|
< Earlier Kibitzing · PAGE 7 OF 15 ·
Later Kibitzing> |
Dec-05-16
 | | AylerKupp: <zanzibar> Not much work on my part, most of it was done by Stockfish 8, and it neither gets tired or complains. :-) But, yeah, I don't trust ANY computer line until I've checked it manually to verify that it does not adversely suffer from the horizon effect. This can be done by sliding forward but that requires some time as well as thought, both of which I usually have in short supply. So, as a shortcut I do what I call "leaping" forward; I simply restart the analysis at the position given by the end of the line. My objective is not, as in sliding forward, to find the best line but just to see if the engine's initial evaluation was reasonable. If after leaping forward the resulting evaluation of the new final position is similar to the original evaluation's final position, then I believe that I can conclude that the initial evaluation was reasonable. But if the two evaluations are greatly different, then I know that the original evaluation was suspect (possibly as a result of too aggressive pruning of the search tree) and that forward sliding is needed to determine where the engine went wrong. I can then do a binary-like search (restarting the analysis at the position after 1/2 of the moves were played) and checking that evaluation, and then if necessary restarting the analysis again after either 1/4 or 3/4 of the moves were played until I find the engine's error. As far as the lack of symmetry after 1.e4 e6 2.d4 d5 vs. 1.d4 e6 2.e4 Nf6 remember that chess engines are notoriously non-deterministic; if you run an analysis of a given position on the same computer and to the same depth, both the line evaluations and move ranking are very likely to be different, particularly if during transpositions the evaluations are close. And if the evaluations are close then the move rankings will easily change at each search ply. So, after 1.d4 e6, Stockfish 8's evaluations of 2.c4 and 2.e4 were very close (and, in fact, both 2.Nf3 and 2.Bf4 crept in as one of Stockfish 8's top 3 choices), so whichever of White's second move is evaluated the highest highly depends on the search depth. For example, at d=30 the evaluations were 2.c4 [+0.09], 2.e4 [+0.06], and 2.Nf3 [0.00]. The [0.03] centipawn evaluation difference between 2.c4 and 2.e4 has no practical significance but the engine can only select one move as being "best" so it selects the one with the highest evaluation, 2.c4 in this case. And, if two (or more) moves having the same evaluation, it typically selects either the first of those moves found in its search tree or the last move found in its search tree, so the selection is also engine dependent. And at d=31 the move rankings were reversed; 2.e4 was evaluated at [+0.04] and 2.c4 was evaluated at [+0.02] so again the evaluation differences were not significant. But at this search depth Stockfish 8 evaluated 2.Nf3 as better than either of them, [+0.10]. Again an insignificant difference but Stockfish had to pick one of the 3 moves as being "best", so at this depth it chose 2.Nf3. And at d=32 2.c4 was once again on top with an evaluation of [+0.12] compared to 2.Nf3's [+0.10] and 2.e4's [+0.09]. I suspect that similar considerations arise after 1.e4 e6 2.d4 but I'm too lazy to check it out. It's like the weather in some parts of the world. If you don't like it at this moment, wait a little while, it will be different. For example, if you want to find out the weather patterns in Zanzibar, check this site: https://www.expertafrica.com/zanzib.... Particularly convenient to know if you want to time your visit to the wildebeest and other migrations. |
|
| Dec-05-16 | | zanzibar: <So, as a shortcut I do what I call "leaping" forward; I simply restart the analysis at the position given by the end of the line.> I like to do walkthroughs - i.e. rather than trust the algorithm to find a line all the way out to n-ply, let it go to (n-m)-ply. Take the best move and repeat. I think this is a more effective way to run for a fixed about of time. I agree with most of the commentary about the asymmetry, it was an funny point to note about how engines work. OK, thanks for the reply - I really got a kick out of the last paragraph about Zanzibar! |
|
Dec-06-16
| | AylerKupp: <zanzibar> Oh, I agree about restarting the engine at (n-m)-ply, that's the basic forward sliding approach. And, if you make m = n -1 than that's like having an engine (or different engines) play a game or doing forward sliding one move at a time. In a recent Team game I named this method a "creeping barrage" in reference to trench warfare during WW I when a teammate used it to try to find the best move from any given position. Yes, I figured that you would like the reference to Zanzibar. :-) |
|
| Dec-06-16 | | zanzibar: Maybe this is ignorant, but what is the optimal strategy to use an engine for actual game play? There must be a better strategy than to just divide the time limit by the number of moves and uniformly play out the game, regardless of complications on the board. (I obviously realize that computers burn the road up calculating tactics - which is the typical challenge for humans). |
|
| Dec-06-16 | | Appaz: <zanzibar> You can read a little about it here: https://chessprogramming.wikispaces... |
|
| Dec-06-16 | | zanzibar: <Appaz> I just took a look, but there's little, if any, discussion of the actual algorithms used - specifically for engine play. There is mention of Hyatt's formula for Cray Blitz (somewhat an old example), and it seems to just use the simple minded approach mentioned above(*). (*) With the finesse of factoring out the book moves, an obvious improvement. |
|
| Dec-06-16 | | zanzibar: I wonder if engines have any idea of criticality - e.g. if there could be a metric suggesting a position deserves extra time to ponder vs. other positions. |
|
| Dec-06-16 | | Appaz: <zanzibar> The basic algorithm is probably very simple, as you said, TimeLeft/NumberOfMovesToTimeControl, but obviously with some adjusting parameters. For example, I would expect that if the PV gives a worse score than the current position, then more time is used. Once I saw a nice article on this but I couldn't find it right now. Things will also be complicated when the engine is running several threads and Sf had a bug in TCEC Season 8 were it used some milliseconds to much because of the overhead when killing its threads and lost on time. Google "stockfish time management bug tcec 8" to read more about this. This is a science of its own and good time managment will be a major factor for an engines playing strength. Stockfish has had some major improvements in this area the two last years, so you should be able to find some stuff over at the support and discussion pages for it. |
|
| Dec-07-16 | | zanzibar: I suppose another simple-minded approach would be to somehow fold the calculation depth into a feedback loop of some kind or another. Not sure how much time/energy I have for pursuing the material in the links <Appaz> mentions (thanks), but it struck my curiosity for a moment. Reminds me a little of the problem of scheduling in computer OS theory. |
|
Mar-27-17
 | | tamar: Stockfish currently beating Houdini 4 Pro
at odds of two pawns (a and h pawn versus Houdini without castling rights) I would not have thought two pawns could be overcome, but the engine makes almost instant use of the open files!
http://www.viewchess.com/cbreader/2... |
|
| Mar-27-17 | | blackdranzer: Nerds ! |
|
| Mar-27-17 | | Appaz: Thanks! |
|
| Jun-25-17 | | scholes: Latest version of stockfish and Komodo are playing a 150 games match on 32 core hardware at time control 60 minutes + 15 seconds. Games are broadcast live here
http://www.fastgm.de/pgn4web/live1.... |
|
Jun-25-17
 | | offramp: It's good to see that Houdini is completely to the vesher's around taken beink. At the moment I think it is winning +21 -6 with about a thousand 150-move draws. And Stockfish is the FREE program and Houdini is the one that costs $$$£££$$$!!! |
|
Jun-26-17
| | offramp: There is a lot of unusual/mind-bending chess in these games against Houdini.
Here is one oddity.
[Event "Test"]
[Site "FastGM"]
[Date "2017.06.25"]
[Round "102"]
[White "Komodo 11.01 64-bit"]
[Black "Stockfish 170606 x64"]
[Result "1/2-1/2"]
[ECO "C41"]
[WhiteElo "3000"]
[BlackElo "3000"]
[PlyCount "87"]
[EventDate "2017.??.??"]
[TimeControl "3600+15"]
1. e4 e5 2. Nf3 d6 3. d4 exd4 4. Nxd4 Nf6 5. Nc3 Be7 6. Bf4 0.38/27 123 Nc6
-0.38/31 188 7. Qd2 0.42/29 119 O-O -0.35/35 321 8. O-O-O 0.23/28 60
Nxd4 -0.37/37 373 9. Qxd4 0.36/28 47 Be6 -0.43/33 209 10. Kb1 0.41/28 75
a6 -0.31/31 63 11. Bc4 0.35/29 202 Nd7 -0.17/28 57 12. Nd5 0.40/29 208
b5 -0.14/31 166 13. Be2 0.48/27 42 Re8 -0.24/26 11 14. Nxe7+ 0.49/29
62 Qxe7 -0.32/33 195 15. Rhe1 0.47/29 92 Rab8 -0.33/31 137 16. Bg3 0.
48/30 180 f6 -0.25/33 156 17. f4 0.60/29 99 a5 -0.08/31 42 18. Bf3 0.
51/30 159 a4 -0.12/31 55 19. Qb4 0.47/28 131 Bf7 -0.08/29 45 20. Bg4
0.37/29 185 h6 -0.08/33 114 21. h4 0.21/28 126 Be6 0.00/31 34 22. Bf3
0.36/27 37 Bf7 0.00/33 52 23. h5 0.24/31 191 c5 0.00/34 54 24. Qd2 0.
23/29 41 Rb6 0.00/36 20 25. Bg4 0.16/31 130 Reb8 0.00/37 56 26. Bh4 0.
19/28 41 Qe8 0.00/34 72 27. Qe2 0.00/29 32 b4 0.00/29 13 28. Bf2 0.13/
30 65 a3 0.00/38 113 29. b3 0.00/31 40 Ne5 0.00/40 42 30. Bf5 0.00/32
42 Bxh5 0.00/30 77 31. g4 0.00/33 33 Bf7 -0.18/30 10 32. Bg3 0.00/30 29
Ng6 0.00/33 90 33. Qd2 0.07/31 130 Ne7 0.00/35 75 34. g5 0.00/32 52
Nxf5 0.00/36 12 35. exf5 0.00/33 38 Qc8 0.00/41 81 36. gxh6 0.00/35 47
gxh6 0.00/40 55 37. Bh4 0.00/37 46 Qxf5 0.00/43 49 38. Qg2+ 0.00/39 47
Kh8 0.00/43 16 39. Re7 0.00/42 59 Rg8 0.00/47 47 40. Qh1 0.00/42 65 Rg7
0.00/47 13 41. Qa8+ 0.00/44 82 Rg8 0.00/50 98 42. Qh1 0.00/45 94 Rg7
0.00/50 10 43. Qa8+ 0.00/41 54 Rg8 0.00/52 92 44. Qh1 0.00/50 122 3-fold
repetition 1/2-1/2
The last few moves:
click for larger view
40...Rg7
41. Qa8+.
click for larger view
41...Rg8
42. Qh1 Rg7
43. Qa8+ Rg8
44. Qh1 ½-½
Long tentacles! |
|
Jun-26-17
 | | Tabanus: Transposing lines:
1. [+0.06]: 1...e6 2.c4 Nf6 3.Nf3 b6 4.a3 Bb7 (as if 1.d4 e6 2.e4 is worse)
3. [+0.11]: 1...Nf6 2.c4 e6 3.Nf3 b6 4.a3 Bb7 |
|
| Jun-28-17 | | WorstPlayerEver: Stockfish (White) totally busts the Nimzo here: 1. d4 Nf6 2. c4 e6 3. Nf3 Bb4 4. Bd2 Qe7 5. g3 Nc6 6. e3 +0.31/34 189 6... Bxd2 -0.18/30 99 7. Nfxd2 +0.32/35 130 7... e5 -0.14/32 120 8. d5 +0.34/33 104 8... Nb8 -0.15/29 59 9. Nc3 +0.37/33 523 9... a5 -0.13/30 103 10. Bd3 +0.30/32 257 10... d6 -0.10/29 99 11. f3 +0.29/31 107 11... Bh3 -0.09/31 145 12. Qe2 +0.31/35 196 12... h5 -0.14/32 351 13. O-O-O +0.40/31 109 13... Na6 -0.14/31 54 14. Bc2 +0.45/32 169 14... g6 -0.19/28 36 15. a3 +0.26/33 333 15... Nc5 -0.13/32 147 16. Kb1 +0.22/33 225 16... Kf8 -0.11/30 125 17. e4 +0.22/33 77 17... Kg7 -0.11/27 53 18. Nb3 +0.23/32 294 18... Nxb3 -0.05/31 50 19. Bxb3 +0.23/35 135 19... c6 -0.05/32 43 20. Bc2 +0.20/35 83 20... cxd5 -0.08/31 155 21. exd5 +0.25/29 45 21... Qc7 0.00/31 43 22. Rhe1 +0.23/32 57 22... Rac8 -0.05/31 82 23. Bd3 +0.12/28 7 23... Nd7 0.00/32 38 24. Nb5 +0.33/33 80 24... Qb6 0.00/34 38 25. Qe3 +0.27/33 46 25... Rhd8 0.00/36 59 26. Qg5 +0.29/28 8 26... Nc5 0.00/38 95 27. Bc2 +0.24/35 98 27... Re8 0.00/39 134 28. Rg1 +0.31/31 24 28... Na6 0.00/39 112 29. g4 +0.58/32 35 29... hxg4 0.00/37 65 30. Bd3 +0.43/33 7 30... e4 0.00/38 122 31. fxe4 +0.54/35 46 31... Nc5 -0.09/38 182 32. Qe3 +0.77/32 22 32... Qd8 -0.14/36 76 33. Qd4 +0.81/31 98 33... Re5 -0.32/38 322 34. Bc2 +0.75/34 62 34... Qe7 -0.33/37 73 35. Rd2 +0.75/36 47 35... Kg8 -0.38/32 34 36. Nc3 +0.66/33 16 36... Nd7 -0.38/35 89 37. Rf2 +0.75/35 63 37... Nf6 -0.38/36 255 38. Bd3 +0.81/38 20 38... Nd7 -0.38/37 32 39. Ka2 +0.81/40 33 39... Nc5 -0.38/35 34 40. Rg3 +0.81/40 102 40... Qg5 -0.38/31 24 41. Rg1 +0.81/39 48 41... Rce8 -0.39/34 52 42. Rd1 +1.09/34 59 42... Ra8 -0.39/35 65 43. Rdd2 +1.13/34 15 43... Qe7 -0.45/33 42 44. Rf4 +1.03/38 49 44... Rc8 -0.51/32 74 45. Bc2 +0.96/36 95 45... Rf8 -0.54/31 25 46. Rff2 +0.96/35 32 46... Ra8 -0.65/33 191 47. Rf4 +0.96/38 35 47... Rf8 -0.71/29 29 48. Re2 +0.96/38 15 48... Kg7 -0.71/32 46 49. Qd2 +1.10/29 13 49... Qd8 -0.71/32 58 50. b3 +1.12/28 23 50... Nd7 -0.66/27 13 51. Nb5 +1.07/30 36 51... Qb6 -0.59/26 16 52. Qc3 +1.11/34 58 52... Nf6 -0.37/25 12 53. Bd3 +1.07/27 4 53... Nd7 -0.41/25 11 54. Rd2 +1.10/31 21 54... Qe3 -0.47/28 19 55. Rff2 +1.06/32 34 55... Qb6 -0.54/29 15 56. Bc2 +1.05/30 19 56... Nf6 -0.53/28 33 57. Kb2 +1.01/30 14 57... Rc8 -0.56/30 48 58. Bd3 +1.05/30 15 58... Nd7 -0.68/30 36 59. Rf4 +1.21/26 7 59... f6 -0.70/30 24 60. Nd4 +1.29/28 8 60... Nc5 -0.70/30 8 61. Ka2 +1.22/28 2 61... Rf8 -0.72/30 20 62. Re2 +1.21/32 21 62... Kg8 -0.76/28 13 63. Bc2 +1.29/32 19 63... Qd8 -0.78/30 54 64. Rff2 +1.29/31 30 64... Kf7 -0.81/28 14 65. Kb2 +1.21/32 15 65... Qb6 -0.81/31 18 66. Qe3 +1.35/26 6 66... Nd7 -0.81/28 7 67. Rd2 +1.27/24 2 67... Rh8 -0.82/32 27 68. Qg3 +1.30/31 12 68... Ke7 -0.83/30 15 69. Ne6 +1.51/25 8 69... g5 -0.82/31 15 70. Nd4 +1.49/30 13 70... Kf7 -0.82/28 8 71. Qe3 +1.49/33 15 71... Kg6 -0.83/28 8 72. Rde2 +1.49/32 12 72... Ra8 -1.05/28 25 73. Nb5 +1.49/34 16 73... Ra6 -1.08/28 28 74. Qg3 +1.49/34 15 74... Kg7 -0.98/27 15 75. Qd3 +1.49/36 13 75... Kf7 -1.02/26 14 76. Qc3 +1.49/34 32 76... Nf8 -0.92/27 21 77. Nd4 +1.49/35 11 77... Nd7 -1.02/26 15 78. Re3 +1.49/30 19 78... Ra8 -1.04/26 7 79. Kb1 +1.49/35 15 79... Rae8 -0.92/27 6 80. Ka2 +1.49/33 29 80... R8e7 -0.82/25 14 81. Rd2 +1.49/29 13 81... Kg7 -1.09/26 21 82. Rg3 +1.56/28 11 82... Kf8 -1.24/24 16 83. Qe3 +1.72/25 11 83... Re8 -1.32/25 20 84. Ne6 +1.79/26 7 84... Kg8 -1.55/27 17 85. Qxb6 +1.68/25 4 85... Nxb6 -1.48/28 8 86. Bd3 +1.81/28 12 86... Kf7 -1.43/26 16 87. b4 +2.23/28 29 87... Rc8 -1.52/26 18 88. bxa5 +2.86/25 7 88... Nd7 -2.45/28 12 89. Rb2 +2.97/28 13 89... Rxe6 -2.28/27 6 90. dxe6 +2.91/26 3 90... Kxe6 -2.30/28 6 91. Rxb7 +3.20/29 28 91... Nc5 -2.41/28 16 92. Rb5 +3.37/31 33 92... Rh8 -2.97/28 19 93. Kb2 +3.64/29 7 93... Kd7 -3.04/28 12 94. Kc3 +3.89/27 16 94... Kc7 -3.20/29 22 95. Rg1 +4.04/28 10 95... g3 -3.48/32 27 96. hxg3 +4.14/27 4 96... Bd7 -3.43/30 11 97. Bc2 +4.39/30 35 97... Rh3 -3.71/31 25 98. Rbb1 +4.49/28 8 98... Bc6 -3.75/29 7 99. Rbe1 +4.47/26 4 99... g4 -3.70/32 11 100. Re3 +4.64/28 10 100... Rh5 -3.65/29 10 101. Rf1 +4.84/29 17 101... Nd7 -3.76/31 7 102. Rf5 +5.04/26 7 102... Rh8 -4.01/28 11 103. a4 +5.06/24 3 103... Rh6 -4.30/28 16 104. Rf4 +5.26/32 8 104... Ne5 -4.68/27 14 105. a6 +6.22/27 33 105... Kb6 -5.08/27 20 106. Re1 +6.45/27 9 106... Kc7 -5.27/29 17 107. a7 +6.87/26 11 107... Rh3 -5.59/30 13 108. Rg1 +6.79/23 2 108... Kb6 -6.69/27 14 109. Rxf6 +7.50/25 10 109... Kxa7 -6.86/28 9 110. Rxd6 +7.62/22 3 110... Nf3 -7.38/29 21 111. Rb1 +9.36/26 9 111... Rxg3 -7.94/27 10 112. e5 +9.99/23 5 112... Rh3 -10.05/28 18 113. e6 +13.94/24 13 113... Be8 -11.87/28 42 114. Kb4 +30.13/25 14 114... Rh2 -19.09/25 15 115. Bf5 +56.47/24 11 115... Ra2 -250.00/32 8 116. Kc5 +128.40/32 11 116... Re2 -250.00/35 28 117. Rd8 +128.45/33 42 117... Nd4 -250.00/31 5 118. Rxe8 +M12/56 9 118... Nxf5 -250.00/30 24 Black resigns |
|
| Jun-28-17 | | WorstPlayerEver: Next game (136) Stockfish plays the Nimzo with Black and seems to neutralize White's position quite quickly: 1. d4 Nf6 2. c4 e6 3. Nf3 Bb4 4. Bd2 Qe7 5. g3 Nc6 6. Nc3 +0.29/30 92 6... Bxc3 -0.21/35 205 7. Bxc3 +0.23/30 46 7... Ne4 -0.31/27 15 8. Qc1 +0.22/32 209 8... O-O -0.17/30 61 9. Bg2 +0.35/28 58 9... d6 -0.20/31 54 10. d5 +0.22/30 126 10... Nd8 -0.13/34 227 11. Nd2 +0.23/28 169 11... Nxc3 -0.21/36 145 12. Qxc3 +0.30/28 50 12... e5 -0.13/32 23 13. O-O +0.26/28 95 13... f5 -0.06/33 120 |
|
| Jun-28-17 | | WorstPlayerEver: After 14. c5
click for larger view |
|
| Jun-28-17 | | WorstPlayerEver: Stockfish immediately went aggressive by playing 16... h5 14. c5 +0.27/29 241 14... Nf7 -0.11/31 182 15. Rac1 +0.23/31 171 15... b6 -0.10/30 69 16. b4 +0.15/28 104 16... h5 -0.07/31 328 17. Nf3 +0.30/26 123 17... bxc5 -0.04/27 207 18. bxc5 +0.30/24 54 18... f4 0.00/29 64 19. Qa3 +0.23/26 71 19... a5 0.00/29 78 20. Rc4 +0.32/27 63 
click for larger view |
|
| Jun-28-17 | | WorstPlayerEver: Gonna be a draw:
20. Rc4 +0.32/27 63 20... Bb7 -0.15/30 137 21. Rd1 +0.28/30 157 21... Ba6 -0.08/33 110 22. Re4 +0.22/29 63 22... fxg3 -0.29/29 16 23. hxg3 +0.31/28 34 23... Qf6 -0.08/31 91 24. Rf1 +0.22/26 45 24... Bc8 -0.20/28 38 25. Qc3 +0.21/28 192 25... Ng5 -0.08/31 56 26. Nxg5 +0.25/27 43 26... Qxg5 -0.03/32 141 27. Kh2 +0.20/30 257 27... Bf5 0.00/34 234 28. Rh4 +0.28/30 98 28... e4 0.00/40 107 
click for larger view |
|
| Jun-29-17 | | WorstPlayerEver: If someone's interested... another exciting game: 1. d4 Nf6 2. c4 g6 3. Nc3 Bg7 4. e4 d6 5. Nf3 O-O 6. Be2 e5 7. O-O Nc6 8. d5 Ne7 9. Ne1 Nd7 10. f3 +0.56/34 328 10... b6 -0.57/29 314 11. Rb1 +0.62/32 207 11... Nf6 -0.58/29 192 12. Nd3 +0.68/28 65 12... Nh5 -0.58/27 99 13. g4 +0.60/29 112 13... Nf6 -0.67/28 67 14. Rf2 +0.62/28 83 14... h6 -0.43/27 116 15. h4 +0.91/32 135 
click for larger viewhttp://www.fastgm.de/pgn4web/live1.... |
|
| Jun-29-17 | | WorstPlayerEver: Curiosity; after 15... Nd7 16. Rg2 Black has moved their Queen Knight 5 times out of 15 moves:
click for larger view |
|
| Jun-29-17 | | WorstPlayerEver: Game 148 (or 149,150 lol) after 39... Rc8-g8, don't you think there are still a lot of errrz to make in this postion?
click for larger view |
|
| Jun-29-17 | | WorstPlayerEver: PS still to recognize the KID here. Komodo-Stockfish |
|
 |
|
|
< Earlier Kibitzing · PAGE 7 OF 15 ·
Later Kibitzing> |
Kibitzing
