Junior 2010 : The new Tal ?

[Damir]

As far as I heared there may be an update coming soon... When, I think is best to ask Amir&co. CCRL-CEGT have stoped testing Deep Junior 2010,because there is an update on its way

[M ANSARI]

I wonder if this is something to look at. I guess mate is mate, but it would be good to setup a few positions to see why Junior doesn't see this trivial mate in 2 and goes for a much longer mate.

[d]6k1/5R2/6K1/8/8/8/8/8 w - - 0 117

Here Junior played Kf6 instead of the obvious rook waiting move and then mate.

[Rolf]

I wonder if this is something to look at. I guess mate is mate, but it would be good to setup a few positions to see why Junior doesn't see this trivial mate in 2 and goes for a much longer mate.

I slightly disagree. Kf6 shows that Junior knows the so called triangle manoever. Kf6-Kg(f)5!-Kg6 which simply changed that Black must move in difference to the starting position, so a really smart thing. To move the R is much "longer" because it has to calculate so many fields on the f-file. From the aspect of efficieny the triangle might be optimal but I'm not a programmer yet.

[Robert Flesher]

I wonder if this is something to look at. I guess mate is mate, but it would be good to setup a few positions to see why Junior doesn't see this trivial mate in 2 and goes for a much longer mate.

I slightly disagree. Kf6 shows that Junior knows the so called triangle manoever. Kf6-Kg(f)5!-Kg6 which simply changed that Black must move in difference to the starting position, so a really smart thing. To move the R is much "longer" because it has to calculate so many fields on the f-file. From the aspect of efficieny the triangle might be optimal but I'm not a programmer yet.

Correct!

The basic fundamentals of zugzwang. I do not use tablebases for this very reason. I like to see what the engine understands.

[Robert Flesher]

I wonder if this is something to look at. I guess mate is mate, but it would be good to setup a few positions to see why Junior doesn't see this trivial mate in 2 and goes for a much longer mate.

[d]6k1/5R2/6K1/8/8/8/8/8 w - - 0 117

Here Junior played Kf6 instead of the obvious rook waiting move and then mate.

...but Junior does see it in analysis mode

68: Junior 2010 UCI - Rybka 3, Microsoft, Blitz:10'
6k1/5R2/6K1/8/8/8/8/8 w - - 0 1

Analysis by Junior 2010 UCI:

117.Rf2 Kh8
+- (#2) Depth: 3 00:00:00
117.Rf2 Kh8
+- (#2) Depth: 3 00:00:00
117.Rf2 Kh8
+- (#2) Depth: 12 00:00:00 2kN
117.Rf2 Kh8
+- (#2) Depth: 12 00:00:00 2kN
117.Rf2 Kh8
+- (#2) Depth: 12 00:00:00 2kN
117.Rf2 Kh8
+- (#2) Depth: 12 00:00:00 3kN
117.Rf2 Kh8
+- (#2) Depth: 21 00:00:00 7kN
117.Rf2 Kh8
+- (#2) Depth: 24 00:00:00 10kN
117.Rf2 Kh8
+- (#2) Depth: 24 00:00:00 10kN
117.Rf2 Kh8
+- (#2) Depth: 27 00:00:00 13kN
117.Rf2 Kh8
+- (#2) Depth: 27 00:00:00 13kN
117.Rf2 Kh8
+- (#2) Depth: 30 00:00:00 17kN
117.Rf2 Kh8
+- (#2) Depth: 33 00:00:00 21kN
117.Rf2 Kh8
+- (#2) Depth: 36 00:00:00 26kN
117.Rf2 Kh8
+- (#2) Depth: 39 00:00:00 32kN
117.Rf2 Kh8
+- (#2) Depth: 42 00:00:00 38kN
117.Rf2 Kh8
+- (#2) Depth: 45 00:00:00 45kN
117.Rf2 Kh8
+- (#2) Depth: 48 00:00:00 53kN
117.Rf2 Kh8
+- (#2) Depth: 51 00:00:00 62kN
117.Rf2 Kh8
+- (#2) Depth: 54 00:00:00 71kN
117.Rf2 Kh8
+- (#2) Depth: 57 00:00:00 82kN
117.Rf2 Kh8 118.Rf8#
+- (#2) Depth: 60 00:00:00 93kN
117.Rf2 Kh8 118.Rf8#
+- (#2) Depth: 60 00:00:00 105kN

(, Microsoft 20.01.2010)

[perejaslav]

[Don]

I tried this position on Doch, and Doch plays g4 in 2 minutes on 1 CPU, slow core 2 machine. So Doch would not play this in Blitz.

Doch does have pawn storm code in it. I'm pretty sure this move would require enormous CPU time without this kind of code - but I could of course be wrong.

One of the Rybka clones plays this move in 15 seconds.

Stockfish 1.6 would choose this move after 15 seconds but shortly thereafter decides to play something else and does not return to g4 in any reasonable amount of time.

So it appears to me that this is basically an evaluation issue and not a move that is likely to played based on tactics.

I think it illustrates that you cannot have a strong program without putting a lot of attention into the evaluation. At some point it will become the bottleneck because I doubt doch would find this move (assuming it's the best move) in any reasonable amount of time unless the evaluation is helping it.

Here is a game Junior 2010 one core against Rybka 3.0 on 4 cores. I am always amazed at the complications Junior can create. Move # 23. g4 !!


[d]3rr1k1/1qbb1ppp/p4n2/n1p1p3/PpP5/1P1PN2P/1BBN1PP1/R2QR1K1 w - - 0 23





[Event "Microsoft, Blitz:10'"]
[Site "Microsoft"]
[Date "2010.01.18"]
[Round "3"]
[White "Junior 2010 UCI"]
[Black "Rybka 3"]
[Result "1-0"]
[ECO "C92"]
[PlyCount "243"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 d6 8. c3
O-O 9. h3 Bb7 10. d3 Na5 11. Bc2 c5 12. Nbd2 Re8 13. Nf1 Qc7 14. Ng3 d5 15.
exd5 Bxd5 16. Nf5 Bd8 17. a4 b4 18. Ne3 Bc6 19. Nd2 Qb7 20. c4 Bc7 21. b3 Rad8
22. Bb2 Bd7 23. g4 Be6 24. g5 Nd7 25. Ne4 Bxh3 26. Qh5 Be6 27. Kh2 Qc6 28. Rg1
Nb7 29. Raf1 Bb8 30. Bd1 Bd6 31. Rh1 Nf8 32. Kg1 Be7 33. Bxe5 Rxd3 34. Bf3 Bd7
35. Nf6+ Bxf6 36. Bxc6 Rxe5 37. Be4 g6 38. Bxd3 gxh5 39. gxf6 Na5 40. Bc2 Bc6
41. Rh3 Nd7 42. Rd1 Nxf6 43. Rd8+ Kg7 44. Nf5+ Kg6 45. Nd6+ Kg7 46. f4 Re1+ 47.
Kf2 Rc1 48. Nf5+ Kg6 49. Ne3+ Kg7 50. Rg3+ Ng4+ 51. Ke2 f5 52. Kd2 Rh1 53.
Nxf5+ Kf7 54. Rc8 Rh2+ 55. Kc1 Ke6 56. Ng7+ Kd7 57. Re8 Rh1+ 58. Kd2 Kc7 59.
Re1 Rh2+ 60. Re2 Rh1 61. Ne6+ Kb6 62. Bxh7 h4 63. Rxg4 Nxb3+ 64. Ke3 Rh3+ 65.
Kf2 Rh2+ 66. Kf1 Nd2+ 67. Rxd2 Rxd2 68. a5+ Kxa5 69. Nxc5 Kb6 70. Nb3 Rd8 71.
Rxh4 a5 72. Rh6 a4 73. Be4 axb3 74. Rxc6+ Ka5 75. Rc5+ Ka4 76. Rd5 Rh8 77. c5
b2 78. c6 Rh4 79. Rf5 Rh1+ 80. Ke2 Rc1 81. Kd2 Rc4 82. Bb1 b3 83. Re5 Rxc6 84.
Re1 Kb4 85. f5 Rd6+ 86. Ke3 Rc6 87. Be4 Rf6 88. Kd4 Rf8 89. Rf1 Rd8+ 90. Ke3
Rf8 91. f6 Kc5 92. f7 Kc4 93. Bd3+ Kd5 94. Kd2 Kc5 95. Kc3 Kd6 96. Kxb2 Kd5 97.
Rf3 Ke5 98. Bc4 Kd4 99. Rf4+ Ke5 100. Rg4 Kf6 101. Rg8 Rxf7 102. Bxf7 Kxf7 103.
Rg5 Ke6 104. Kxb3 Kf7 105. Kc4 Kf6 106. Rh5 Kg6 107. Re5 Kf7 108. Kd5 Kf6 109.
Kd6 Kg7 110. Rf5 Kh6 111. Ke5 Kg7 112. Rf6 Kg8 113. Ke6 Kg7 114. Kf5 Kh7 115.
Rf7+ Kh8 116. Kg6 Kg8 117. Kf6 Kh8 118. Rb7 Kg8 119. Rb8+ Kh7 120. Rb1 Kg8 121.
Rh1 Kf8 122. Rh8# 1-0

[fern]

In fact, Don, beyond an horizon of, say, 100 moves, evaluation bis everything. Inside this 10 moves, tactic is evberything instead.
Which is the use to make a tree until move 45 if you does not know which is the best of all the results?
If with specific tactics chess cannot be solved, then the shortcuts of positional acumen are neccesary.

my best
Fern

[M ANSARI]

I am not convinced g4 is the best move in this position. Of course it was a nice win by Junior, but it also looks like black could have defended better. Junior will win a spectacular game every once in a while, but to get that one win it will lose 20 in equally spectacular fashion. Some of the most beautiful attacking games I have ever seen was with Rybka Dynamic on high contempt, but the increased amount of losses drown out the few dramatic wins.

[d]3rr1k1/1nbn1ppp/p1q1b3/2p1p1PQ/PpP1N3/1P1PN3/1BB2P1K/5RR1 b - - 0 29

In this position Rybka played Bb8 moving an important defending piece to an inactive square. With more time Rybka 3 would choose Bd6, which would make much more sense as they bring in the piece to defend rather than move away from the attack. In such a sharp position any small error can immediately result in a dramatic loss. That is the beauty of chess, especially in human chess. Here is a 5 minute analysis of R3 on the position mentioned


3rr1k1/1nbn1ppp/p1q1b3/2p1p1PQ/PpP1N3/1P1PN3/1BB2P1K/5RR1 b - - 0 1

Analysis by Rybka 3:

29...Nd6 30.Nf6+
-/+ (-0.75) Depth: 6 00:00:00 0kN
29...Nd6 30.Nf6+
-/+ (-0.75) Depth: 7 00:00:00 0kN
29...Bb8 30.Rg3
=/+ (-0.63) Depth: 8 00:00:00 0kN
29...Rf8 30.Nf6+ gxf6 31.gxf6+
=/+ (-0.58) Depth: 9 00:00:00 0kN
29...Rb8 30.Rg3
=/+ (-0.40) Depth: 10 00:00:00 0kN
29...Rc8 30.Rg3
=/+ (-0.46) Depth: 10 00:00:00 0kN
29...Rb8 30.Rg3
=/+ (-0.40) Depth: 11 00:00:00 0kN
29...Nf8 30.Nf6+
= (0.00) Depth: 12 00:00:00 0kN
29...Nf8 30.Nf6+
= (0.00) Depth: 13 00:00:00 34kN
29...Bd6 30.Rh1
= (0.00) Depth: 14 00:00:00 34kN
29...Bd6 30.Rh1
= (0.00) Depth: 15 00:00:00 34kN
29...Bd6 30.Rh1
= (0.00) Depth: 16 00:00:00 34kN

(, Microsoft 20.01.2010)

[M ANSARI]

Actually Bd6 doesn't seem to work if white plays rh1! as seen with further analysis.

Bd6 Rh1! g6 Qh4 h5 gxh6 Be7 h7+ Kh8 Qg3 Bf8 f4! looks very strong.

One thing for sure is it needs more analysis. I wish I had Junior 2010 to check and see what moves it would have chosen.