Announcement: Open 10x8 WB engine tournament

[TonyJH]

As I didn't make rules for this, I would like to know if there are any objections from your remaining opponents this cycle (TJchess and Smirf Donation) if I would decide to replace it now. They would be disadvantaged a little compared to their competitors if they had to play a stronger BigLion.

In any case I can replace it before the next cycle starts.

No objection from me.

[hgm]

No veto from SMIRF against replacements, if SMIRF then also would be replaced by MS-173g.

Reinhard.

You already have it?

[smrf]

Yes. See e.g. at http://www.chessbox.de/Compu/schachsmirf_e.html.

[hgm]

OK, I installed both the new Smirf and BigLion. They are still playing under the name of the previous versions (this is needed to not mess up the standings and cross table), don't be confused about that. I will fix that for the next cycle.

[Matthias Gemuh]

OK, I installed both the new Smirf and BigLion. They are still playing under the name of the previous versions (this is needed to not mess up the standings and cross table), don't be confused about that. I will fix that for the next cycle.

I still doubt the stability of the new BigLion80, so shout out loud if you notice a crash/hang. He has however outplayed Smirf once today to score a 1:1

Matthias.

[hgm]

Yes, it seems Smirf miscalculated on 14. Axe2 {+0.91/12}, while after the obvious recapture its score drops to -1.83/11.

High time I make the html viewer under-promotion resistant, btw...

[hgm]

Another interesting game in which the Archbishop plays a dominant role is this one, from the second cycle. (Note the first 6 moves were forced from a PGN to set up the Bird-Chess opening position.)

Code: Select all

[Event "Computer Chess Game"]
[Site "SCHAAK_PC"]
[Date "2008.04.05"]
[Round "1.4"]
[White "TSCP Gothic"]
[Black "Smirf Donation"]
[Result "1-0"]
[TimeControl "3300+5"]
[Variant "capablanca"]
[FEN "rnbqckabnr/pppppppppp/10/10/10/10/PPPPPPPPPP/RNBQCKABNR w KQkq - 0 1"]
[SetUp "1"]
[Number "7"]

{--------------
r n b q c k a b n r
p p p p p p p p p p
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
P P P P P P P P P P
R N B Q C K A B N R
white to play
--------------}
1. Cf3 Cf6 2. Cd4 Cd5 3. Cb5 Cb4 4. Cc3 Cc6 5. Qe1 Qe8 6. Cd1 Cd8 
{ Engines start to play from here: }
7. d4 {+0.04/12 1:45} d6 {-0.19/11 2:52} 8. Nc3 {+0.15/11 1:41}
g6 {-0.17/10 2:37} 9. Nh3 {+0.29/11 1:38} Nh6 {-0.11/10 1:45} 10.
Nd5 {+0.23/10 1:34} Na6 {+0.22/10 1:11} 11. c3 {+0.03/10 1:31}
c6 {+0.45/10 1:40} 12. Ne3 {+0.10/11 1:28} Bg7 {+0.57/11 1:36} 13.
j4 {+0.02/11 1:25} Af6 {+0.56/11 1:56} 14. j5 {-0.09/11 1:22}
Ni4 {+0.74/10 1:32} 15. Rj3 {-0.15/11 1:20} h5 {+0.79/10 1:36} 16.
g3 {-0.12/10 1:17} O-O {+0.70/9 1:30} 17. Bg2 {-0.14/10 1:14}
c5 {+0.83/9 1:39} 18. dxc5 {+0.17/10 1:12} Nxc5 {+0.99/9 1:09} 19.
Nj4 {+0.01/9 1:09} Be6 {+1.48/9 1:20} 20. h3 {-0.29/10 1:07}
Nh6 {+1.00/10 1:51} 21. h4 {-0.29/10 1:05} Qb5 {+0.89/8 1:40} 22.
j6 {+0.08/10 1:03} ixj6 {+1.03/9 1:11} 23. Nc2 {+0.05/10 1:01}
Qa4 {+1.28/9 1:31} 24. Nd4 {+0.08/10 59} Ni4 {+1.11/9 2:18} 25.
Nxe6 {+0.08/10 57} Nxe6 {+0.64/10 1:37} 26. b3 {+0.24/10 55}
Qa5 {+0.81/9 1:06} 27. Qd2 {+0.25/10 53} Rh6 {+1.06/11 1:53} 28.
Bb2 {+0.36/10 51} Ag4 {+1.06/9 2:36} 29. Rh3 {+0.46/9 49}
Rh8 {+1.26/9 1:19} 30. Qd5 {+0.32/9 48} Qa6 {+1.35/9 1:11} 31.
i3 {+0.31/9 46} Rc8 {+1.63/8 1:05} 32. Af3 {+0.19/9 44} Rc5 {+1.52/9 59}
33. Qd3 {+0.06/10 43} Qxd3 {+1.28/10 55} 34. Cxd3 {+0.04/10 42}
Af5 {+1.32/10 1:05} 35. e4 {+0.01/10 40} Ah6 {+1.48/10 47} 36.
b4 {-0.13/10 39} Rc4 {+1.40/10 58} 37. Cd5 {-0.01/9 37} Cc8 {+1.87/9 36}
38. Ad1 {-0.54/9 36} Nc7 {+1.65/8 34} 39. Cd3 {-0.31/9 35} Nb5 {+2.31/9 52}
40. Rc1 {-0.35/9 34} Axc1 {+2.67/9 27} 41. Cxc1 {-0.93/11 33}
Nxc3 {+2.49/9 23} 42. Ae3 {-1.01/10 32} Rxb4 {+3.49/9 15} 43.
Bxc3 {-1.31/11 30} Cxc3 {+4.07/10 19} 44. Cxc3 {-1.21/11 29}
Bxc3 {+4.01/11 19} 45. Bf3 {-1.29/11 28} Bd4 {+4.01/9 17} 46.
Ac2 {-1.52/11 27} a5 {+3.44/10 11} 47. Nxh5 {-1.64/11 27}
gxh5 {+4.65/11 15} 48. e5+ {-0.79/11 26} Ki7 {+4.11/11 11} 49.
exd6 {-1.16/11 25} exd6 {+4.17/10 13} 50. a3 {-1.56/11 24}
Rc8 {+4.53/12 16} 51. Ad3 {-1.93/11 23} Rb1+ {+5.10/11 7} 52.
Kg2 {-1.89/10 22} Rj1 {+4.28/10 10} 53. Bxh5 {-1.40/11 22}
Rj2+ {+4.38/11 9} 54. Kh1 {-1.54/10 21} Rc3 {+4.36/11 8} 55.
Af4+ {-2.00/11 20} Nh6 {+4.14/12 11} 56. Rh2 {-1.93/11 19}
Rxh2+ {+3.99/12 6} 57. Kxh2 {-1.59/11 19} Rc2 {+3.99/11 4} 58.
Axd6 {-1.50/11 18} Rxf2+ {+4.06/11 5} 59. Ki1 {-1.42/10 18} b6 {+3.89/10 5}
60. a4 {-1.47/11 17} Bc5 {+4.04/10 2} 61. Ae4 {-1.48/11 16} f5 {+3.92/11 4}
62. Ad3 {-1.31/12 16} Bd4 {+3.90/12 8} 63. i4 {-1.48/12 15} j5 {+4.18/12 5}
64. i5 {-0.60/11 15} Ni4 {+3.37/12 3} 65. Ac1+ {-0.17/12 14}
Kh8 {+2.46/13 3} 66. Bxi4 {-0.11/13 14} jxi4 {+2.46/12 3} 67.
Ah6 {+0.00/12 13} Rf1+ {+1.99/12 2} 68. Kh2 {+0.00/11 13} Kh7 {+1.80/12 2}
69. Aj5+ {+0.01/12 12} Kh8 {+0.98/15 4} 70. Axi4 {+0.65/12 12}
Rf2+ {+1.07/14 3} 71. Kh3 {+0.80/11 11} Kg7 {+1.07/12 5} 72.
Ah6+ {+1.26/12 11} Kg6 {+0.49/12 1} 73. Ai8+ {+2.20/12 11} Kf6 {-0.10/13 1}
74. Axj7 {+2.26/12 10} Ke7 {-0.12/12 4} 75. Ah6 {+3.05/12 10}
Kd7 {-0.34/11 3} 76. i6 {+3.26/13 10} Bh8 {-2.26/11 0} 77. Ae3 {+3.26/12 9}
Re2 {-2.09/12 2} 78. Axb6+ {+3.26/12 9} Ke7 {-2.21/13 1} 79.
Axa5 {+3.27/12 9} Re6 {-2.15/11 2} 80. h5 {+3.27/12 8} Bi7 {-2.24/11 0} 81.
Ac3 {+3.26/12 8} Re3 {-1.91/10 0} 82. Ad5+ {+3.78/11 8} Kd7 {-1.89/11 0}
83. a5 {+5.66/11 7} f4 {-1.31/10} 84. a6 {+5.80/11 7} Rxg3+ {-3.13/11 0}
85. Ki2 {+6.01/11 7} Kd6 {-4.12/11 0} 86. Ae4+ {+6.69/12 7}
Ke5 {-5.34/13 2} 87. Axg3 {+7.80/13 6} fxg3 {-8.19/14 1} 88.
a7 {+7.94/14 6} g2 {-8.58/13 0} 89. Kh2 {+8.83/14 6} g1=R {-8.83/12 0} 90.
Kxg1 {+8.97/13 6} Be3+ {-10.17/13 0} 91. Kf1 {+9.08/14 6}
Bxa7 {-10.11/12 0} 92. i7 {+9.11/14 5} Be3 {-11.09/11 1} 93.
i8=Q {+9.23/13 5} Ba7 {-11.18/10 1} 94. Qh8+ {+17.88/12 5}
Kf4 {-13.64/11 0} 95. h6 {+18.08/12 5} Ke3 {-15.22/11 1} 96.
Qe5+ {+99.83/12 3} Kd3 {-16.68/12 1} 97. Qf5+ {+99.83/1 0}
Kc3 {-20.54/11 0} 98. Qa5+ {+99.83/8 0} Kb2 {-24.37/10 1} 99.
Qxa7 {+99.83/8 0} Kc2 100. Qc5+ {+99.83/2 0} Kb3 101. Qb6+ {+99.83/4 0} Ka4
102. Qd4+ {+99.83/6 0} Kb5 103. h7 {+99.91/7 0} Kc6 104. h8=Q {+99.91/4 0}
Kb5 105. Qb8+ {+99.95/1 0} Ka6 106. Qbb6# {+99.99/1 0}
{Xboard adjudication: Checkmate} 1-0

After some initial skirmishes, in which TSCP-G (white) sacs the j-Pawn to expose Smirf's King, and managed to destroy Smirf's bishop pair, the folowing position occurs:



Smirf decides here to sacrifice its Archbishop against Rook plus 3 Pawns: 40. ..., Axc1 41. Cxc1, Nxc3 42. Ae3, Rxb4. TSCP-G avoids the actual loss of the third Pawn (e4, which is hanging by the R+N attach, while at the same time Bb2 is hanging) by giving its Bishop plus the pair advantage for a Knight through 43. Bxc3, which is tantamount to losing a Pawn (on 10x8 even an unpaired Bishop is worth half a Pawn more than a Knight).

TSCP-G then mysteriously sacs a Knight with 47. Nxh5, for a discovered check on Smirf's King. It seems all this sacrifies buys it is somewhat more reduction of the black King safety, and it is very questionable if in this stage of the game King safety is still that much of a concern.

The material imbalance thus achieved is preserved as the players trade down to the end-game, until the following position occurs after 63. i4:



TSCP-G is leading here by an Archbishop against Rook + Knight + Pawn, with the additional advantage that Smirf's edge Pawn is badly doubled. (Black's Rook-on-2nd is not worth very much here, as the second rank is already evacuated.) Due to bad evaluation, however, TSCP-G evaluates this positions as -1.48, while Smirf even has it at -4.13 (for white).

Smirf then pushes its doubled j-Pawn, effectively initiating a trade that solves it by swapping its N+P for white's B (63. ..., j5 64. i5, Ni4 65. Ac1+, Kh8 66. Bxi4, jxi4, after which the black i-Pawn is undefendable). The situation on the board is then A+4P vs R+B+5P, which TSVP-G evaluates as equal, and Smirf as -2.46 (for white).

But black has basically no attack chances at all: he cannot use the R+B to attack white Pawns more often than they can be defended, as the a-Pawn is not on the Bishop's color, and can be easily defended by the Archbishop against any Rook attack, and the other white Pawns are under care of the King. On the contrary, due to its King being in the open, black has to take great care that its R and B will not be lost through checking combinations.

The Archbishop then goes on the rampage, and quickly gobbles up most of black's Pawns, leaving white with connected passers. White sacs its Archbishop for the Rook (87. Axg3)when it sees that the lone Bishop will not be able to stop the passers on both wings, and subsequently wins easily by promoting to two Queens.

This game gives a typical example of how the Archbishop beats Rook + minor in the end-game (in the presence of Pawns).

[smrf]

Dear Harm,

SMIRF bases only on its own code. Thus it does not surprise, that there are a lot of weaknesses to be found. The slightly more mature MS-173g (or the coming MS173h) might have a little fewer of them.

But the claimed wrong behaviour has nothing to do with maybe unprecise piece exchange values. SMIRF is my first playing chess program, even now having a lot of repaired parts. But when reviewing all those still existing want-to-be extinguished design weaknesses of it, the playing strength of SMIRF is completely surprising.

There will be a completely rewritten program Octopus as a successor someday. Its detail evaluation will be better and probably 10 times faster. SMIRF does not even support a mobility factor yet. And SMIRF's reuse of cached results is a shame.

Thus I suggest to avoid conclusions on wrong or right piece exchange values from its behaviour. Maybe that could be tried to be investigated by two similar mature programs someday in the future.

Reinhard.

[hgm]

Well, I only point out that Smirf (and TSCP to a lesser extent) give an unrealistically low evaluation of the white position after 63. i4. I cannot with certainty explain why they do this, of course. The problem with root scores is that they come from leaves many moves away, so you never really know what exactly the engines are evaluating. It could be that Smirf only evaluates end-leaf positions where he first has captured 4 Pawns, because a search bug prevents him to search captures of the opponent. But I don't consider that very likely.

The persistency with which Smirf keeps such strange scores over many moves suggests the problem is in the evaluation. Search bugs manifest themselves by the score being high when the engine starts the tactical exchange, and then suddenly dropping as the opponent plays the replies that the search overlooked. Smirf usually seems very happy after it has sacrificed an Archbishop for a pair of minors or a Rook and a few Pawns.

And I think you did confirm that Smirf uses piecevalues like N=3.05, B=3.66 and A=6.89. And such piece values would exactly lead to misevaluations of the type we see, where an inferior or nearly lost material imbalance is evaluated as +2 or +3. Even if everything else works perfectly.

It might be a good idea to have Smirf also print the evaluation of the root position itself. For me this helped to detect many evaluation bugs. Even better (but a lot harder to implement) would be to follow the PV to the end leave where the score comes from, and print that position (so it can be correlated with the score).

The main reason I show the game is to illustrate the power of the Archbishop in the end-game, where it can easily overwhelm R+N. I don't think it is due to bad play of Smirf (e.g. due to bugs or time pressure) that it loses from the position after 64. i4. This is simply a lost end-game for black.

[smrf]

Well, Harm, repeatedly I have tried to explain, that at an exchange there are vanished not only the values of disappearing pieces, but also the positional evaluations of influenced squares will change. Thus the value of any piece is not merely defined by its average exchange value. But it is a hard task, to separate those two components the right way. In the "-X" versions of SMIRF I have tried to use a new mixture of those parts. And it seems to be a promising approach. Because of that, experiences coming from playing the old donation "BC-" version of SMIRF are a little bit out of my focus ...

Nevertheless SMIRF seems to be at the end of its life cycle.

Reinhard.