Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 on Skulltrail - A very interesting lesson
1 2 Previous Next  
- - By SR (****) Date 2008-08-19 13:04 Edited 2008-08-19 13:30
Yesterday I was visiting Paul "National" and I had the opportunity to sit in the drivers seat of his overclocked 8 CPU skulltrail. Incredible experience! According to the node count the his skull is about 10 times faster than my engine, but after my visit I saw evidence that its somehow is doing better than that.

We loaded Rybka 3 default book and of we went. The book "forced" R3 to play the king Indian and I was allowed to play the same move I originally played agaist my own engine (before its stopped playing the Kings Indian). The game was not played as a rated game so I had full access to the single main line in infinite mode. We also decided when R3 should move, but usually making sure it reached at least depth 17 or 18 (and believe me the depth was climbing fast on the skull) and that the evaluation was stable.

[Event """]
[Site "?"]
[Date "2008.08.19"]
[Round "?"]
[White "Skulltrail"]
[Black "Human+ access to machine output "]
[Result "0-1"]
[ECO "E97"]
[Annotator ""]
[PlyCount "70"]
[TimeControl "600+3"]

{136MB, Rybka3.ctg, OLD} 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. Nd2 Ne8 10. b4 f5 11. c5 h5

r1bqnrk1/ppp1n1b1/3p2p1/2PPpp1p/1P2P3/2N5/P2NBPPP/R1BQ1RK1 w - h6 0 12

11...h5 is my "anti-computer" move in this line. It leaves the theory and makes it very hard for any engine to spot whites problems before its to late.

12. Qb3 f4 13. h3 Kh7 14. a4 Kh6 15. a5 g5 16. a6 b6 17. cxb6 cxb6 18. Qd1 Nf6 19. Nb5 Ng6 20.
Ra2 g4 21. hxg4 hxg4 22. Rc2 Rf7 23. Rc6 Bf8

r1bq1b2/p4r2/PpRp1nnk/1N1Pp3/1P2Ppp1/8/3NBPP1/2BQ1RK1 w - - 0 24


It turns out that this position is extremely complex. During the game I explained to Paul that I would expect Rybka3 score of +0.80
to begin to drop. Usually depth 16 was reached very fast but in this position R3 suddently started to do some serious branching but after around 20 min we reached level 16. The white score suddently dramatically dropped and became almost -2.00 paws. OK nice that humans still understand something the engines fail to grasp.

24. Nc4 Rh7

r1bq1b2/p6r/PpRp1nnk/1N1Pp3/1PN1Ppp1/8/4BPP1/2BQ1RK1 w - - 0 25

Later I analysed the position in the diagram on my PC analyse overnight. In the morning the score had dropped to +0.00 at depth 15.

Back to the game. The game continued 25. g3 Bd7 26. Nbxd6 Bxd6 27. Rxd6 Kg7 28. gxf4 Qh8 29. Rxd7+ Nxd7 30. Bxg4 Nf6 31. Be3 Rh1+ 32. Kg2 Rxf1 33. Qxf1 Nxg4 34. Qh1 Qxh1+ 35. Kxh1 Nxf4 0-1

When I returned home I had another look at the game and realised that the feat of Pauls Skull was truely amasing.

I tried to analyse the game on my PC and I was astonished how slow is was compared to the skulltrail. In fact the position is incredible complex so Pauls Skull  must have seen a hell lot of lines already after 23....Bf8

The score of the Skulltrail drops at depth 16, but it require encredible deep and complex analysis to see this. My sp 3.2 GHz engine still did not have a clue what is happening 14 plies later!! And this was just in the main line. Many side lines (of other white attempts) are also extremely complex.

r6q/p2n2kr/Pp4n1/3Pp3/1PN1PPB1/8/5P2/2BQ1RK1 b - - 0 30


In this position R3 running on my modest PC (at depth 12) did still did still not have clue that white is lost.  How is this possible??? Well, I looked at the side lines and discovered that there is still are some tricky moves that are quite difficult to find.

I am now convinced that R3 is doing a lot more clever stuff than just simple search. My computer was calculating 12 plies ahead from the position in the diagram and still did not spot whites problems (after 5 min search).  Yet the skulltrail discovered whites problems essentially 14+12=26 plies before the problems arose. Scary stuff....

I am very curious to reach to the bottom of all this.  Is white really "lost" in all lines after 23....Bf8?  During the game I certainly though white had some problems, but it turns out that white has quite a number of tricky recources. Its just that black can be even more trickier. 

What are the main lines? It was first when I began to analyse some of the sidelines that I realised just how complex the position is.

I think the position is an excellent opportunity to see the benefits of Aquarium.
Without use of aquarium the required analysis is only for the heavy fish  
Parent - - By Kappatoo (*****) Date 2008-08-19 13:19
Surprising that the King's Indian still works that well against the engines. Just look at white's 12th and 13th move - awful!
Parent - - By SR (****) Date 2008-08-19 13:41
Well, without these imperfections the engines would really rule ;-)
Parent - - By Roland Rösler (****) Date 2008-08-20 03:05
Hundreds of times I have seen human (Top GM) creating very complex positions and then come the crash! Too complex!
If I´m right, that you play the opening line until 11. ... h5 and then you always followed the moves from Skulltrail up to move 24 (25), I´m not so much impressed. Give the black pieces after 24. Nc4 to Anand, Carlsen or Moro and you will see they have no chance against Rybka 3 on Skulltrail with white.

PS: It would be interesting to hear, what 2750+ players think about position after move 24.
Parent - - By SR (****) Date 2008-08-20 07:21 Edited 2008-08-20 07:23
I did not say you had to be impressed.  The position on Pauls skull was new to me already after 14.a4 (that is the correct move, but part of the wrong plan). Why would Anand,Carlsen or More get into the position after 24.Nc4 in the first place?  It has nothing to do about having seen all the moves up to move 24 already before. Its simply that if Anand,CArlsen or Moro had white they would for sure not have played 13. h3 and they would certainly have prefered to play the standard white break through from the main lines in the Kings Indians.

The whole way the pawn structure was fixed on the queen side in a way I had not seen before was new to me, but was not very relevant. White only two plans were to desparately attack the d6 pawn or try to infiltrate along the c-file, possible with a knight getting to c7. 

Personally, I am much more happy with most of my draws, especially when they are in lines  that are new to me and when the game is played under proper rated conditions. My latest game from was such a draw (all setting was default except the hash of 512 that I forgot to reset).

[Event "Blitz:10'+3""]
[Date "2008.08.20"]
[White "Riis, Soren"]
[Black "Rybka 3 32-bit"]
[Result "1/2-1/2"]
[ECO "A58"]

{512MB, Rybka3.ctg, OLD} 1. d4 {2} Nf6 {1} 2. c4 {1} c5 {0} 3. d5 {1} b5 {0} 4.
cxb5 {1} a6 {0} 5. bxa6 {1} g6 {0} 6. Nf3 {2} Bxa6 {0} 7. Nc3 {1} Bg7 {0} 8. g3
{1} d6 {0} 9. Bg2 {1} Nbd7 {0} 10. Rb1 {2} Nb6 {0} 11. b3 {1} Bb7 {0} 12. Nh4 {
1} O-O {0} 13. O-O {1} Ne8 {0} 14. Bb2 {2} Bxc3 {0} 15. Bxc3 {0} Rxa2 {0} 16.
b4 {0} c4 {22} 17. Qd4 {1} Nf6 {15} 18. e4 {6} Qc7 {62} 19. Ra1 {117} Rfa8 {13}
20. g4 {15} h6 {53} 21. f4 {23} Rxa1 {21} 22. Rxa1 {0} Rxa1+ {21} 23. Bxa1 {0} c3
{29} 24. Qxc3 {75} Qxc3 {14} 25. Bxc3 {0} Nxg4 {24} 26. Bf3 {20} Nf6 {25} 27. Kf2
{81} Ba6 {0} 28. Bd4 {20} Nbd7 {0} 29. Ke3 {31} Bc4 {0} 30. Kd2 {52} Kf8 {12}
31. Kc3 {5} Ba6 {59} 32. Ng2 {10} g5 {10} 33. fxg5 {12} hxg5 {9} 34. Ne3 {1}
Ke8 {8} 35. Nc4 {19} g4 {11} 36. Bg2 {7} Nh5 {31} 37. Bf1 {26} Kd8 {0} 38. Be2
{15} Nhf6 {7} 39. Bd3 {12} Bxc4 {8} 40. Kxc4 {4} Kc7 {8} 41. Kb5 {31} Kb7 {17}
42. Ka5 {6} Ne5 {7} 43. Bxe5 {7} dxe5 {1} 44. Kb5 {16} Nh7 {10} 45. Be2 {40}
Ng5 {0} 46. Bxg4 {5} Nxe4 {38} 47. h4 {14} f5 {16} 48. Bh3 {77} Ng3 {0} 49. Kc4
{10} Kb6 {7} 50. h5 {14} Nxh5 {1} 51. Bxf5 {1} Nf6 {9} 52. Be6 {12} Kc7 {1} 53.
Kb5 {27} Kd6 {6} 54. Kc4 {51} e4 {0} 55. Kd4 {6} e3 {1} 56. Kxe3 {1} Nxd5+ {8}
57. Bxd5 {6} Kxd5 {0} 58. Kd3 {1} e6 {0} 59. b5 {6} Kc5 {0} 60. Ke4 {1} Kxb5 {0
} 61. Ke5 {0} Ka4 {0} 62. Kxe6 {0} 1/2-1/2

Please do not ask me why Rybka plays the Benko Gambit. It started to do it a few days ago, but after I studied the main line and recomended moves for 20 miniuttes or so, I found a way to handle the opening that suits human play (when you play computers the key is obviously to play lines that are easy to play, and where the plans are relatively clearcut).

r2qnrk1/1b2ppbp/1n1p2p1/2pP4/7N/1PN3P1/P3PPBP/1RBQ1RK1 w - - 0 14

Problem Find a good move to play against a chess engine

Answer: The recomended theory move in Jeroens book is 14.Qd2 and maybe I should look into that if I am more ambitious. Anyway, the move 14.Bb2 was my "home preparation" and the real point is 16.b4! that solves all problems white might have agaist a top engine.

6k1/1bq1pp2/1n1p1npp/3P4/1PpQPPPN/8/6BP/B5K1 b - - 0 23

White have had very logical and easy moves to play. The point of the white play is to force black to give up his black squared bishop and get dangerous play along the diagonal a1-h8. The rook from f8 needs to be forced away (this expalins 19.Ra1) as black otherwise can play e7-e5 and after d5xe6ep, fxe6 the rook protects the knight of f6 and black is ready to play e5 with the type of  position where any computer would beat a human.

6k1/1b2pp2/1n1p1npp/3P4/1P2PP1N/2B2B2/7P/6K1 w - - 0 27


The smoke has settled. White has a fine position (bishop pair and a dangerous b-pawn), while whites only concern is to avoid black to destroy his pawn center (e.g. by playing f5). White has nothing to fear, except when you play R3 every minute error will be punished.

8/8/4K3/8/k7/8/8/8 b - - 0 62


The final position
Parent - - By Roland Rösler (****) Date 2008-08-20 08:05
Its simply that if Anand,CArlsen or Moro had white they would for sure not have played 13. h3 and ...

Here you are quite right! They never would play 13. h3??. But I think, this is part of the problem. They have black and would feel very good (maybe they are amuzed?) about the patzer with white. But be sure, they will lose against this patzer! They think, they are very comfortable after 24. Nc4, but no way for them for win or draw! Here you can see, that chess is too complex for humans! Larry makes all his points against his poor opponents in this way. Best positions for the humans, but after some time too complex for the humans! That´s all! It´s no science. When he gave knight odds to the opponent (FM), it was easy for the human. No experiments! Keep it easy and stupid! And the FM won!!

PS: Best human chess players don´t make it easy and stupid. Then they wouldn´t be best human chess players. They all have their plans and traps (sometimes) to win against other humans. But computer chess has much more dimensions. Humans have only four dimensions, but cc has at least 10 (or 11)! :-) Sorry for this support of super string theory.
Parent - - By SR (****) Date 2008-08-20 08:13 Edited 2008-08-20 08:28
They think, they are very comfortable after 24. Nc4, but no way for them for win or draw!

If I can win this type of position, they probably could win the positions blindfold ;-)  I am virtually certain that Carlsen would have crushed R3 after 24.----Nc4 and possibly even if he had black after move 12. How can you suggest that they have no way to win or draw? Which of the black moves that occured in the game after 24....Nc4 do you think they would miss?  The move are not that difficult to find for a human player. Or are you suggesting they have white and would think they are confortable after 24.Nc4? With the white pieces they would certainly not feel confortable after 24.Nc4 as black has all kinds of dangerous possibilities.  They would probably judge the position as unclear...
Parent - By Roland Rösler (****) Date 2008-08-20 08:42
Or are you suggesting they have white and would think they are confortable ...

No! No! No! I said it three times, that human with black can´t win or draw after move 24. Nc4. Okay, you followed the line on Skulltrail and all was easy! I followed human top tournaments many times live and I´m sure, they would blunder!!
Parent - By stvs (***) Date 2008-08-23 02:30
very good.......ur total results vs r3 soren? seems u mastered the r3!
Parent - - By Arkansaw (***) Date 2008-08-19 14:12
Well, it's the same engine, different hardware only affects the time needed for evaluation
Parent - - By SR (****) Date 2008-08-19 14:56 Edited 2008-08-19 15:22
I am not sure its that simple. I think R3 is pretty "non-deterministic" (on a 8 core engine) and it might only hit some of the key moves on some of the runs (though eventually if you are willing to wait for days it will find every thing). Can Pauls dramatic fall in evaluation be reproduced on other engines?? My PC was not able to reproduce it despite that a simple node count suggest my engine should have dropped its evaluation after approx. 3 hours. Even with the benefit of two extra plies that score did not drop after more than 6 hours calculations! Was Pauls Skull just lucky ;-) or is the mp version fundamentally different for the sp version?

r6q/p2n2kr/Pp4n1/3Pp3/1PN1PPB1/8/5P2/2BQ1RK1 b - - 0 30


One of the difficulties during many of my runs (after having cleared the hash) R3 thinks that 30....Rh1+ is an only move (well not in the position but for example when I analysed the position after move 24 overnight). 

The engine have to find 30...Nf6 and then realise that 31.Bf3 does not work (31....Rg8!) and that after a move like 31.Be3 black has 31....Rh1+ 32.Kg2,Rxf1!  These are not easy moves (for a computer) to find in advance.
Parent - - By Hamlet (**) Date 2008-08-19 15:50
24. xxx fail low after 15-ply :-), heavy one.
More later
Parent - - By Hamlet (**) Date 2008-08-19 15:55
Another one :-) in row. First:
24.Nd2-c4 Nf6-e8 25.Be2xg4 Bc8xa6 26.Nb5-c3 Ba6xc4 27.Rc6xc4 Ne8-f6 28.Bg4-e6 Rf7-c7 29.Rc4xc7 Qd8xc7 30.Qd1-d3 a7-a6 31.Bc1-a3
  ±  (1.07)   Depth: 14   00:00:09  1599kN
24.Qd1-c2 Bc8-d7 25.Rc6-c3 Rf7-h7 26.Nb5-c7 Ra8-c8 27.Nc7-e6 Rc8xc3 28.Qc2xc3 Qd8-c8 29.Qc3xc8 Bd7xc8 30.Be2-b5 Bc8xe6 31.d5xe6 Rh7-g7
  ²  (0.32)   Depth: 15   00:03:39  38814kN

24.Nd2-c4 failed low (15-ply missing) and then right now 24.Qd1-c2 failed low in 16-ply.
Parent - - By Hamlet (**) Date 2008-08-19 16:36 Edited 2008-08-19 16:39
Yes, several fail lows. Here is some analysis:

r1bq1b2/p4r2/PpRp1nnk/1N1Pp3/1P2Ppp1/8/3NBPP1/2BQ1RK1 w - - 0 1


Analysis by Rybka 3 good:

24.Qd1-a4
  ²  (0.63)   Depth: 2   00:00:00
24.Qd1-c2
  ±  (0.81)   Depth: 2   00:00:00
24.Bc1-b2
  ±  (0.77)   Depth: 2   00:00:00
24.Bc1-b2
  ±  (0.77)   Depth: 4   00:00:00
24.Bc1-b2
  ±  (0.77)   Depth: 5   00:00:00
24.Qd1-c2
  ±  (0.97)   Depth: 5   00:00:00
24.Qd1-c2 Bc8xa6
  ±  (1.03)   Depth: 5   00:00:00
24.Qd1-c2 Bc8xa6 25.Nb5xd6 Bf8xd6 26.Be2xa6
  ±  (0.86)   Depth: 6   00:00:00  7kN
24.Nd2-c4
  ±  (1.06)   Depth: 7   00:00:00  25kN
24.Nd2-c4
  ±  (1.06)   Depth: 7   00:00:00  25kN
24.Nd2-c4 Rf7-d7 25.Qd1-a4 Ng6-h4
  ±  (0.97)   Depth: 7   00:00:00  52kN
24.Nd2-c4 Rf7-d7 25.Rf1-e1 Bc8xa6 26.Nb5xd6 g4-g3 27.f2xg3 Kh6-h7 28.b4-b5
  ±  (0.95)   Depth: 8   00:00:00  81kN
24.Nd2-c4 Rf7-d7 25.Rf1-e1 Bc8xa6 26.Nb5xd6 g4-g3 27.f2xg3 Kh6-h7 28.b4-b5
  ±  (0.95)   Depth: 10   00:00:00  83kN
24.Nd2-c4 Rf7-d7 25.Rf1-e1 Bc8xa6 26.Nb5xd6 g4-g3 27.f2xg3 Kh6-h7 28.b4-b5
  ±  (0.95)   Depth: 10   00:00:00  103kN
24.Nd2-c4
  ±  (1.15)   Depth: 11   00:00:02  305kN
24.Nd2-c4 Nf6-e8 25.Be2xg4 Bc8xa6
  ±  (1.02)   Depth: 11   00:00:03  485kN
24.Nd2-c4 Nf6-e8 25.Be2xg4 Bc8xa6 26.Nb5-c3 Ba6xc4 27.Rc6xc4 Ne8-f6 28.Bg4-e6 Rf7-c7 29.Rc4-c6 a7-a5 30.Nc3-b5 Rc7-h7
  ±  (0.99)   Depth: 12   00:00:04  654kN
24.Nd2-c4 Nf6-e8 25.Be2xg4 Bc8xa6 26.Nb5-c3 Ba6xc4 27.Rc6xc4 Ne8-f6 28.Bg4-e6 Rf7-c7 29.Rc4xc7 Qd8xc7 30.Qd1-d3 a7-a6 31.Bc1-a3
  ±  (1.07)   Depth: 13   00:00:06  1002kN
24.Nd2-c4 Nf6-e8 25.Be2xg4 Bc8xa6 26.Nb5-c3 Ba6xc4 27.Rc6xc4 Ne8-f6 28.Bg4-e6 Rf7-c7 29.Rc4xc7 Qd8xc7 30.Qd1-d3 a7-a6 31.Bc1-a3
  ±  (1.07)   Depth: 14   00:00:09  1599kN
24.Qd1-c2 Bc8-d7 25.Rc6-c3 Rf7-h7 26.Nb5-c7 Ra8-c8 27.Nc7-e6 Rc8xc3 28.Qc2xc3 Qd8-c8 29.Qc3xc8 Bd7xc8 30.Be2-b5 Bc8xe6 31.d5xe6 Rh7-g7
  ²  (0.32)   Depth: 15   00:03:39  38814kN
24.g2-g3 Rf7-h7 25.Kg1-g2 Kh6-g7 26.Rf1-h1 Bc8-d7 27.Rh1xh7+
  =  (0.06)   Depth: 16   00:35:17  391mN
24.g2-g3 Rf7-h7 25.Kg1-g2 Bc8-d7 26.Rf1-h1+ Kh6-g7 27.Rh1xh7+ Kg7xh7 28.Rc6-c3 f4-f3+ 29.Nd2xf3 g4xf3+ 30.Rc3xf3 Bd7-e8 31.Qd1-h1+ Kh7-g8 32.Qh1-h3 Be8-d7 33.Qh3-h1 Qd8-e7
  =  (0.00)   Depth: 17   00:37:37  416mN
24.g2-g3 Rf7-h7 25.Kg1-g2 Bc8-d7 26.Rf1-h1+ Kh6-g7 27.Rh1xh7+ Kg7xh7 28.Rc6-c3 f4-f3+ 29.Nd2xf3 g4xf3+ 30.Rc3xf3 Bd7-e8 31.Rf3-c3 Be8xb5 32.Be2xb5 Ra8-c8 33.Bb5-c6 Bf8-g7 34.Qd1-h1+ Kh7-g8 35.f2-f3 Rc8-c7 36.Bc1-g5 Qd8-c8
  =  (0.06)   Depth: 18   00:53:26  599mN
Parent - - By SR (****) Date 2008-08-19 19:32 Edited 2008-08-19 20:09
Interesting! From this analysis it looks like 24.Nc4 could have been a decisive mistake and that 24.g3 is a much better choice that keep the position unclear/balanced.

Added Hang on a minutte! After 24.g3 black plays 24....Kg7 preparing to play 25....f4-f3 and white might strangulated.
I am not sure the engines fully appriciate blacks posibilities.....

Added Here is a little sample line where my R3 suddently dropped its evaluation after the thematic move 28....Nf4!

r1bqnbk1/p4r2/PpRp4/1N1Pp3/1P2Pnp1/3B1pP1/QB1N1P2/5RK1 w - - 0 29


The line went: 24.g3,Kg7! 25.Qc2, f3! 26.Bd3,Ne8 27.Qa2,Kg8 28. Bb2, Nf4! winning. I am sure a faster engine will shuffle the white pieces around in a different way, but I still think black might have a decisive advantage.
Parent - By NATIONAL12 (Gold) Date 2008-08-19 20:19
Soren i am sorry i cannot help you at the moment,i am doing a lot of AN in match against hamsters.

regards Paul

ps you have created an excellent thread which invites discussion,you must be one of 6 strongest players on the forum.the others know who they are.
Parent - - By turbojuice1122 (Gold) Date 2008-08-19 15:58
Vas has stated before that an n-ply search with the MP version will definitely be stronger than an n-ply search in the SP version, so this goes along with your line of thinking.
Parent - By M ANSARI (*****) Date 2008-08-19 16:20
When I first got a dual core (on a laptop), I couldn't believe how fast it was.  Then I got a Quadcore and was equally impressed and couldn't use the dual core anymore.  This again happened when I managed a very high overclock of the Quad and yet again when I got an Octa.  As you go up in hardware, you wonder how you ever managed with less hardware.  At the moment I have a hard time even using the 3.2Ghz Octa when the overclocked one does so much better.  I think this trend will go on and on.
Parent - By Arkansaw (***) Date 2008-08-19 22:41 Edited 2008-08-19 22:45
Yes, this position is complicated enough to make the evals behave that way. I just ran it for a bit, and at depth 13 it is already taking way too much time than usual

I still don't find the Skulltrail amazing....for the amount of investment, that order of performance is only to be expected.
Parent - By NATIONAL12 (Gold) Date 2008-08-19 16:46
i remember the eval changed from 0.90 to -3.60 in the space of one move.this was when we let Skulltrail run for 20+ mins to see if could improve its position.i think it dropped to -2 approx.it was very instructive to watch.
ps Soren left it entirely to me on when to let Skulltrail move (R3)
Parent - - By Ernst (***) Date 2008-08-19 16:58
With default book you probably mean the book provided with the program and not Rybka3.ctg?
Parent - - By NATIONAL12 (Gold) Date 2008-08-19 17:13
yes,but Soren asked which book i wished to use,i have quite a number.i chose the default book that came with the engine as it is not one i normally use.
Parent - - By Hamlet (**) Date 2008-08-19 17:28
Rybka 2 book don't play this variation.
Parent - By NATIONAL12 (Gold) Date 2008-08-19 17:38
i havent checked and cant at the moment,no doubt Soren will clarify.
Parent - By NATIONAL12 (Gold) Date 2008-08-19 17:44
book used was Rybka 3Gen
Parent - - By SR (****) Date 2008-08-19 18:32
Quite a lot of books enters this line on their own accord (even Jeroen's Rybka3 book), but the Kings Indian might not have been in the book of Rybka2. Its in the default book of Rybka3 that was used on Paul's engine. Of course with engine learning set on optimal this is a minute problem as the book will stop playing the the line after basically just one bad experience.
Parent - By Ernst (***) Date 2008-08-19 19:20
9.Nd2 would not be my choice in this opening.
Parent - By Banned for Life (Gold) Date 2008-08-19 20:37
Soren,

I seem to remember suggesting a while back that engines have a hard time with KID and its associated pawn storm, and Vas telling me that this would no longer be true in R3. Thanks for showing that this is still an open issue!!!

Regards,
Alan
Parent - - By M ANSARI (*****) Date 2008-08-19 19:57
No need for Skulltrail ... even a "lowly" Q6600 can see that white is in deep trouble :).  Actually looking quickly at the position white has to protect mate on h1 ... for that he needs a white bishop or queen.  Since queen and bishop cannot be coordinated then the white bishop is key.  With Nf6! black attacks this crucial piece and looks like Bf3 is forced then Rg8! seems to win since white cannot seem to use counterplay of the relatively exposed black King without getting mated first.  This ofcourse needs precise calculation which computers are just fantastic at.  At least in this position I think things are quite easy for an engine to calculate but I haven't checked the earlier positions.

New game, Blitz:4'+2"
r6q/p2n2kr/Pp4n1/3Pp3/1PN1PPB1/8/5P2/2BQ1RK1 b - - 0 1


Analysis by Rybka 3 :

1...Rh1+
  ±  (1.18)   Depth: 2   00:00:00
1...Rh1+
  ±  (1.23)   Depth: 4   00:00:00
1...Rh1+
  ±  (1.03)   Depth: 4   00:00:00
1...Rh1+
  ±  (0.83)   Depth: 4   00:00:00
1...Rh1+
  ²  (0.68)   Depth: 4   00:00:00
1...Rh1+
  ²  (0.48)   Depth: 5   00:00:00
1...Rh1+
  ²  (0.28)   Depth: 5   00:00:00
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3
  ²  (0.30)   Depth: 5   00:00:00
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Bf5 Nh4+ 5.Ke2
  ²  (0.36)   Depth: 6   00:00:00  20kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Bf5 Nh4+ 5.Ke2 Nxf5
  ²  (0.37)   Depth: 7   00:00:00  31kN
1...Rh1+
  =  (0.17)   Depth: 8   00:00:00  83kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Bf5 Nh4+ 5.Ke2 Nxf5 6.exf5 Qh5+ 7.Ke1 exf4 8.Rg1+ Kh7 9.Bxf4 Qxd1+ 10.Kxd1 Rxf2 11.Rh1+ Kg7 12.Rg1+
  =  (0.00)   Depth: 8   00:00:00  159kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Bf5 Nh4+ 5.Ke2 Nxf5 6.exf5 Qh5+ 7.Ke1 exf4 8.Rg1+ Kh7 9.Bxf4 Qxd1+ 10.Kxd1 Rxf2 11.Rh1+ Kg7 12.Rg1+
  =  (0.00)   Depth: 9   00:00:00  177kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Bf5 Nh4+ 5.Ke2 Nxf5 6.exf5 Qh5+ 7.Ke1 exf4 8.Rg1+ Kh7 9.Bxf4 Qxd1+ 10.Kxd1 Rxf2 11.Rh1+ Kg7 12.Rg1+ Kh7 13.Rh1+ Kg7 14.Rg1+ Kh7 15.Rh1+ Kg7 16.Rg1+ Kh7
  =  (0.00)   Depth: 10   00:00:01  338kN
1...Rh1+
  =  (-0.20)   Depth: 11   00:00:03  978kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Bf5 Nh4+ 5.Ke2 Nxf5 6.exf5 Qh5+ 7.Ke1 exf4 8.Bxf4 Re8+ 9.Kd2 Rxf2+ 10.Rxf2 Ne4+ 11.Kc2 Qxd1+ 12.Kxd1 Nxf2+ 13.Kc2 Re4 14.Be5+ Kg8 15.Kc3 Nd1+ 16.Kd3
  =  (0.00)   Depth: 11   00:00:04  1063kN
1...Rh1+
  =  (-0.20)   Depth: 12   00:00:09  2724kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Ke3 exf4+ 5.Kd3 b5 6.Bb2 bxc4+ 7.Kxc4 Rb8 8.Kb3 Qf8
  =  (-0.25)   Depth: 12   00:00:25  7052kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Ke3 exf4+ 5.Kd3 b5 6.Bb2 bxc4+ 7.Kxc4 Rb8 8.Kb3 Qf8
  =  (-0.09)   Depth: 13   00:00:29  8033kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Ke3 exf4+ 5.Kd3 b5 6.Bb2 bxc4+ 7.Kxc4 Rb8 8.Rg1 Qf8 9.Bc3 Ne5+ 10.Kb3 Kh6 11.Bf5 Rxf2 12.Rh1+ Kg5 13.Rg1+ Kh6 14.Rh1+ Kg5 15.Rg1+ Kh6 16.Rh1+ Kg5
  =  (0.00)   Depth: 14   00:00:48  12762kN
1...Rh1+ 2.Kg2 Rh2+ 3.Kf3 Nf6 4.Ke3 exf4+ 5.Kd3 b5 6.Bb2 bxc4+ 7.Kxc4 Rb8 8.Rg1 Qf8 9.Bc3 Ne5+ 10.Kb3 Kh6 11.Bf5 Rxf2 12.Rh1+ Kg5 13.Rg1+ Kh6 14.Rh1+ Kg5 15.Rg1+ Kh6 16.Rh1+ Kg5
  =  (0.00)   Depth: 15   00:01:43  26993kN
1...Nf6
  =  (-0.20)   Depth: 15   00:02:01  33000kN
1...Nf6
  ³  (-0.40)   Depth: 15   00:02:15  37277kN
1...Nf6
  µ  (-0.80)   Depth: 15   00:02:38  44670kN
1...Nf6
  -+  (-1.60)   Depth: 15   00:03:21  58258kN
1...Nf6 2.Bf3 Rg8 3.Nd6 Kf8 4.Nf5 exf4 5.Ng3 fxg3 6.fxg3 Ne5 7.Bf4 Nfg4 8.Bxe5 Nxe5 9.Bg2+ Rf7 10.Rxf7+ Nxf7 11.Qf3 Qd4+ 12.Qf2 Qxf2+ 13.Kxf2 Ke7 14.Bh3 Ne5 15.Kg2 Kd6 16.Be6 Rg7
  -+  (-2.91)   Depth: 15   00:05:20  95828kN
1...Nf6 2.Bf3 Rg8 3.Nd6 Kf8 4.Nf5 exf4 5.Ng3 fxg3 6.fxg3 Ne5 7.Bf4 Nfg4 8.Bxe5 Nxe5 9.Bg2+ Rf7 10.Rxf7+ Nxf7 11.Qf3 Qd4+ 12.Qf2 Qxf2+ 13.Kxf2 Ke7 14.Bh3 Ne5 15.Kg2 Kd6 16.Be6 Rg7
  -+  (-2.91)   Depth: 16   00:06:28  119mN
1...Nf6
  -+  (-3.11)   Depth: 17   00:10:08  192mN

(,  19.08.2008)
Parent - - By SR (****) Date 2008-08-19 20:21
I think you misunderstood me. Yes your Q6600 engine found the win at depth 15, but was talking about the difficulties handling the position not as root position. What I ment is that my R3 had great difficulties to appreciate this winning line in the diagram 5-6 plies earlier. In the diagram it took my engine around 6 min, so yours Q6600 is probably running twice the speed of my 3.2GHz P4 outdated CPU.

I agree 100% with what you said about the position and that black obviously is winning. I never claimed I had problems understanding blacks possibilities in this type of position. The whole point of blacks attack (right from the opening) is that it is relatively simple to play for a human who have the sufficient chess knowledge and can make the correct decisions in a few key positions.
Parent - By M ANSARI (*****) Date 2008-08-19 22:07
Ah ok ... anyway I had forgotten the other position on so I have good analysis from the Q6600 ... it seems my Quad here thinks it is a draw.

New game
r1bq1b2/p4r2/PpRp1nnk/1N1Pp3/1P2Ppp1/8/3NBPP1/2BQ1RK1 w - - 0 1


Analysis by Rybka 3 :

1.Qa4
  ²  (0.63)   Depth: 2   00:00:00
1.Qc2
  ±  (0.81)   Depth: 2   00:00:00
1.Bb2
  ±  (0.88)   Depth: 3   00:00:00
1.Bb2
  ±  (0.88)   Depth: 3   00:00:00
1.Bb2
  ±  (0.74)   Depth: 4   00:00:00
1.Qc2 Bxa6
  ±  (0.71)   Depth: 5   00:00:00
1.Qb3 Rh7
  ±  (0.85)   Depth: 5   00:00:00
1.Qb3 Rh7 2.Re1
  ±  (0.82)   Depth: 6   00:00:00  14kN
1.Qb3 Bd7 2.Nxd6 Bxd6 3.Rxd6 Qe7
  ±  (0.71)   Depth: 7   00:00:00  19kN
1.Nc4
  ±  (1.02)   Depth: 7   00:00:00  26kN
1.Nc4 Rd7 2.Qa4 Kh7
  ±  (0.96)   Depth: 7   00:00:00  46kN
1.Nc4
  ±  (1.16)   Depth: 8   00:00:00  54kN
1.Nc4
  ±  (1.36)   Depth: 8   00:00:00  56kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6
  ±  (1.15)   Depth: 8   00:00:00  99kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6 5.Be6 Rg7
  ±  (1.00)   Depth: 9   00:00:00  139kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6 5.Be6 Rc7 6.Rc6
  ±  (1.07)   Depth: 10   00:00:01  198kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6 5.Be6 Rc7 6.Rc6 a5 7.bxa5
  ±  (1.07)   Depth: 11   00:00:01  292kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6 5.Be6 Rc7 6.Rc6 a5 7.bxa5
  ±  (1.07)   Depth: 12   00:00:02  476kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6 5.Be6 Rc7 6.Rc6 a5 7.bxa5 Rxa5 8.Qd3
  ±  (1.01)   Depth: 13   00:00:04  1051kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bxc4 4.Rxc4 Nf6 5.Be6 Rc7 6.Rc6 a5 7.bxa5 Rxa5 8.Qd3 Ne7 9.Rxc7 Qxc7 10.g3
  ±  (0.94)   Depth: 14   00:00:13  3955kN
1.Nc4 Ne8 2.Bxg4 Bxa6 3.Nc3 Bb7 4.Qd3 Bxc6 5.dxc6 Rg7 6.Rd1 Qg5 7.Qh3+ Nh4 8.f3 Nf6 9.g3
  ±  (0.90)   Depth: 15   00:00:39  11416kN
1.Nc4 Rh7 2.Nbxd6 Bxd6 3.Nxd6 Bd7 4.Nf5+ Kg5 5.Rxf6 Qxf6 6.Bxg4 Rah8 7.f3 b5 8.d6 Bxf5 9.exf5 Ne7 10.Kf2 Nxf5 11.Re1 Nxd6 12.Qd5 Re8 13.Bxf4+ Qxf4 14.Qxd6 e4 15.Qd5+ Qe5 16.Qxe5+
  ³  (-0.52)   Depth: 16   00:31:03  542mN
1.Qc2 Ne8 2.g3 Rh7 3.Qd1 Kg7 4.Bxg4 Bxa6 5.Qa4 Bxb5 6.Qxb5 Qg5 7.Bf5 Qh5 8.Kg2 Qh2+ 9.Kf3 Qh5+ 10.Bg4 Qg5 11.Be6 fxg3 12.fxg3 Rh2 13.Rf2 Qh5+ 14.Ke3 Qg5+ 15.Ke2 Qxg3 16.Rxh2
  =  (0.05)   Depth: 16   00:48:37  830mN
1.Qc2 Ne8 2.g3 Rh7 3.Qd1 Kg7 4.Bxg4 Bxa6 5.Qa4 Bxb5 6.Qxb5 Qg5 7.Bf5 Qh5 8.Kg2 Qh2+ 9.Kf3 Qh5+ 10.Bg4 Qg5 11.Be6 fxg3 12.fxg3 Rh2 13.Rf2 Qh5+ 14.Ke3 Qg5+ 15.Ke2 Qxg3 16.Rxh2
  =  (0.05)   Depth: 17   00:57:05  978mN
1.Qc2 Ne8 2.g3 Rh7 3.Qd1 Kg7 4.Bxg4 Bxa6 5.Qa4 Bxb5 6.Qxb5 Qg5 7.Bf5 Qh5 8.Kg2 Qh2+ 9.Kf3 Qh5+ 10.Bg4 Qg5 11.Be6 fxg3 12.fxg3 Rh2 13.Rf2 Qh5+ 14.Ke3 Qg5+ 15.Ke2 Qxg3 16.Rxh2
  =  (0.05)   Depth: 18   01:09:02  1190mN
1.Qc2 Ne8 2.g3 Rh7 3.Qd1 Kg7 4.Bxg4 Bxa6 5.Qa4 Bxb5 6.Qxb5 Qg5 7.Bf5 Qh5 8.Kg2 Qh2+ 9.Kf3 Qh5+ 10.Kg2 Qh2+ 11.Kf3 Qh5+ 12.Kg2 Qh2+ 13.Kf3 Qh5+ 14.Kg2 Qh2+ 15.Kf3 Qh5+ 16.Kg2
  =  (0.00)   Depth: 19   01:39:15  1734mN

(,  20.08.2008)
Parent - By NATIONAL12 (Gold) Date 2008-08-19 20:22
you must remember we had all this fun in a little over an hour
Parent - - By NATIONAL12 (Gold) Date 2008-08-19 20:32
as an aside to this, i find the new chessbase GUI with R3,has dropped my core temps on second processor by 5-6 degrees centigrade.now both are running approx 77c.srange if this is a coincidence.this is air cooled.
Parent - - By Banned for Life (Gold) Date 2008-08-19 20:40
R3 crashes a lot less than Zappa on my 3.6 GHz air cooled, marginally stable Q6600 when the room gets warm, so I'm guessing that R3 is a lot less stressful for the processor.

Regards,
Alan
Parent - By NATIONAL12 (Gold) Date 2008-08-19 21:00
my temps have dropped with Zappa as well in this new GUI,running Zappa at the moment 76-79c
Parent - By brunjes (**) Date 2008-08-19 23:23
Better yet, run Rybka 2.3.2a against it.  :-)

There are always positions where the older version of an engine will spot a particular move faster than its newer, shinier sibling and this appears to be one of them.

Roy
Parent - - By Christian Packi (****) Date 2008-08-19 20:27
May i ask you what your fide rating is? I really like your posts and you seem to know how to give rybka trouble.
Parent - - By SR (****) Date 2008-08-19 20:48
Thanks for your kind words. About the fide rating see ;-)

http://www.chessbase.com/newsdetail.asp?newsid=3465  
Parent - By Christian Packi (****) Date 2008-08-19 21:38
Nice articel. I remember the Crafty test and thought it was quite dubious too.
Parent - - By Hamlet (**) Date 2008-08-19 22:35
Soren. I should try correspondence chess. :-) You would be surprised how weak some 2500+ iccf players are. they just play rybka moves.
I know from experience.
regards,
hamlet
Parent - By Hamlet (**) Date 2008-08-19 22:42
Or even worse. I had one opponent 2400+ who played only deep fritz 10 moves. :-(. he lost.
Parent - By Kapaun (****) Date 2008-08-19 23:34
Playing Rybka moves is fine and will get you very far. As long as you manage to identify the two or three occasions per game where it won't - and find the appropriate move, of course... :-)
Parent - By Uri Blass (*****) Date 2008-08-22 21:16
No surprise
Rybka is strong.

I have iccf rating above 2600 and I guess that the team of me and rybka3 cannot score more than 60% against Rybka3 in correspondence games.

Uri
Parent - - By Arkansaw (***) Date 2008-08-19 23:41
Okay, I can safely conclude that this is a wide tree with several deep lines.

After one hour, we see:

Various alternatives with none sharing the same eval
only 75 positions evaluated at depth 13, which is far from the usual 10s/position rate, and clearly insufficient to instil confidence in the current main line
Attachment: test.JPG (150k)
Parent - - By SR (****) Date 2008-08-20 00:05 Edited 2008-08-20 00:17
Now try to "help" IDeA by suggesting moves for black. This interactive feauture is the real advantage of IDeA. It allows to combine the human knowledge with the engine calculations.  Scroll through the tree and follow the line IDeA currently favor for white. Often the suggested black move is clearly enferious because Rybka expect you to nicely cover d6, while you want to slowly (or sometime pretty fast) build up a king attack.  The point is that by playing moves YOU (the user) consider interesting for black, the moves automatically becomes a part of the IDeA.

If your suggestion is bad you have just waisted Rybkas time, but if it is good Rybka will take your suggestion into account and change its overall evaluation. I do not think one needs to be a super strong player to enjoy this type of interactive analysis. Its also a very good way of learning about the strength and weaknesses of your plans.

PS. Another way your can help IDeA build up a tree is to try play out some of the lines to a greater depth. Try for example to enter the remaining moves of the game I played. After 25.g3 suggest the move 25...Kg7 and then if possible probably suggest 26....f3 in the next move.  If you get that position ignore every thing (rooks, pawns, knight etc) and try to get a queen and rook to the h-file OR try to bring a black knight to f4 (where is can often not be taken because it will open up lines to the white king). Experiment with this type of ideas and the evaluations in the IDeA will soon change dramatically!
Parent - - By Arkansaw (***) Date 2008-08-20 01:35
I'm considering what is the best way to deduce which moves are worth trying, assuming that I know nothing about the original position and the lines that you described.

Right now, apart from the only 'good' move g3 (which has just dipped below 0.00), all the other options look worse and they seem to get cut off pretty early. The total position count is still very low and with few transpositions yet, the position counts do reflect how deep a certain move is explored, and g3 having a significantly longer line is certainly the most reliable.
Attachment: test2.JPG (168k)
Parent - - By SR (****) Date 2008-08-20 06:26
What is your current main line after 25.g3  (e.g. the line with the highest score for white, and best responcs for black)?
Parent - - By Arkansaw (***) Date 2008-08-20 11:20
Just got back from work :)

g3 is still occupying the giant share of position count, and a perpetual line has been reached. The alternatives are being explored very slowly as there are way too many positions that require 1min+ to resolve at depth 13 (somewhat better now for g3 that has already reached some very deep lines). After the first move, there seem to be few alternatives for either side; the evals often differ so much so early that IDeA by itself will not spend too much time exploring them.
Attachment: test3.JPG (186k)
Parent - - By SR (****) Date 2008-08-20 11:34
How does white respond to 25.g3,Kh7? What are whites candidate move after 25....Kg7 in your IDeA?  I will then make further suggestions for black! 
Parent - - By Arkansaw (***) Date 2008-08-20 11:38
unexplored until now, at depth 13 it differs from the best move Rh7 by +0.85
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka 3 on Skulltrail - A very interesting lesson
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill