Author: Uri Blass
Date: 21:53:08 05/09/02
Go up one level in this thread
On May 09, 2002 at 18:22:26, Vine Smith wrote:
>On May 09, 2002 at 12:59:36, Uri Blass wrote:
>
>>On May 09, 2002 at 09:39:27, Vine Smith wrote:
>>
>>>On May 09, 2002 at 09:20:48, Uri Blass wrote:
>>>
>>>>On May 09, 2002 at 09:14:07, Vine Smith wrote:
>>>>
>>>>>On May 09, 2002 at 08:56:31, Uri Blass wrote:
>>>>>
>>>>>>On May 09, 2002 at 08:05:27, Vine Smith wrote:
>>>>>>
>>>>>>>On May 09, 2002 at 07:15:24, Uri Blass wrote:
>>>>>>>
>>>>>>>>On May 09, 2002 at 06:46:45, Vine Smith wrote:
>>>>>>>>
>>>>>>>>>On May 09, 2002 at 06:17:15, Uri Blass wrote:
>>>>>>>>>
>>>>>>>>>>On May 09, 2002 at 06:04:57, Vine Smith wrote:
>>>>>>>>>>
>>>>>>>>>>>On May 09, 2002 at 03:02:52, Slater Wold wrote:
>>>>>>>>>>>
>>>>>>>>>>>>[D]8/8/8/3Q3k/p2p1q2/P2P4/6K1/8 b - - 0 71
>>>>>>>>>>>>
>>>>>>>>>>>>This is from a game I played last nite against Crafty 18.14. I was using Deep
>>>>>>>>>>>>Junior 7.
>>>>>>>>>>>>
>>>>>>>>>>>>I am unsure what the other guy was using (HW), although I know it was an SMP
>>>>>>>>>>>>system. No idea about TBs, or anything like that. I have over 20GB of TBs,
>>>>>>>>>>>>3/4/5 and a lot of 6's.
>>>>>>>>>>>>
>>>>>>>>>>>>Here is the entire game:
>>>>>>>>>>>>
>>>>>>>>>>>>Crafty 18.14 - Deep Junior 7 [A29]
>>>>>>>>>>>>(2), 08.05.2002
>>>>>>>>>>>>[-0.12]
>>>>>>>>>>>>
>>>>>>>>>>>> 128MB, jbook.ctg
>>>>>>>>>>>> 1.c4 Nf6 2.Nc3 e5 3.Nf3 Nc6 4.g3 d5 5.cxd5 Nxd5 6.Bg2 Nb6 7.0-0 Be6 last book
>>>>>>>>>>>>move -0.12/15 9 8.d3 (Qc2) 8...f6 2054kN/s -0.05/15 8 9.Be3 Nd4 2268kN/s
>>>>>>>>>>>>-0.06/16 4 10.Nh4 (Rc1) 10...g5 1924kN/s -0.11/14 6 11.Nf3 c6 2174kN/s
>>>>>>>>>>>>0.00/14 0 12.Ne4 g4 2152kN/s 0.19/15 6 13.Nxd4 exd4 2016kN/s 0.00/16 0
>>>>>>>>>>>>14.Bf4 (Bd2) 14...Nd5 1921kN/s -0.02/16 6 15.Bd2 (Qd2) 15...Rg8 2055kN/s
>>>>>>>>>>>>0.09/14 8 16.Rc1 Qb6 2188kN/s 0.00/16 0 17.Qa4 (b4) 17...a5 2048kN/s
>>>>>>>>>>>>0.13/14 6 18.a3 f5 2393kN/s -0.08/15 2 19.Ng5 Bd7 2324kN/s 0.00/15 0
>>>>>>>>>>>>20.b4 (Nf7) 20...Bh6 2301kN/s -0.19/13 15 21.f4 gxf3 2222kN/s -0.18/14 0
>>>>>>>>>>>>22.Nxf3 Be3+ 2349kN/s -0.22/14 8 23.Bxe3 (Kh1) 23...Nxe3 2014kN/s -0.37/12
>>>>>>>>>>>>3 24.Rf2 f4 2619kN/s -0.33/14 3 25.gxf4 (Rc5) 25...Bh3 2673kN/s -0.81/13 9
>>>>>>>>>>>> 26.Ne1 Qd8 2779kN/s -0.51/14 13 27.Qb3 (Rc5) 27...Rg6 2880kN/s -0.64/13 17
>>>>>>>>>>>> 28.Rc5 (f5) 28...a4 2781kN/s -0.67/12 9 29.Qa2 Qf6 2806kN/s -0.66/13 14
>>>>>>>>>>>>30.Rg5 (b5) 30...Be6 2238kN/s -0.46/15 16 31.Rxg6 Qxg6 2227kN/s -0.59/15 11
>>>>>>>>>>>> 32.Qd2 (Qb2) 32...Bd5 2270kN/s -0.49/16 24 33.f5 Qg4 2529kN/s -0.27/16 20
>>>>>>>>>>>>34.b5 0-0-0 2814kN/s -0.21/14 7 35.Qa5 (bxc6) 35...Rg8 2948kN/s -0.62/13 7
>>>>>>>>>>>>36.bxc6 (Qa8+) 36...Bxc6 2885kN/s -0.15/14 23 37.Qc5 (Qe5) 37...Qg5 2636kN/s
>>>>>>>>>>>>-0.55/14 7 38.Nf3 Qg7 2690kN/s -0.60/14 1 39.Ne1 Qf6 2500kN/s 0.00/16 0
>>>>>>>>>>>>40.Nf3 h5 2564kN/s -0.75/15 0 41.h4 (Kh1) 41...Rg3 2683kN/s -0.47/13 6
>>>>>>>>>>>>42.Kh2 Rg7 2734kN/s -0.45/14 0 43.Bh3 Ng4+ 2351kN/s -0.20/15 0 44.Bxg4
>>>>>>>>>>>>hxg4 2413kN/s -0.49/15 4 45.Ng5 Rxg5 2568kN/s -0.14/15 14 46.hxg5 Qxg5
>>>>>>>>>>>>2699kN/s -0.23/15 8 47.Kg1 g3 2802kN/s -0.28/16 8 48.Rg2 Qe3+ 2526kN/s
>>>>>>>>>>>>-0.52/16 22 49.Kf1 Kd7 3054kN/s 0.00/17 24 50.Rxg3 (Qc4) 50...Qf4+ 2399kN/s
>>>>>>>>>>>> -0.67/14 3 51.Rf3 Bxf3 2273kN/s 0.00/15 0 52.exf3 Qxf3+ 2139kN/s 0.00/15
>>>>>>>>>>>>0 53.Kg1 Qg3+ 2184kN/s -0.37/14 2 54.Kh1 Qd6 2666kN/s -0.11/14 20 55.Qa7
>>>>>>>>>>>>(Qb5+) 55...Qh6+ 2006kN/s -0.37/14 4 56.Kg2 (Kg1) 56...Qd2+ 2173kN/s
>>>>>>>>>>>>-0.37/14 3 57.Kg1 Qe3+ 2338kN/s -0.37/15 0 58.Kg2 (Kh1) 58...Qg5+ 2401kN/s
>>>>>>>>>>>>-0.07/14 9 59.Kh2 Qh4+ 2340kN/s -0.07/14 4 60.Kg2 Qg4+ 2334kN/s 0.00/16 0
>>>>>>>>>>>> 61.Kh2 (Kh1) 61...Qe2+ 2860kN/s 0.00/15 20 62.Kg1 (Kh1) 62...Qd1+ 2315kN/s
>>>>>>>>>>>>0.06/13 2 63.Kf2 Qc2+ 2867kN/s 0.00/14 0 64.Ke1 (Kg1) 64...Qc1+ 2631kN/s
>>>>>>>>>>>>0.00/12 4 65.Kf2 Qf4+ 2656kN/s 0.00/13 0 66.Kg2 Ke7 2760kN/s 0.00/15 0
>>>>>>>>>>>>67.Qxb7+ Kf6 2193kN/s #1/16 0 68.Qc6+ (Qe4) 68...Kxf5 1880kN/s 0.00/16 8
>>>>>>>>>>>>69.Qc8+ Kg6 1704kN/s 0.00/17 13 70.Qg8+ Kh5 1559kN/s #1/18 0 71.Qd5+
>>>>>>>>>>>>(Qh8+) 71...Qg5+ 2119kN/s 0.00/18 2 72.Qxg5+ Kxg5 0kN/s 0.00/3 0 73.Kg3
>>>>>>>>>>>>Kf5 398kN/s #20/62 0 74.Kf3 Ke5 93kN/s #20/62 0 75.Kg4 Kd5 1135kN/s
>>>>>>>>>>>>#20/61 0 76.Kf4 Kc5 1792kN/s #20/61 0 77.Ke5 Kc6 1521kN/s #20/62 0
>>>>>>>>>>>>78.Kxd4 Kd6 #20/1 0 79.Kc4 Kc6 #18/1 0 80.Kb4 Kd5 #17/1 0 81.Kxa4 Kc5
>>>>>>>>>>>>#16/1 0 82.Ka5 Kc6 #16/1 0 83.d4 Kd5 #14/1 0 84.Kb6 Kxd4 #13/1 0
>>>>>>>>>>>>85.a4 Kc4 #12/1 0 86.a5 Kd5 #11/1 0 87.a6 Kc4 #10/1 0 88.a7 Kd4 #9/1
>>>>>>>>>>>>0 89.a8Q Ke3 #9/1 0 90.Qf8 Ke4 #7/1 0 91.Kc6 Kd3 #6/1 0 92.Qf3+ Kd2
>>>>>>>>>>>>#5/1 0 93.Kd5 Ke1 #4/1 0 94.Qg2 Kd1 #4/1 0 95.Kd4 Kc1 #2/1 0 96.Kc3
>>>>>>>>>>>>Kb1 #1/1 0 97.Qb2# 1-0
>>>>>>>>>>>>
>>>>>>>>>>>>And PGN:
>>>>>>>>>>>>
>>>>>>>>>>>>[Event "FICS rated blitz game"]
>>>>>>>>>>>>[Site "FICS, Fremont, California USA"]
>>>>>>>>>>>>[Date "2002.05.08"]
>>>>>>>>>>>>[Time "00:42:28"]
>>>>>>>>>>>>[Round "-"]
>>>>>>>>>>>>[White "tjfritz"]
>>>>>>>>>>>>[Black "Project"]
>>>>>>>>>>>>[WhiteElo "2450"]
>>>>>>>>>>>>[BlackElo "2579"]
>>>>>>>>>>>>[TimeControl "300+5"]
>>>>>>>>>>>>[Mode "ICS"]
>>>>>>>>>>>>[Result "1-0"]
>>>>>>>>>>>>
>>>>>>>>>>>>1. c4 Nf6 2. Nc3 e5 3. Nf3 Nc6 4. g3 d5 5. cxd5 Nxd5 6. Bg2 Nb6 7. O-O Be6
>>>>>>>>>>>>8. d3 f6 9. Be3 Nd4 10. Nh4 g5 11. Nf3 c6 12. Ne4 g4 13. Nxd4 exd4 14. Bf4
>>>>>>>>>>>>Nd5 15. Bd2 Rg8 16. Rc1 Qb6 17. Qa4 a5 18. a3 f5 19. Ng5 Bd7 20. b4 Bh6
>>>>>>>>>>>>21. f4 gxf3 22. Nxf3 Be3+ 23. Bxe3 Nxe3 24. Rf2 f4 25. gxf4 Bh3 26. Ne1
>>>>>>>>>>>>Qd8 27. Qb3 Rg6 28. Rc5 a4 29. Qa2 Qf6 30. Rg5 Be6 31. Rxg6 Qxg6 32. Qd2
>>>>>>>>>>>>Bd5 33. f5 Qg4 34. b5 O-O-O 35. Qa5 Rg8 36. bxc6 Bxc6 37. Qc5 Qg5 38. Nf3
>>>>>>>>>>>>Qg7 39. Ne1 Qf6 40. Nf3 h5 41. h4 Rg3 42. Kh2 Rg7 43. Bh3 Ng4+ 44. Bxg4
>>>>>>>>>>>>hxg4 45. Ng5 Rxg5 46. hxg5 Qxg5 47. Kg1 g3 48. Rg2 Qe3+ 49. Kf1 Kd7 50. Rxg3
>>>>>>>>>>>>Qf4+ 51. Rf3 Bxf3 52. exf3 Qxf3+ 53. Kg1 Qg3+ 54. Kh1 Qd6 55. Qa7 Qh6+ 56. Kg2
>>>>>>>>>>>>Qd2+ 57. Kg1 Qe3+ 58. Kg2 Qg5+ 59. Kh2 Qh4+ 60. Kg2 Qg4+ 61. Kh2 Qe2+ 62. Kg1
>>>>>>>>>>>>Qd1+ 63. Kf2 Qc2+ 64. Ke1 Qc1+ 65. Kf2 Qf4+ 66. Kg2 Ke7 67. Qxb7+ Kf6 68. Qc6+
>>>>>>>>>>>>Kxf5 69. Qc8+ Kg6 70. Qg8+ Kh5 71. Qd5+ Qg5+ 72. Qxg5+ Kxg5 73. Kg3 Kf5
>>>>>>>>>>>>74. Kf3 Ke5 75. Kg4 Kd5 76. Kf4 Kc5 77. Ke5 Kc6 78. Kxd4 Kd6 79. Kc4 Kc6
>>>>>>>>>>>>80. Kb4 Kd5 81. Kxa4 Kc5 82. Ka5 Kc6 83. d4 Kd5 84. Kb6 Kxd4 85. a4 Kc4
>>>>>>>>>>>>86. a5 Kd5 87. a6 Kc4 88. a7 Kd4 89. a8=Q Ke3 90. Qf8 Ke4 91. Kc6 Kd3 92. Qf3+
>>>>>>>>>>>>Kd2 93. Kd5 Ke1 94. Qg2 Kd1 95. Kd4 Kc1 96. Kc3 Kb1 97. Qb2#
>>>>>>>>>>>>{Black checkmated} 1-0
>>>>>>>>>>>
>>>>>>>>>>>This is obviously some kind of bug in Junior, possibly induced by lack of time
>>>>>>>>>>>to "think".
>>>>>>>>>>
>>>>>>>>>>I do not think that it is a bug.
>>>>>>>>>>It is a typical computer mistake.
>>>>>>>>>>
>>>>>>>>>>computers do not understand that the pawn endgame is lost before trading queens
>>>>>>>>>>Tablebases cannot help much here because if they can see that white win a pawn
>>>>>>>>>>they will not let trade queens but winning the pawn is too deep for them to see
>>>>>>>>>>because of typical bad search rules(they considers the line Qg5+ Qxg5 Kxg5 only
>>>>>>>>>>for a very short time) and before winning the pawn the position seems to be
>>>>>>>>>>better for black because the black pawns are more advanced then the white pawns
>>>>>>>>>>and the black king is also closer to the cntre of the board.
>>>>>>>>>>
>>>>>>>>>>Uri
>>>>>>>>>
>>>>>>>>>Junior's file shows that it looked 18 ply ahead when selecting 71...Qg5+? The
>>>>>>>>>pawn on d4 was lost 14 ply later. So which of the very obvious king advances by
>>>>>>>>>White did it prune out of the search, and why would that not represent a bug,
>>>>>>>>>meaning unintended behavior by the program?
>>>>>>>>>
>>>>>>>>>Regards,
>>>>>>>>>Vine
>>>>>>>>
>>>>>>>>
>>>>>>>>Junior depth does not mean plies.
>>>>>>>>depth 18 of Junior means only 9 plies of brute force depth.
>>>>>>>>
>>>>>>>>There are rare cases when Junior cannot see simple tactics when this is not a
>>>>>>>>question of time but this does not seem to be one of thee cases because if you
>>>>>>>>give Junior to play at long time control(for example 10 hours/20 moves) then
>>>>>>>>Junior can fail low at depth 21.
>>>>>>>>
>>>>>>>>I did not wait to see how much time it needs to find Kh4 and maybe someone with
>>>>>>>>faster hardware can do it.
>>>>>>>>
>>>>>>>>Uri
>>>>>>>
>>>>>>>Faster hardware is not necessary. On my Celeron 433, Junior 7 takes a disgusting
>>>>>>>4 minutes and 15 seconds to both see something is wrong with Qg5+ and change to
>>>>>>>moving the king. On this same machine, some other much better times:
>>>>>>
>>>>>>Here is what happens on my p800
>>>>>>Junior in analysis mode can see that something is wrong with Qg5+ at depth 19
>>>>>>
>>>>>>
>>>>>>Crafty 18.10 - Blass,U
>>>>>>[D]8/8/8/3Q3k/p2p1q2/P2P4/6K1/8 b - - 0 1
>>>>>>
>>>>>>Analysis by Junior 7:
>>>>>>
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2
>>>>>> = (-0.19) Depth: 3 00:00:00
>>>>>>1...Kh4 2.Qd8+ Kh5
>>>>>> = (0.00) Depth: 6 00:00:00 6kN
>>>>>>1...Kh4 2.Qd8+ Kh5
>>>>>> = (0.00) Depth: 6 00:00:00 6kN
>>>>>>1...Kh4 2.Qd8+ Kh5
>>>>>> = (0.00) Depth: 6 00:00:00 6kN
>>>>>>1...Kh4 2.Qd8+ Kh5
>>>>>> = (0.00) Depth: 6 00:00:00 6kN
>>>>>>1...Kh4 2.Qf3 Qd2+ 3.Qf2+ Qxf2+ 4.Kxf2 Kg4 5.Ke1 Kg3 6.Kd2
>>>>>> = (-0.10) Depth: 9 00:00:00 99kN, tb=9
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2 Ke6 5.Kd2 Kd5
>>>>>> = (-0.15) Depth: 9 00:00:00 99kN, tb=9
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2 Ke6 5.Kd2 Kd5 6.Kc2
>>>>>> = (-0.20) Depth: 12 00:00:00 162kN, tb=17
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Kg3 Ke5 5.Kf3
>>>>>> = (0.00) Depth: 15 00:00:04 1819kN, tb=476
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Kg3 Ke5 5.Kf3
>>>>>> = (0.00) Depth: 17 00:00:17 5523kN, tb=3099
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kg3 Kf5 4.Kf3 Ke5 5.Kg4 Kd6
>>>>>> = (0.06) Depth: 19 00:00:44 14829kN, tb=10707
>>>>>>1...Kh4 2.Qd8+ Kg4 3.Qd7+ Kh4
>>>>>> = (0.00) Depth: 19 00:03:55 97484kN, tb=52801
>>>>>>
>>>>>>(Blass, Tel-aviv 09.05.2002)
>>>>>>
>>>>>>Here is what happens at 120/40 time control when Junior does not want a draw and
>>>>>>evaluates draws as 0.15 pawns against itself so it does not understand that Kh4
>>>>>>is better than Qg5+ because both moves lead to draw if you use Junior point of
>>>>>>view.
>>>>>>
>>>>>>Blass,U - Junior 7
>>>>>>8/8/8/3Q3k/p2p1q2/P2P4/6K1/8 b - - 0 1
>>>>>>
>>>>>>Analysis by Junior 7:
>>>>>>
>>>>>>1...Kg6 2.Qc6+ Kf5 3.Qxa4
>>>>>> = (0.10) Depth: 3 00:00:00
>>>>>>1...Kh4 2.Qd8+ Qg5+ 3.Qxg5+ Kxg5
>>>>>> = (-0.14) Depth: 3 00:00:00
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2
>>>>>> = (-0.19) Depth: 3 00:00:00
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2 Kf4
>>>>>> ² (0.28) Depth: 6 00:00:00 1kN
>>>>>>1...Kh4 2.Qd8+ Kg4 3.Qd7+ Kg5 4.Qd8+ Qf6 5.Qa5+ Kf4 6.Qxa4
>>>>>> = (0.09) Depth: 6 00:00:00 8kN
>>>>>>1...Kh6 2.Qe6+ Kg7 3.Qd7+ Kf6 4.Qd8+
>>>>>> = (0.04) Depth: 6 00:00:00 33kN, tb=2
>>>>>>1...Kh6 2.Qe6+ Kg7 3.Qd7+ Kf6 4.Qd8+ Ke6 5.Qe8+ Kd5 6.Qb5+ Ke6
>>>>>> = (0.00) Depth: 9 00:00:00 161kN, tb=9
>>>>>>1...Kh4 2.Qf3 Qd2+ 3.Qf2+ Qxf2+ 4.Kxf2 Kg4 5.Ke2 Kf4 6.Kd2
>>>>>> = (-0.09) Depth: 9 00:00:00 207kN, tb=14
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2 Ke5 5.Kd2 Kd5
>>>>>> = (-0.15) Depth: 9 00:00:00 214kN, tb=15
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Ke2 Ke5 5.Kd2 Kd5 6.Kc2
>>>>>> = (-0.20) Depth: 12 00:00:00 248kN, tb=15
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Kg3 Ke5 5.Kf3 Kd5
>>>>>> = (0.06) Depth: 15 00:00:05 2364kN, tb=710
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Kg3 Ke5 5.Kf3 Kd5 6.Kf4 Kc6 7.Kf5 Kb5 8.Ke5
>>>>>> = (0.06) Depth: 18 00:00:14 5750kN, tb=2565
>>>>>>1...Qg5+ 2.Qxg5+ Kxg5 3.Kf3 Kf5 4.Kg3 Ke5 5.Kf3
>>>>>> = (0.00) Depth: 20 00:00:34 13620kN, tb=9657
>>>>>>1...Qg5+ 2.Qxg5+
>>>>>> ² (0.45) Depth: 21 00:02:11 62803kN, tb=39566
>>>>>>1...Kh4 2.Qd8+ Kg4 3.Qd7+ Kg5
>>>>>> = (0.00) Depth: 21 00:11:41 383694kN, tb=159805
>>>>>>
>>>>>>(Blass, Tel-aviv 09.05.2002)
>>>>>>
>>>>>>If I try Junior at 600/40 it may jump directly from 18 to 21 and again I guess
>>>>>>that I need to wait a long time to get Kh4.
>>>>>>
>>>>>>
>>>>>>>
>>>>>>>Phalanx XXII: 3 seconds to see the pawn on d4 falling, 25 seconds to move the
>>>>>>>king though with no draw score
>>>>>>>Pharaon 2.35 (an old beta): 9 seconds to see both that Qg5+ loses and to move
>>>>>>>the king with a draw score
>>>>>>>Yace 0.99.57: 12 seconds to see that Qg5+ is kind of bad (like -.8 or something)
>>>>>>>and move the king (don't recall if it gave a draw score)
>>>>>>>Fritz 6: 15 seconds with details like Yace above
>>>>>>>A huge host of other Winboard engines: 20-40 seconds to move the king
>>>>>>>Some good engines that took longer: Little Goliath 3.5, Nejmet 3.05
>>>>>>>
>>>>>>>So Junior 7 is not searching this kind of position correctly, even given the
>>>>>>>typical weaknesses of programs in evaluating certain endings. And of course I'm
>>>>>>>quite aware that 18 ply is not brute force, that's why I specifically asked
>>>>>>>which of the obvious king advances was being pruned out of the selective search
>>>>>>>-- a legitimate question I think.
>>>>>>
>>>>>>I think that the question should be what is the move that Junior
>>>>>>does not extend in the line.
>>>>>>
>>>>>>Qg5+ Qxg5 Kxg5 Kg3 Kf5 Kf3 Ke5 Kg4 Ke6 Kf4 Kd5 Kf5 Kc5 Ke5.
>>>>>>
>>>>>>You can see that there are 11 quiet king moves in this line.
>>>>>>
>>>>>>Junior only search depth 9+extensions.
>>>>>>
>>>>>>After Kxg5 Junior can see positive score of more than +0.15 at depth 24 and
>>>>>>cannot see it at depth 21 so I guess that Junior extend transition to pawn
>>>>>>endgame but it seems not to be enough when it does not know which moves to
>>>>>>extend in the pawn endgame.
>>>>>>
>>>>>>Regards,
>>>>>>Uri
>>>>>
>>>>>Well, it's not surprising that there are 11 quiet king moves after the queens
>>>>>are exchanged, because the only pieces left to move after that are kings! You
>>>>>can see from the Junior search lines you posted what knowledge has been left out
>>>>>-- the concept of the opposition, the basis of all pawn endgames. Instead, it
>>>>>steers the White king toward the pawns with 3.Kf3 after the queen exchange,
>>>>>which prevents it from seeing any problems, likely until the brute force portion
>>>>>of the search gets close enough for non-selective extensions to see the pawn on
>>>>>d4 falling. I don't believe Junior really doesn't know about the opposition,
>>>>>though -- instead, I think what we're seeing here is either an unwise shortcut,
>>>>>like the opposition code is not used in searching when queens are on, even if
>>>>>the search extends into a position with queens off, or a bug, a malfunction in
>>>>>how the programmers intended Junior to search.
>>>>>
>>>>>Regards,
>>>>>Vine
>>>>
>>>>
>>>>I believe that Junior does not know about the opposition because Amir does not
>>>>consider knowledge about pawn endgame as important knowledge and believe that it
>>>>is usually solved by search.
>>>>
>>>>
>>>>Here is the analysis of Junior after trading queens
>>>>Junior can find Kg3 only at depth 21.
>>>>
>>>>
>>>>Blass,U - Chess Tiger 14.0
>>>>[D]8/8/8/6k1/p2p4/P2P4/6K1/8 w - - 0 1
>>>>
>>>>Analysis by Junior 7:
>>>>
>>>>3.Kh1 Kf4
>>>> ³ (-0.27) Depth: 3 00:00:00
>>>>3.Kf3 Kf5
>>>> = (-0.09) Depth: 3 00:00:00
>>>>3.Kf3 Kf5 4.Ke2
>>>> = (-0.20) Depth: 6 00:00:00
>>>>3.Kf3 Kf5 4.Ke2 Kf4 5.Kd2
>>>> = (-0.20) Depth: 9 00:00:00
>>>>3.Kf3 Kf5 4.Ke2 Ke6 5.Kd2 Kd5
>>>> = (-0.15) Depth: 12 00:00:00 1kN
>>>>3.Kf3 Kf5 4.Ke2 Ke6 5.Kd2 Kd5 6.Kc2 Kc5
>>>> = (-0.15) Depth: 15 00:00:00 2kN
>>>>3.Kf3 Kf5 4.Kg3 Ke5 5.Kf3
>>>> = (0.00) Depth: 18 00:00:00 7kN, tb=11
>>>>3.Kf3 Kf5 4.Kg3 Kg5
>>>> = (0.00) Depth: 21 00:00:00 13kN, tb=40
>>>>3.Kg3 Kf5 4.Kf3 Ke5 5.Kg4 Kd6
>>>> = (0.06) Depth: 21 00:00:00 15kN, tb=60
>>>>3.Kg3 Kf5 4.Kf3 Ke5 5.Kg4 Kd6 6.Kf4 Kd5 7.Kf5 Kd6 8.Ke4 Kc5
>>>> = (0.11) Depth: 24 00:00:00 20kN, tb=82
>>>>3.Kg3!
>>>> ² (0.41) Depth: 27 00:00:00 28kN, tb=157
>>>>3.Kg3!
>>>> ± (0.71) Depth: 30 00:00:00 45kN, tb=280
>>>>3.Kg3! Kf5 4.Kf3 Ke5 5.Kg4 Kd6 6.Kf5 Kd5 7.Kf4 Kc5 8.Ke5 Kc6 9.Kxd4
>>>> +- (#20) Depth: 30 00:00:00 53kN, tb=313
>>>>3.Kg3! Kf5 4.Kf3 Ke5 5.Kg4 Kd6 6.Kf5 Kd5 7.Kf4 Kc5 8.Ke5 Kc6 9.Kxd4
>>>> +- (#20) Depth: 30 00:00:00 56kN, tb=328
>>>>
>>>>(Blass, Tel-aviv 09.05.2002)
>>>>
>>>>Uri
>>>
>>>Then Ban (what about Bushinsky?) has made a bad decision. Junior will
>>>misevaluate entry into pawn endings even at tournament time controls on fast
>>>hardware when sufficiently distant from positions where opposition counts. Tiger
>>>obviously knows something about opposition, at least not to prune opposition
>>>moves out of its search, and it's not by any means a slow searcher overburdened
>>>with useless knowledge.
>>
>>Junior probably does not prune the moves out of it search but simply does not
>>extend them.
>>
>>There are usually 3 kind of moves:
>>1)moves that the program extends(for example programs extend checks)
>>2)moves that the program does not extend and does not prune
>>3)moves that the program prunes(for example it may be stupid sacrifices when the
>>remaining depth is small)
>>
>>I believe that the king moves that Junior does not extend are in group 2 and not
>>in group 3.
>>
>>I do not know if tiger has knowledge to extend the king moves that were
>>discussed but I guess that it has no knowledge about opposition.
>>
>>If it knows to extend the king moves then it may be for different reasons.
>>
>>Maybe christophe and Amir can respond if my guess is correct.
>>
>>Uri
>
>Well, forget about the question of "knowledge" for the moment, and just look at
>the times I posted from Junior 7 and other programs to avoid Qg5+ (since every
>king move is a draw). Don't you think something is wrong when not only top
>commercial programs like Tiger, Rebel, Fritz (ver 6!), but also many Winboard
>programs, not even the very best ones like Fritz and Crafty, but even second
>tier ones are solving this position in a fraction of the time it took Junior (4
>minutes 15 seconds)? Some of them didn't know exactly what they were doing,
>judging from their evaluations, but others clearly did. Junior doesn't just lack
>knowledge here, something is hampering its search of the pawn endgame versus
>other programs, not all or even most of which may have any code for detecting
>the opposition. I still think Junior is doing something wrong in this position
>which the programmers did not intend.
>
>Vine
All version of Junior had problems with pawn endgames(I do not know about
Junior6 but I know about Junior4.6 and Junior5).
The situation was clearly worse with Junior4.6 when it could not solve a
position that every top program could solve by hash tables
(I suspect that it used hash table only for better order of moves and not to
prune the tree but I do not know).
The truth is that other winboard engines use different search algorithm and
Junior's search algorithm is not good for pawn endgames.
I also suspect that Junior still does not use hash tables in an efficient way
and it can explain some problems in pawn endgames.
Junior also have almost no evaluation about pawn endgame and I post another
example to prove it.
Junior does not evaluate unstoppable pawns in the endgame and here is the
proof(Junior find by search that 1.a4 is winning but it has clearly negative
evaluation at small depth when other top programs have no problem to see that
white is better from the first ply).
New position
[D]8/2ppp3/8/8/8/8/P2K4/7k w - - 0 1
Analysis by Junior 7:
1.a4 Kg2
µ (-1.27) Depth: 3 00:00:00
1.a4 Kg2 2.Kd3 Kf2
-+ (-1.55) Depth: 6 00:00:00
1.a4!
µ (-1.25) Depth: 9 00:00:00 2kN
1.a4! Kg2 2.a5 Kf3 3.a6 Ke4 4.a7 Kd4 5.a8Q
+- (2.72) Depth: 9 00:00:00 13kN
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.