Author: Uri Blass
Date: 14:20:11 04/20/03
Go up one level in this thread
On April 20, 2003 at 16:22:25, Sune Fischer wrote: >On April 20, 2003 at 15:25:49, Uri Blass wrote: > >>On April 20, 2003 at 15:04:22, Sune Fischer wrote: >> >>>On April 20, 2003 at 03:23:11, Uri Blass wrote: >>> >>>>>>>[D]8/r7/8/5bk1/8/5B2/5RPP/6K1 b >>>>>>> >>>>>>>Terry >>>>>> >>>>>>Crafty sees this instantly: >>>>>> >>>>>> 4-> 0.05 2.00 1. ... Ra1+ 2. Rf1 Ra2 3. g3 Rd2 >>>>>> 5 0.05 ++ 1. ... Ra1+!! >>>>>> 5 0.06 0.00 1. ... Ra1+ 2. Rf1 Rxf1+ 3. Kxf1 Bh3 >>>>>> 4. gxh3 Kg6 5. Kf2 >>>>>> 5-> 0.07 0.00 1. ... Ra1+ 2. Rf1 Rxf1+ 3. Kxf1 Bh3 >>>>>> 4. gxh3 Kg6 5. Kf2 >>>>>> 6 0.08 0.00 1. ... Ra1+ 2. Rf1 Rxf1+ 3. Kxf1 Bh3 >>>>>> 4. gxh3 Kg6 5. Kf2 Kg5 >>>>>> 6-> 0.09 0.00 1. ... Ra1+ 2. Rf1 Rxf1+ 3. Kxf1 Bh3 >>>>>> 4. gxh3 Kg6 5. Kf2 Kg5 >>>>>> 7 0.24 0.01 1. ... Ra1+ 2. Rf1 Rxf1+ 3. Kxf1 Bh3 >>>>>> 4. gxh3 <EGTB> >>>>> >>>>>Good Job! I should test all my chess software on this ending, to see if they >>>>>"understand" the draw with a passed pawn or doubled pawns on the h-file with the >>>>>wrong coloured Bishop. >>>>> >>>>>But remeber, the Mach III is slow and old, with no EGTB and solves this position >>>>>quickly! >>>>> >>>>>Dan Spracklen had to give it some endgame understanding in the code; Which to me >>>>>says what could he have come up with, had he remained creative in the world of >>>>>CC? I believe he ended his chess programming career in 1994 when his last >>>>>programme was for Saiteck. >>>>> >>>>>Terry >>>> >>>>I think that every programmer can teach his(her) program to have that knolwedge >>>>if (s)he is interested in it. >>> >>>All you have to do is to support the egtbs. >> >>No > >Yes :) > >>You do not need to have tablebases for this and most people have not enough >>memory for all the 5 piece tablebases(KBPP vs K is 5). > >It's a simple TB draw 4 moves in, the PV of Crafty also shows that. It proves nothing. >I believe any engine with a not completely broken TB implementation can find >this draw. It might be a good position to check the implementation, though. No I believe that most people have not the KBPP vs K tablebases so it is bad position. I do not so I cannot use it to test nothing about tablebases and I do not plan to download the KBPP vs K tablebases only for testing because I prefer to use memory of the computer for other purposes. I think that programs should know every 4 vs 1 position by knowledge and it is a waste of time even to call tablebases in 4 vs 1 positions. > >>I do not like egtbs and they can also be counter productive >>because you may be happy with a lost KR vs KPPP (Rook is more) when KR vs KPP >>is drawn that unfortunately you reject because of tablebases. > >Getting out of bed in the morning can also be dangerous ;) >If they help more often than they hurt, then I think the decision is easy. > >>It is possible that tablebases are productive only after you have enough >>knowledge in the evaluation. > >I believe it is the opposite actually, the more knowledge you have the less you >need the table bases. A smart engine without TBs may not declare mate in 23 but >it will still win in 30 moves. > I think that it is dependent on the knowledge. I did not think about specific knowledge for 5 piece positions but about knowledge that is also common to 6 and 7 piece positions(I agree that knowledge like knowledge to win KRB vs KR or KBN vs K may do tablebases less important but I did not think about that knowledge but about knowledge in evaluation that is relevant also for other positions). It may be possible to use it in a productive way to avoid problems that I mentioned(the program prefers loss by KR vs KPPP and not drawn KR vs KPP) by having rules to trust tablebases scores only in part of the cases but I do not like it. I also do not like learning to use another code that is not mine when there are a lot of other things that should be improved. >>>I doubt Crafty will solve this one "on its own". >>> >>>-S. >> >>Crafty has the relevant knowledge and I do not see it as hard to solve it after >>adding the relevant knowledge. >> >>The releavant knowledge is relevant not only for that case but also for other >>cases. > >Maybe, maybe not (have you checked the code?). I do not need to do it. I can simply give crafty to analyze and I do not have the relevant tablebases. >I don't think anyone is highly motivated to program lots of simple 4-5 piece >specific endgames when the engine most often is used with TBs. Seems time would >be better spent in the areas out of reach for the TBs. This is exactly the point. This knowledge is mainly for position that is out of the tablebases and also for position that is in the tablebases. KBPP vs KP is also draw. 7k/2p5/8/8/8/7P/7P/5BK1 w - - 0 1 >I also think there are too many exceptions to make simple rules for what is won, >drawn or lost in these endgames. Often one tempi is enough to change the >outcome, if you add general knowledge you are likely to be wrong in lots of >cases too. Your opinion. My opinion is that there are some rules that are always right. I dislike tablebases at this moment. I also do not like to add to my code something that is even not alway correct(tablebases can tell you mate in 90 and it is a drawn by the 50 move rule. I know that programs with tablebases usually have no rule to translate mate in 90 to some static score that is less than mate because of the 50 move rule and I do not like copying from what other people are doing. Uri
This page took 0.01 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.