Author: martin fierz
Date: 02:03:42 12/23/01
Go up one level in this thread
On December 22, 2001 at 23:55:03, K. Burcham wrote: >On December 22, 2001 at 20:33:43, martin fierz wrote: > >>On December 22, 2001 at 03:36:46, K. Burcham wrote: >> >>>in this game there was a big swing from a positive score of about five to a >>>losing score of a mate in 11 in the final position. i have had several questions >>>about this game since i found this out at that time. this game has been >>>discussed here before. this was not a blitz game. shredders score >>>gradually increased to the +5, and back to a mate in 11 with 80 moves. all this >>>made me curious. here are my questions about this game. >>> >>>questions: >>> 1. was there one or more blunders? >>> 2. will some of the other top programs have this large eval swing? >>> 4. how miscalculated was this plan by shredder to push its own two pawns? >>> 5. can we find a key move in this game that will end in a draw? >>> 6. will shredder duplicate the same eval, in an analysis? >>> 7. one of the top programmers that post here, made this statement: >>> "well we know we cant trust shredders eval, so we should not >>> use this program to analyze games". i assumed he was basing >>> that comment on this game. so is his statement true? >>> 8. did deep junior eval also swing, the same as shredder? >>> 9. when did deep junior see the problem shredder was in? >>> 10. does this game indicate that deep junior can see far ahead >>> of most programs? >>> >>>now i am able to get some accurate answers. with dual 1533 and 1 gig ram, >>>i can run deep junior7. in the list below you will see i also used other >>>programs to compare. i tried to keep most depths from about 12 to 15. >>>deep junior was getting depths of 4 to 5 more than other programs. >>>sometimes the post would hang during analysis. but i think this info >>>is accurate. i started with move 21. shreddder was in book until about move 11. >>>deep junior also did not have 11...Rc8 in its book. it seems stefan >>>was using a different book than the commercial shredder5. maybe this was the >>>shredder5.32 book, not sure. at move 21 shredder eval was up about one pawn. >>> >>>all scores are for white, deep junior7. >>> >>>deep junior7, shredder5, chess tiger 14.0, deep shredder, crafty 18.08 >>> >>>21.Qh3 -.30 -1.25 -1.28 -.92 -.69 >>>31.Na4 -1.25 -3.05 -2.04 -2.88 -.2.05 >>>47.hxg4 -1.95 -4.08 -2.88 -4.25 -3.19 >>>55.Kh3 -1.76 -4.92 -4.52 -4.95 -3.00 >>>57.g5 -2.63 -5.37 -4.98 -5.40 -4.20 >>>61.Kh4 -1.75 -5.58 -3.92 -5.40 -3.25 >>>...a4 -1.08 -5.51 -3.62 -5.09 -3.00 >>>62.f6 -.92 -3.60 -3.60 -3.95 -2.95 >>>...Rb8 -.37 -2.00 -2.90 -2.80 -2.46 >>>63.Bf3 .00 -.27 -1.38 .00 -2.44 >>>... >>>64.g6 +.20 +.45 .00 +.37 +.19 >>>...Rd8 +3.50 +2.33 +3.88 +5.50(Qc2) +4.36 >>>66.Qc2 +4.00 +4.75 +4.08 +5.15 +2.47 >>>69.Nxd5 +3.16 +7.06 +3.48 +5.40 +2.33 >>>...Kf7 +5.75 +7.35 +6.85 +7.50 +6.62 >>>71.Qf3 +6.10 +9.00 +10.85 +10.11 +8.44 >>> >>>80...Kb6 mate in 11 stefan resigns >>> >>>with the above info all of my questions have been answered. >>>i am not saying that any of these programs would have played the same lines as >>>shredder to lead into this same game. but it is obvious that junior did not see >>>the win any sooner than the other programs. although dj eval was lower than >>>shredder considerably. shredder5 eval and large eval swing was a little more >>>than the others. but most programs evals were similiar. also at move 71 >>>deep junior eval is still 3 points lower than others. so DJ7 eval is >>>conservative on both ends. it seems to me that deep junior had no idea >>>that it could promote and mate, before the other programs could see this. >>>but in this game deep junior played for the win, and was the victor. >>> >>>i also studied some of the key moves in this game. after using ct14 and deep >>>shredder i found that on move 48...Bxf5, 49.exf5, the two pawns could now >>>advance forward for the queen. in the Maastricht game shredder did not see >>>this combination. instead if shredder had played 48...Nc2, >>>shredder wins. of course i know this was just one game. but dj was running >>>on 2x1533 and getting deep search. with this only one move change, shredder >>>wins. also slate and i tried another line with changing just one move in >>>this game and shredder had a draw with slate. slate if you read this, >>>i found another line that shredder could win. >>>here is the one move change, 48...Nc2 that shredder wins this game. >>> >>>[Event "Maastrict 2001 with move 48...Nc2"] >>>[Site "?"] >>>[Date "2001.12.22"] >>>[Round "6"] >>>[White "Deep Junior"] >>>[Black "Shredder5"] >>>[Result "0-1"] >>>[SetUp "1"] >>>[FEN "1r5k/2qb3p/3p3b/p3pN2/N3P1P1/2P1n1P1/6B1/R3Q2K b - -"] >>> >>>1... Nc2 2. Qd1 Nxa1 3. Nxh6 Bxa4 4. Qxa4 Nb3 5. g5 Nc5 6. Qa2 >>>Rf8 7. Qd5 a4 8. Nf5 a3 9. g4 Qd7 10. Qa2 >>>Qa4 11. Kh2 Nxe4 12. c4 Rb8 13. c5 Qb3 14. Qe2 >>>a2 15. Qa6 Qb2 16. Qa7 Rg8 17. g6 hxg6 18. Nh6 >>>a1=Q 19. Qxa1 Qxa1 20. Nxg8 Nc3 21. Kh3 Ne2 22. Ne7 0-1 >>> >>> >>>this is 48.Nf5 >>> [D] 1r5k/2qb3p/3p3b/p3pN2/N3P1P1/2P1n1P1/6B1/R3Q2K b - - >>> >>>here is the complete game. >>>[Event "WMCCC"] >>>[Site "Maastricht"] >>>[Date "2001.08.22"] >>>[Round "6"] >>>[White "Junior"] >>>[Black "Shredder"] >>>[Result "1-0"] >>> >>>1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be3 >>>e6 7. Be2 Qc7 8. f4 b5 9. a3 Bb7 10. Bf3 Nbd7 11. Nde2 Rc8 >>>12. O-O Be7 13. Kh1 O-O 14. b4 Nb6 15. Qd3 Rfd8 16. f5 Nc4 >>>17. fxe6 fxe6 18. Bd4 Rd7 19. Bxf6 Bxf6 20. Bg4 Re7 21. Qh3 >>>Rce8 22. a4 bxa4 23. Bh5 Rf8 24. Bg4 Bc8 25. Rxa4 Ree8 >>>26. Raa1 g6 27. Qg3 Qb6 28. Bh3 Qxb4 29. Qd3 Bg7 30. Rfb1 >>>Qc5 31. Na4 Qc7 32. c3 Rf6 33. Nd4 Ref8 34. Nf3 Rf4 35. Nd4 >>>Re8 36. Rf1 Rxf1+ 37. Rxf1 Bd7 38. Ra1 Kh8 39. Qf1 g5 >>>40. Qd1 Ne5 41. g3 g4 42. Bg2 a5 43. Qc2 Rb8 44. Qc1 Nc4 >>>45. Qe1 Bh6 46. h3 Ne3 47. hxg4 e5 48. Nf5 Bxf5 49. exf5 >>>Nc2 50. Qd1 Nxa1 51. Qxa1 e4 52. Qd1 e3 53. Bf3 Bg7 54. Kg2 >>>Re8 55. Kh3 Qc4 56. Be2 Qc6 57. g5 Be5 58. Nb2 d5 59. Nd3 >>>Bxc3 60. Nf4 d4 61. Kh4 a4 62. f6 Rb8 63. Bf3 Qb5 64. g6 >>>Rd8 65. g7+ Kg8 66. Qc2 Rd6 67. Qa2+ Qb3 68. Bd5+ Rxd5 >>>69. Nxd5 Kf7 70. Qg2 d3 71. Qf3 Qb8 72. Qh5+ Ke6 73. Nxc3 >>>e2 74. Qg4+ Kf7 75. Kg5 h6+ 76. Kxh6 e1=Q 77. Qg6+ Ke6 >>>78. f7+ Kd7 79. Qxd3+ Kc7 80. Qc4+ Kb6 1-0 >>> >>>final position. stefan resigns with two queens on board. 80...Qb6 >>>and mate in 11. >>> >>> [D] 1q6/5PP1/1k5K/8/p1Q5/2N3P1/8/4q3 w - - >>> >>>kburcham >> >>dear kburcham, >>interesting analysis, but i think you have not got the right answer... >>hmm, so you are basically saying that Bxf5 is supposed to be *the* mistake? >>i looked at the game, and if one of my chess pupils had been black, he would >>have gotten a serious talking-to! let me explain: >>the position after 51. Qxa1 is clearly winning for black. if you ask any strong >>chess player, he will tell you what the plan is: exchange queens - at any price, >>for a pawn or two even. the *only* thing that can go wrong for black in this >>position is that he will have a problem with his king, and this only happens >>when there are queens on the board. besides, an exchange is not worth much in >>the middle game, but decisive once you enter the endgame, and there is no other >>pair of rooks on the board. this is what all strong humans know, and what >>obviously even top programs have not got programmed into their evaluation. so, >>returning to the position after 51.Qxa1, how should black proceed? since his >>position is already completely winning, he doesnt have to win a second time by >>queening the e-pawn. i'll suggest a line, you can try to improve for white. my >>line is 'human' in the sense that i will just attempt to exchange queens. i do >>this by activating my queen first, and look for a possible exchange. The white >>moves are made by fritz 5.32 running in chessbase. mostly the move indicated got >>by far the best evaluation. >> >>51. ..Qc4 52. Qd1 Rg8 53. Bf3 e4 54. Be2 Qa2 (menacing Qd2, game over) 55. Nb6? >>Qd2 56. Qxd2 Bxd2 and i'm not going to look at this position any more, it is so >>winning for black. >>i find it very interesting that fritz 5.32 doesnt mind the queen exchange, and >>would also not play 55. ...Qd2! itself. this clearly shows that this version of >>fritz has no clue about what is going on here. since you have all top programs, >>could you give this a try and see which program would exchange the queen here? >>alternatively, white could play 55. Qxd6 Qxe2 56. Qxh6 Qxg4 57. Qf6+ Qg7 and >>black has again achieved his goal of exchanging the queens. >> >>believe me, there is no other viable plan for black in this position than the >>exchange of queens. i have made this experience too many times as the guy who >>should have done anything to get the queens off the board, the last time our >>team lost a match 4.5-3.5 in the highest swiss league where i allowed a >>perpetual check in a winning position. i have seen my pupils lose games a piece >>up because they were checkmated. of course these are kids, but it's the same >>here. there is a nice website called www.chesscafe.com, there there used to be a >>column of some guy teaching weaker players. his motto was GTS, "go to sleep": if >>you are a piece up, trade the queens then you can GTS - else you always have >>to keep your eyes open for surprises, just like in this game. do the test and >>see which program will play ...Qd2. of course, other moves win too there, but >>that's not the point - the point is, do the programs know = have it coded in >>their evaluation, that being an exchange up is worth more once the queens are >>gone? >> >>thanks & best regards >> martin > > >hello martin----- >yes i see. thanks for the lesson. 51...Qc4 works well. forces queen >exchange. i played this through with gambit tiger 2.0 and deep junior7. >this line wins for shredder as you point out. that is amazing how we can change >one move and the program wins. this is not the first time i have seen this. >not a blunder, no eval change with either move. > >but here is a line tiger wanted to change in yours. >i have not studied this line but i will. > >-3.10 17 147134136 53.Qxd6(Bf3) Bg7 Bd5 Qf1+ Kh2 Qe2+ Kh3 Re8 Qd7 Rb8 Qd6 Rf8 >Qe6 Qe1 Nc5 e4 Qxe4 Qf1+ Qg2 see sune's line in the other post for a convincing way to win after 53. Qxd6 >here is a position tiger plays into after changing to your 51...Qc4. tiger says >mate in 9. >[D] 8/8/2N5/6k1/4p3/6P1/3b2K1/4r3 b - - 14 1 > >also i am going to let deep junior run all night 2x1533mhz, to see if it will >play 55...Qd2 and force the queen exchange. it is at depth 23, and has not >played it yet. > >kburcham i really find it interesting that these programs search so deep without wanting to exchange the queens. the GTS thing is a very human idea, because we make lots of mistakes. one might think that computers wouldnt have to adhere to it, because they will not miscalculate. but your position here is a beautiful example that even strong programs can stumble because they don't know about it. this means that their evaluation functions do not know that an ending with R vs minor piece is a decisive advantage. in the middle game the exchange is not worth much, maybe a bit more than a pawn, but in the endgame it's much more. any comments from programmers? how do your evaluation functions evaluate being an exchange up in middle- and endgame? and in RR vs R+minor piece and just R vs minor piece? cheers martin
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.