Author: Uri Blass
Date: 11:20:46 11/27/03
Go up one level in this thread
On November 27, 2003 at 14:05:53, Robert Hyatt wrote: >On November 26, 2003 at 23:45:47, Uri Blass wrote: > >>On November 26, 2003 at 23:15:57, Robert Hyatt wrote: >> >>>On November 26, 2003 at 22:12:43, Uri Blass wrote: >>> >>>>On November 26, 2003 at 21:14:14, Robert Hyatt wrote: >>>> >>>>>On November 26, 2003 at 20:20:26, Peter Berger wrote: >>>>> >>>>>>For information about setup and rules: >>>>>>http://www.talkchess.com/forums/1/message.html?329237 >>>>>> >>>>>>Looks strangely familiar? - yes, the controversial setup of game 2 again. Space >>>>>>advantage vs. pawn - OK, nothing new, we get this in every other game. >>>>>> >>>>>>This time after the "novelty" 27. Nf4 Junior reached less than nothing - it's >>>>>>just too drawish. >>>>>> >>>>>>In 1985 there was some upset in our little rural chessclub - a computer had >>>>>>proven hat KRBKR was a forced win, amazing. Dunno how this came up, and of >>>>>>course these days we know better - many if not most of these positions are a >>>>>>draw. >>>>>> >>>>>>But surely not all of them - KRBKR is dangerous to the extreme, especially if >>>>>>the defending king is restricted to the back row - Crafty deals with these >>>>>>positions as a draw statically in general (my humble and very possibly wrong >>>>>>judgement after watching the game and evals , no analysis of source coude ) - >>>>>>dary judgement by the engine. >>>>>> >>>>>>Not to have to deal with general criticism about setup I spent quite some time >>>>>>tonight with the best and fastest computer I could come up with in analysis - of >>>>>>course with all the 5-piece tablebases, to make sure the problem is for real. >>>>>> >>>>>>It is - Crafty is not aware that KRBKR is just mean, even with 5 piece >>>>>>tablebases, something very similar (identical in fact) to the actual game would >>>>>>have happened. >>>>> >>>>>I will look, but two things. If it ended up in a lost KRB vs KR, that can >>>>>happen. But if it has tables it will likely _not_ end up in that position >>>>>as it would avoid trading down to it... >>>> >>>>The problem was losing KRB vs KRPP based on looking in the game. >>>> >>>>I am not sure about the losing mistake of Crafty but I think it was Rh1 in the >>>>following position >>>> >>>>[D]5k2/7R/7p/p3K3/2B5/8/8/5r2 b - - 0 64 >>>> >>>>> >>>>>That's the purpose of tables. >>>>> >>>>>Losing the occasional KRB vs KR is not something I worry about, because most >>>>>all chess program users have at least the full 5-piece set of tables which >>>>>avoids 99.9% of the problems, where KRB vs KR is won is a very tiny percentage >>>>>of all KRB vs KR endings... >>>>> >>>>>When you say something "identical" would have happened with the tables, you >>>>>overlook the power of probing the tables _deep_ in the search, so you simply >>>>>don't trade into such lost positions... Crafty doesn't trade into such a >>>>>position and _then_ realize "crap, lost position, shouldn't have done that." >>>>> >>>>>That's why the tables are important. >>>>> >>>>>Here, if a program thinks KRB vs KR is winning, it will be wrong most >>>>>of the time. You look _really_ silly winning a piece, trading your last >>>>>pawn, to end up in a nearly forced draw with KRB vs KR. >>>> >>>>Junior did not evaluate it as winning a piece but evaluated it as advantage for >>>>the side with the bishop(I am sure that Junior could see by search at least >>>>winning one pawn to get KRB vs KRP but Junior never evaluated it as something >>>>close to +2 and the evaluation was +0.5 or +0.8 in most of the endgame). >>>> >>>>Uri >>> >>> >>>It was _still_ wrong. In KRB vs KRP, in 99% of the cases, KRP draws or >>>wins. Thinking the KRB side is better is simply wrong. >> >>I think that evaluation is dependent on the position and KRB vs KRP is not >>enough information. >> >>winning for the pawn can happen only if the pawn is very advanced(otherwise the >>bishop can sacrifice itself for the pawn for a draw) so it is clear that in the >>relevant position there were only chances for the bishop. >> >>The fact that the opponent king was in the last rank also helped to increase the >>chances. >> >>It is not the first case when I see that KRB wins KRP in comp-comp games. >> >> >> Except for the >>>rare exception. I suspect Crafty with tables might well avoid the critical >>>trade but I am not sure, some are very deep. >>> >>>But I'll take the occasional loss for being wrong, just as I accept the >>>occasional case where I don't take the trojan horse but should. It was >>>wrong here (Crafty). But it is right _most_ of the time. A program that >>>thinks KRB vs KR or KRP, or KRN vs KR or KRp, is better for the KRB or KRN >>>is going to be wrong most of the time. Look at the KRB vs KR tablebase >>>summary file... >> >>Statistic about KRB vs KR is not relevant because the position that heppens in >>the board is not random KRB vs KR position. >> >>Uri > >Actually it is, when you think about it. This same thing could have happened >with the black king anywhere on the board and white and black would have >evaluated it the _same_ way, and white would not have won most of those >cases. I do not think that the sides evaluate it the same way when the black king is anywhere in the board. I suspect that it may be the case for Crafty but not for Junior. 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.