>>I remember that Junior got endgames that are not pawn endgames in at least one >>>>>>>>>>>>>of it's tournament game against humans. >>>>>>>>>>>>> >>>>>>>>>>>>>It was an endgame when both sides had knights and trading for pawn endgames was >>>>>>>>>>>>>not relevant in that case. >>>>>>>>>>>>> >>>>>>>>>>>>>Uri >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>>That is the critical case to handle however. Suppose it is a pawn down. And >>>>>>>>>>>>it finds a way to trade knights and win that pawn back. And it ends up in a >>>>>>>>>>>>dead lost pawn ending because of the opponent's distant majority... >>>>>>>>>>>> >>>>>>>>>>>>If you don't know a lot about king and pawn endings, you had better _not_ get >>>>>>>>>>>>into king and pawn + 1 minor piece endings. It is _all_ about king and pawns >>>>>>>>>>>>there... >>>>>>>>>>> >>>>>>>>>>>It depends on the endgame and there are cases when pawn endgames with distant >>>>>>>>>>>pawn majority are not relevant. >>>>>>>>>>> >>>>>>>>>>>I remember that in that game Junior was a pawn up and pawn endgame was simply >>>>>>>>>>>not relevant. >>>>>>>>>>> >>>>>>>>>>>Junior drew the game but not because of not evaluating correctly pawn endgame >>>>>>>>>>>but because it overestimated it's two connected pawns(I remember a position of >>>>>>>>>>>KNPP vs KNP from that game and the opponent could sacrifice it's pawn and >>>>>>>>>>>capture Junior's pawns because the king of Junior was at long distance from the >>>>>>>>>>>pawns to defend them). >>>>>>>>>> >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>Wait. You are arguing with yourself here. First you said "not because of not >>>>>>>>>>evaluating correctly pawn endgames" and then follow that up with "it >>>>>>>>>>overestimated...". That is _exactly_ the point. That is an incorrect >>>>>>>>>>evaluation. >>>>>>>>> >>>>>>>>> >>>>>>>>>Amir did not say that there were no problem in evaluation in the endgame and the >>>>>>>>>point is that the practical problem was not pawn endgames. >>>>>>>>> >>>>>>>>>I did not say that he did not care about endgames but that he did not care about >>>>>>>>>pawn endgames and thought that there are more important things to improve. >>>>>>>>> >>>>>>>>>Uri >>>>>>>> >>>>>>>>Doesn't the game you posted sort of invalidate that statement??? >>>>>>> >>>>>>>I posted a position and the problem in the game was that the evaluation of a >>>>>>>similiar endgame position was not correct. >>>>>>> >>>>>>>The problem was not wrong evaluation of pawn endgame because pawn endgame means >>>>>>>that there are no knights in the board. >>>>>>> >>>>>>>Uri >>>>>> >>>>>> >>>>>>In a position where you have one knight per side, the search will encounter >>>>>>and have to evaluate a _huge_ number of pawn endgames. Because two plies >>>>>>is enough to eliminate the two knights. So the program _will_ have to >>>>>>be able to correctly evaluate pawn endings, or it will trade the knights to >>>>>>win a pawn and lose the ending... >>>>> >>>>>it depends on the position on the board and in that case wrong evaluation of >>>>>pawn endgame was not relevant. >>>>> >>>>>Junior does not evaluate wrong every pawn endgame and even when it evaluates >>>>>pawn endgames wrong it does not mean always that it is going to do a mistake. >>>>> >>>>>Uri >>>> >>>>Um, exactly when did Amir (allegedly) say this anyway? I've been following this >>>>discussion, but for all I know this was said when Junior 4.6 was the latest >>>>version, and the Junior team may have even changed their mind and worked on pawn >>>>endings by now! Or is this a comment made just a couple of months ago or >>>>something? >>>> >>>>Dave >>> >>>I heard it in 1998 when the commercial version was Junior4.6 and Junior4.9 was >>>the program that was playing. >>> >>>It is possible that Amir changed his mind about it >>>I do not know but I know that Junior7 had problems in evaluating the relevent >>>pawn endgame so it cannot find the right move in the position that started this >>>discussion. >> >> Which means that it would lose half a point if it was playing that game (the >>position came from a real game), so pawn endings knowledge _is_ important to >>results... :) >> >> José C. > >Junior never played that game and the games that are relevant are games that >Junior played. > >If I never play 1.d4 and I reply 1.d4 by Nf6 then games that begin with >1.d4 f5 are not relevant for me. > >I do not say that similiar endgames cannot happen to Junior but the question is >what is the probability that it happens to Junior and learning this probability >can be done only from games of Junior. > >Uri The problem with that kind of reasoning is this: The alpha/beta search is done in an exhaustive manner. If your evaluation has holes that produce wrong scores, then if those scores are too high, the search will force the game toward those positions to get the high score, and it will lose. If the scores are too low, the search will push the game away from those positions, and possibly also lose. In any case, if your evaluation spits out a bad number, the search will notice it. And that bad number will attract the game or push it in another direction. If the number if really bad, then the result will be equally bad. The stronger the opposition, the more likely this is to happen, since tactics will overrule poor evaluation in most cases. But when there are no tactics, big holes cause big problems. Even little holes cause big problems...
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.