Author: Vine Smith
Date: 21:58:33 05/20/01
Go up one level in this thread
On May 21, 2001 at 00:05:27, Robert Hyatt wrote: >On May 19, 2001 at 15:56:53, Vine Smith wrote: > >>On May 19, 2001 at 12:29:46, Sune Larsson wrote: >> >>>On May 19, 2001 at 11:33:42, Vine Smith wrote: >>> >>>>On May 19, 2001 at 10:46:48, Sune Larsson wrote: >>>> >>>>>On May 19, 2001 at 10:15:06, Vine Smith wrote: >>>>> >>>>>>On May 19, 2001 at 08:52:03, Sune Larsson wrote: >>>>>> >>>>>>>On May 19, 2001 at 06:22:23, Vine Smith wrote: >>>>>>> >>>>>>>>On May 19, 2001 at 05:26:07, Jeroen Noomen wrote: >>>>>>>> >>>>>>>>>Fritz - Gambit Tiger Leiden round 4 >>>>>>>>>------------------------------------- >>>>>>>>>1. d4 d5 2. c4 dxc4 3. e4 Nc6 4. Be3 Nf6 5. Nc3 e5 6. d5 Na5 7. Nf3 Bd6 >>>>>>>>>8. Qa4+ Bd7 9. Qxa5 a6 10. Na4 Qe7 11. a3 Nxe4 12. Bxc4 b5 13. Bd3 Nf6 >>>>>>>>>14. Nc3 e4 15. Nxe4 Nxe4 16. O-O O-O 17. Rfe1 f5 18. h3 Qf7 19. Rad1 Rfe8 >>>>>>>>>20. Bc2 g6 21. b4 Nc3 22. Rd3 Qg7 23. Bd1 Ne4 24. Bd4 Qf7 25. Rde3 g5 >>>>>>>>>26. Bc2 g4 27. hxg4 fxg4 28. Bxe4 gxf3 29. Bxf3 Qf4 30. Bc5 Qh2+ 31. Kf1 >>>>>>>>>Bf4 32. Rxe8+ Bxe8 33. Be3 Bd6 34. Ke2 Bh5 35. Rd1 Qe5 36. Rd4 Bxf3+ >>>>>>>>>37. gxf3 Qf5 38. Rd3 h5 39. a4 h4 40. axb5 h3 41. Qa1 h2 42. b6 cxb6 >>>>>>>>>43. Qh1 Rc8 44. f4 Qh5+ 45. Kf1 Rc2 46. Bd2 Bxf4 47. d6 Rxd2 48. Rxd2 Bxd2 >>>>>>>>>49. d7 Qd1+ 50. Kg2 Qxh1+ 0-1 >>>>>>>> >>>>>>>>Astonishing positional suicide -- few, if any, humans would ever play 21.b4, >>>>>>>>entombing the queen. Of course, Tiger could afford to open the game in any way >>>>>>>>it pleased after that. It would be interesting to see if the lack of interest in >>>>>>>>liberating the queen is duplicated by many programs; correct or not, I would >>>>>>>>have been so concerned about this offside post that I would have played 17.Qe1 >>>>>>>>immediately. >>>>>>> >>>>>>> >>>>>>> The terms "mobility" and "king safety" comes to mind after this kamikaze... >>>>>>> >>>>>>> Anyway, I made a quick check on PIII 800 with 128 Mb hash. 4 min ponder. >>>>>>> >>>>>>> Fritz 5.32 21.b3 and later plays b3-b4. >>>>>>> Fritz 6 Old 21.b3 >>>>>>> Fritz 6 (Newest) 21.b4 >>>>>>> Deep Fritz 21.b4 >>>>>>> Hiarcs 7.32 21.b4 >>>>>>> >>>>>>> Then we have some non closers: >>>>>>> >>>>>>> Crafty 18.08 21.Bc1 >>>>>>> Shredder 5 21.h4 >>>>>>> Gandalf 4.32h 21.Kf1 or 21.Re2 (shifting after 4.02) >>>>>>> >>>>>>> Sune >>>>>> >>>>>>Horrible that not only Fritz, but also Deep Fritz (this program is going to play >>>>>>Kramnik???) and Hiarcs (with something of a good reputation for positional play, >>>>>>I thought) commit this blunder. Also, any program that doesn't evaluate the >>>>>>resulting position at something like -2.00 or lower has nothing to be proud of. >>>>>>I'm putting together a suite of positional problems to evaluate programs' >>>>>>understanding or lack thereof, and I'll have to include this position as an >>>>>>"avoid move" type of problem (even though it's a little unfair to Fritz, which >>>>>>will of course fail this one). >>>>> >>>>> >>>>>>Can you tell if Gandalf wants to play 21.Re2 in order to follow up with 22.Qe1? >>>>>>If so, that would mark it as the only program so far to demonstrate some concern >>>>>>about the queen's situation. >>>>>> >>>>>>Regards, >>>>>>Vine >>>>> >>>>> Gandalf ponders about lines like 21.Re2 Nf6 22.Ng5 Qe7 23.Qc3, but >>>>> after some 8 minutes again shifts to 21.Kf1 b4 22.a4 = (Gandalf) >>>>> >>>>> Sune >>>> >>>>Ugh! This suggests some generalized failure of program evaluation routines. Even >>>>in Gandalf's line with 21...b4 22.a4, having the queen cut off in this manner >>>>seems fatal. I'm going to investigate whether very long-term searches will >>>>correct this fault, which would at least give these programs the hope that with >>>>the faster hardware of tomorrow, they could avoid such disastrous decisions. >>>> >>>>Regards, >>>>Vine >>> >>> Long-term searches here would sure be interesting - as your coming suite of >>> positional problems. The latter must be pure gold for the programmers, if >>> it's possible to pinpoint down just what the positions are all about - and >>> transform them into programming language... >>> >>> Sune >> >>Fritz 6e at 17 ply still doesn't think it has done a bad thing. It currently >>believes that in reply to 21.b4, Black should play 21...Re7, evaluating the >>position at +0.03. In the entire line it gives, there is no effort made to free >>the queen from her prison. I'm hoping that it will see the light within the next >>3 or 4 ply -- or come up with some astounding method of liberating the queen >>with no bad side effects. >>I started working on the positional suite in connection with a program a friend >>of mine is going to write. I'm arguing for some radical new approach to >>evaluation, but, indeed, the challenge is to formulate it in such a way that he >>can code it successfully. Maybe I'll get lucky and stumble across a new >>frontier. >>The suite is meant to show that something new might be necessary. I have about >>120 positions now, but I keep having to review and replace them after I look at >>test runs. It's difficult -- unlike tactical problems, it's often impossible to >>absolutely prove the validity of some positional continuations. So I try to find >>positions where some particularly fine positional move has been endorsed and >>explained by several IMs and/or GMs in varying sources. But also some very >>elementary problems as well -- many programs have been failing these too. When I >>have something consistent and grouped into useful themes for diagnostic >>purposes, I'd be glad to release it in EPD format to anyone interested. >> >>Regards, >>Vine > >Just as a note, crafty has a couple of terms here that are preventing it from >falling into the b4 hole. 1. King safety doesn't like the queen stuck on the >off-side of the board with no hope to get over to the kingside; 2. the >usual blocked pawns/levers evaluation makes totally blocking things up very >hard.. I thought Crafty's choice was the most sensible of those summarized by Sune Larson. I will run the position after Fritz's 21.b4 to see if Crafty has a strongly negative view of what White has done. That would be important for any program trying to win from this point, as failure to understand that keeping the queen confined must be a priority could result in the premature release of this piece for the sake of less important advantages. I'm going to use 18.9 instead of Larsson's 18.8 (the Chessbase version, I'm guessing), because I prefer to work from the command line. Are there any significant evaluation differences between recent Crafty versions?
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.