Author: Uri Blass
Date: 03:53:01 08/01/02
Go up one level in this thread
On August 01, 2002 at 04:51:07, Gian-Carlo Pascutto wrote: >On August 01, 2002 at 00:57:51, Uri Blass wrote: > >>On July 31, 2002 at 12:12:49, Tim Foden wrote: >> >>>2rq1rk1/pb3ppp/1p2p3/n7/3PP3/3B1N2/P2Q1PPP/3RR1K1 w - - bm d5; c0 " pv d5 exd5 >>>e5 - piece activity"; >> >>Rc1 0.40/9,0.35/10 >> >>After d5 >>Movei suggest exd5 exd5 0.00/9,0.04/10,0.29/11 >> >>After d5 exd5 e5 movei says Nc4 0.75/10,0.78/11 >> >>It seems that the line that is suggested is bad because it simply lose a pawn >>for no compensation. >> >>Movei evaluates mobility and it see no compensation. > >I think that what you're trying to say here is that neither you nor >your program are good enough to _see_ the compensation. After the >line white is winning. The explanation piece activity is misleading and it should be piece activity and mate attack. I also suspect that part of the positions are about tactics. a program may find the line that you post for tactical reasons without positional knowledge. > >An example continuation is Nc4 Qf4 Nb2 Bxh7 Kxh7 Ng5+ Kg6 h4 > >Does movei see here that white is won? Even here Movei has problems to see that white is winning but at least it starts to see that there are moves that do not win for black(I guess it extends the wrong lines so it has problems to get deep here). 8 229 4053 5697759 d8d7 d1b1 b2c4 f4g3 g6h6 g3f4 f8e8 g5f7 h6h7 9 23 8177 11408013 d8d7 h4h5 g6h6 g5e6 h6h7 e6f8 c8f8 d1d2 b2a4 e5e6 d7e8 e6e7 f8h8 9 24 10456 14611316 f7f5 9 162 13524 18983721 f7f5 d1b1 b2c4 g2g4 f5g4 f4g4 d8e7 g5e6 g6h7 e6f8 c8f8 e5e6 9 162 21487 29550359 f7f5 d1b1 b2c4 g2g4 f5g4 f4g4 d8e7 g5e6 g6h7 e6f8 c8f8 e5e6 10 0 27403 37923033 f7f5 h4h5 g6h5 g5e6 d8e7 f4h2 e7h4 e6g7 h5g5 g7e6 g5h5 g2g4 f5g4 e6g7 h5g5 g7e6 g5h5 10 1 37702 52000810 d8d7 10 16 39929 55099718 d8d7 h4h5 g6h6 g5e6 h6h7 e6f8 c8f8 d1d2 b2a4 e5e6 d7e8 e6e7 f8h8 d2d1 10 17 54869 75534171 c8c4 10 20 55483 76404508 c8c4 h4h5 g6h6 g5f7 h6h7 f4f5 h7g8 f7d8 f8f5 d1d2 f5h5 d8b7 c4h4 f2f4 h4h1 g1f2 10 21 60911 83787993 d8e7 10 179 82695 112558034 d8e7 h4h5 g6h6 g5e6 h6h7 f4f5 h7g8 e6f8 c8f8 d1b1 b2c4 b1b3 f8e8 f2f4 e7c5 g1f1 10 179 84380 114765298 d8e7 h4h5 g6h6 g5e6 h6h7 f4f5 h7g8 e6f8 c8f8 d1b1 b2c4 b1b3 f8e8 f2f4 e7c5 g1f1 11 114 165893 222765731 d8e7 f4d2 e7a3 h4h5 g6h5 g5h3 f7f6 e5f6 g7f6 h3f4 h5h6 f4d3 h6g7 d2b2 a3b2 d3b2 > >>I suspect that most of the positions are not correct. >>I expect programs to do better with more time in a good positional test suite. >> >>If this is not the case it means that the test is bad. > >a) It's not because your program is not smart enough to solve the positions >that they're incorrect. They may not all be correct, but your claim that >some of them are incorrect just because your program doesn't solve them is >by far one of the most ludicrous I have ever seen here. I do not say that it is the reason but I tend to believe that big part of positional tests are wrong unless I know about a test to prove that they are correct. A possible test may be beating the top programs that cannot solve it in a match of 2 games. If the right positional move lead to better result then it is possible to do it. > >b) If you don't have the right knowledge, more time will help very little >in a pure positional set. I do not see it in that way. positional test should be a test of positions when programs cannot see winning material. The fact that they cannot see winning material does not mean that more time cannot help them because it is possible that with more time they may find a way to translate positional advantage a to positional advantage b that almost every program knows. 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.