Author: blass uri
Date: 02:40:15 06/14/98
Go up one level in this thread
On June 13, 1998 at 22:57:12, blass uri wrote: > >On June 13, 1998 at 19:39:02, Thorsten Czub wrote: > > >>[Event "k6/200 40/120"] >>[Site "?"] >>[Date "1998.??.??"] >>[Round "?"] >>[White "Claus Carstens"] >>[Black "Junior 4.6"] >>[Result "1-0"] >>[ECO "D00"] >>[Annotator "Czub,T"] >>[PlyCount "37"] >> >>{3072kB, jun_book.ctg >>} 1. d4 {26} 1... d5 {2} 2. e4 {2} 2... dxe4 { >>-0.60/13 130} 3. Nc3 {247} 3... Nf6 {-0.56/15 0} 4. f3 {21} 4... exf3 { >>-0.72/14 177} 5. Nxf3 {8} 5... e6 {-0.77/15 177} 6. Bg5 {20} 6... Nbd7 { >>-0.77/15 692} 7. a3 {12} 7... Be7 {-0.94/16 392} 8. Qd2 {16} 8... O-O { >>-0.97/16 369} 9. Qf2 {17} 9... a5 {-1.12/14 306} 10. Qh4 {53} 10... h6 { >>-1.26/16 583} 11. Bd3 {4} The first test position >>11... hxg5 {-2.81/15 209} 12. Nxg5 {108} 12... >>g6 { >>-2.54/16 473} 13. Qh6 {23} 13... Qe8 {-1.79/12 83} 14. O-O {14} 14... >>Bc5 { >>#6/12 446} 15. Rxf6 {16} 15... Bxd4+ {#6/18 381} 16. Kh1 {12} 16... Nxf6 >>{ >>#4/41 332} 17. Nce4 {16} 17... Bd7 {#3/49 53} 18. Nxf6+ {9} 18... Bxf6 >>{#2/51 0 >>} 19. Qh7# {11} 1-0 >> >> >>[Event "40/120 k6/200"] >>[Site "?"] >>[Date "1998.??.??"] >>[Round "?"] >>[White "Claus Carstens"] >>[Black "Virtual2 "] >>[Result "1-0"] >>[ECO "D00"] >>[Annotator "Czub,T"] >>[PlyCount "47"] >> >>{3072kB >>} 1. d4 {0} 1... d5 {0.07/14 11} 2. e4 dxe4 3. Nc3 Nf6 4. f3 exf3 5. >>Nxf3 5... Bg4 {This is the so called Teichmann-variation. Also not the >>best >>defense against BDG. Computers often play this defense because they like >>the >>bishop on g4 pinning the queen. } 6. h3 Bxf3 7. Qxf3 c6 8. Qf2 e6 9. >>Bg5 Nbd7 >>10. a3 10... Be7 {You see the pattern. Exactly the same structures as in >>the >>first game vs. Junior. Virtual2 plays the same moves, not understanding >>1 % of >>white's plan. } 11. Qh4 O-O 12. Ne4 h6 13. Bd3 The second test position >> hxg5 14. Nxg5 g6 15. Qh6 >>Qe8 >>16. O-O Bd6 17. Rxf6 Nxf6 18. Rf1 Bf4 19. Rxf4 Qd7 20. Rh4 Qxd4+ 21. >>Rxd4 Rfd8 >>22. Rh4 Nh5 23. Qh7+ Kf8 24. Qxf7# 1-0 >> >> >>The problem in those games is NOT hxg. > >I do not agree. >can you show me how white win against Junior or virtualchess in these >games after Nd5 instead of hxg5. >I prefer to play with black in these positions. > >in the second game 15...Qa5+ is better than 15...Qe8 and I am not sure >if >black loses. I believe 15...Qa5+ losing after 16.c3 I believe the mistake in the games was hxg5 I say I believe because I did not prove it by a tree but all the lines I tried were a win for white. I think proving hxg5 is wrong should be an easy task for computer programs if they knew which lines to search. I believe a tree of less than 1000000 nodes should be enough. if the tree has only good moves for white. you do not have to have a better evaluation function to generate this tree but to know what moves have good chances to be good moves and for that reason be analysed. I cannot say in my evaluation function that they are good moves because I know they are good moves only after I analyse it and see that they are good. If I decide they are good moves by evaluation function like Cstal then I do unsound sacrifices. I believe the positions in the games black played are good test positions how much time does it take programs to play a better move than hxg5 fritz after some minutes want to play e5 which is not a better move and after a longer time find Re8 but after a longer time(more than an hour) it go back to the mistake hxg5. Junior4.6 cannot find it in a reasonable time the position in the second game seem to be more difficult. The third test position: The position Genius5 lost against Cstal by playing the mistake c5 is also a good test position for computer programs. r2r2k1/p4p1p/bqp1p1p1/4N3/1b1P4/8/PPQ2PPP/R1BR3K black to move Cstal does not do this mistake but Cstal also do unsound sacrifices. Genius5 liked c5 because it thought black win a pawn and a good computer program should see it is not trurth by search and not by evaluation function because otherwise it can do unsound sacrifices. Genius5 could see it if it knew which lines to search with the same evaluation function. another good test position that show my point that computer programs are weak at tactics is the position kasparov resigned against deeper blue. no program see the draw(I am not talking about seeing the right move that is only a result of hating the altrnatives). 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.