OpenChess

http://strelkachess.narod.ru/Strelka5.html

15/10/2011

After a long break after the tumultuous events in the world
computer chess lately, I decided to publish the latest version
The arrow. This publication was preceded by a lot of work, but it was conducted
in private mode. Were made by dozens of versions of the series 3 and 4.
The published version is 5.

I will not dissemble, and say that Arrow 5 is 100 per cent
the original engine. I used a lot of things that could spy
in other programs. This is primarily fish 3, Hippolyta, Houdini, and Kritter
Stokfish. In most algorithms used Houdini version 1.5a,
for that I bring my gratitude to Robert Gudartu and apologies.

But if the arrow 5 would contain only strangers algorithms
I would not publish it. Post the engine only makes sense if
if there is something new, something others do not. Innovations in the arrow 5
quite a lot. The most important innovations - in the evaluation function.

Unfortunately, not all of the planned and managed to make the test.
Because of this, Arrow 5 can sometimes produce strange results, especially
country may be the importance of evaluating the position. This is due to the fact that the assessment
based not only on the basis of the current position, but also as a function depending
from a chain of moves that led to this position. The weights of the various parameters of the engine
not subjected to rigorous testing because of time constraints, so
I have a big request does not kicked over the strange results of the arrows.

As with previous versions of arrows, the new version is 32-bit and
uniprocessor. Therefore, it is difficult to compete with modern 64-bit
and multi-processor monsters. But in his weight class Arrow 5
quite competitive.

Arrow 5 can be downloaded here

Thanks for the information but I'll pass.

After so many of his Strelka releases, I have to ask the obvious: has the author of Strelka never heard of 64-bit architecture and multiprocessor? Doesn't he get tired of 32-bit @ 1cpu?

I'm kind of glad Critter, Stockfish, IvanHoe series (Houdini/Tactico Pro, Deep Saros, Leopard, Vitruvius, etc), Crafty, Spark etc are available in 32 & 64-bit multiprocessors AND provide much greater Hash size for analysis.

Again, I'll pass his 32-bit @ 1cpu engines and it's limitations....

The engine is actually very strong. It outputs wild evaluation score though.

[Event "Microsoft, Blitz:1'"]
[Site "VirtualBox"]
[Date "2011.10.15"]
[Round "13"]
[White "Houdini 1.5a x64"]
[Black "Strelka 5"]
[Result "0-1"]
[ECO "B19"]
[Annotator "0.52;0.98"]
[PlyCount "144"]
[TimeControl "60"]

{1 Core, 128 MB Hash, Ponder-Off, EGTB-Off W=14.8 ply; 1,418kN/s B=14.0 ply;
873kN/s} 1. e4 c6 2. d4 d5 3. Nc3 dxe4 4. Nxe4 Bf5 {Both last book move} 5. Ng3
{0.52/15 2} Bg6 {0.98/14 2} 6. h4 {(Nf3) 0.43/14 1} h6 {0.88/14 1} 7. Nf3 {
(Bd3) 0.30/14 3} Nf6 {0.92/14 1} 8. h5 {0.24/15 2} Bh7 {0.92/12 0} 9. Bd3 {
0.18/15 1} Bxd3 {0.22/14 1} 10. Qxd3 {0.24/15 1} e6 {0.16/16 5} 11. Be3 {
0.18/14 1} Bd6 {0.18/15 1} 12. Ne2 {(0-0-0) 0.08/15 1} O-O {-0.24/15 1} 13.
O-O-O {0.09/15 1} Nbd7 {-0.18/15 1} 14. Kb1 {0.05/15 2} b5 {(Qc7) -0.22/14 3}
15. Rh4 {0.03/13 1} Nd5 {-0.17/14 1} 16. Bf4 {(g4) 0.05/14 1} Nxf4 {-0.37/13 1}
17. Nxf4 {0.06/12 0} a5 {(Qc7) -0.27/14 3} 18. Ne2 {0.02/14 1} Re8 {
(a4) -0.39/13 1} 19. Rhh1 {(Nc3) 0.01/13 1} Qc7 {(Nf6) -0.55/13 1} 20. Nc3 {
-0.05/14 1} a4 {(f5) -0.54/13 1} 21. g4 {-0.09/15 1} b4 {-0.48/13 1} 22. Ne4 {
-0.23/14 1} Bf4 {(b3) -0.56/13 1} 23. g5 {-0.02/13 2} b3 {-0.54/13 2} 24. a3 {
(Rhg1) -0.13/12 1} c5 {(bxc2+) -0.74/12 1} 25. c3 {(c4) 0.03/12 1} c4 {
-0.27/12 1} 26. Qe2 {0.00/13 1} Ra5 {-0.16/12 1} 27. Nfd2 {0.17/13 1} Bxg5 {
(Rc8) -0.14/13 1} 28. Nxc4 {0.18/14 1} Rd5 {0.22/14 0} 29. Rdg1 {0.17/15 0} Qf4
{(Nf6) 0.22/14 3} 30. Rg4 {0.23/15 1} Qc7 {0.31/15 1} 31. Nxg5 {(Rg2) 0.17/14 1
} Rxg5 {0.21/15 1} 32. Rxg5 {0.23/15 0} hxg5 {0.21/14 0} 33. h6 {0.12/15 1} g6
{(Qf4) 0.14/15 1} 34. Nd2 {(Ka1) 0.22/14 1} Qf4 {0.09/13 0} 35. f3 {
(Ne4) 0.14/14 1} Kh7 {(Nf6) -0.15/13 0} 36. Nc4 {(Ka1) 0.01/12 1} e5 {
(Qf5+) -0.50/14 1} 37. Re1 {-0.09/14 0} Kxh6 {-0.89/13 0} 38. dxe5 {
(Ka1) -0.19/13 1} Kg7 {-0.99/15 0} 39. Qd3 {(Ka1) -0.41/14 1} Rxe5 {
(Qf5) -1.33/13 1} 40. Nxe5 {-0.38/13 0} Nxe5 {-1.33/12 0} 41. Qd8 {-0.29/15 0}
Qh4 {-1.18/15 0} 42. Rd1 {-0.23/13 1} Nxf3 {-1.18/13 0} 43. Ka1 {-0.31/15 0} g4
{-1.57/14 1} 44. Qa8 {(Qxh4) -0.30/16 0} Qh5 {-1.81/14 0} 45. Rf1 {-0.46/15 1}
Qh3 {(Qh2) -1.65/13 0} 46. Rc1 {-0.53/13 1} Qh6 {(Qg2) -2.72/14 1} 47. Rb1 {
-0.56/16 0} Qf4 {-2.65/13 0} 48. Rf1 {-0.54/16 0} Qe3 {(Qh2) -2.30/14 1} 49.
Rh1 {-0.80/14 0} Nh2 {-2.30/13 0} 50. Qd5 {-0.61/14 1} g3 {-2.20/14 0} 51. c4 {
-0.51/14 1} g5 {-2.18/15 0} 52. Qg2 {-0.37/15 1} Qf4 {(Qf3) -2.06/15 3} 53. c5
{-0.57/13 1} Nf3 {-1.82/13 1} 54. Qh3 {-0.70/12 0} Nh4 {-1.80/14 0} 55. c6 {
-0.11/13 2} Kg6 {(g2) -0.11/14 2} 56. Qd7 {-0.11/14 0} g2 {-0.27/12 0} 57. Rb1
{-0.11/13 0} Qf1 {-0.02/13 0} 58. Qd1 {-0.11/15 0} Qf4 {-0.08/14 0} 59. Qd8 {
(Qd7) -0.11/14 0} Qf1 {(Qe4) -0.26/12 0} 60. Qb6 {(Qd1) 0.16/14 0} Qf6 {
0.03/12 0} 61. Qc5 {0.16/12 0} Nf5 {0.14/13 0} 62. c7 {-0.04/14 1} Ne7 {
0.14/12 0} 63. Qxe7 {-0.04/15 0} Qxe7 {0.36/11 0} 64. c8=Q {-0.02/14 0} Qf6 {
0.07/13 0} 65. Qg8+ {-0.04/14 0} Kh6 {0.02/14 0} 66. Qe8 {-0.04/15 0} Qf3 {
-0.20/13 0} 67. Rg1 {(Rc1) -0.04/15 0} Qf1+ {(f5) -5.30/13 0} 68. Qe1 {
-1.78/14 0} Qxe1+ {-5.30/12 0} 69. Rxe1 {-1.62/1 0} f5 {-5.04/12 0} 70. Rg1 {
-7.66/17 1} g4 {-5.02/11 0} 71. Rxg2 {-7.66/17 0} Kg5 {-5.02/10 0} 72. Rh2 {
(Kb1) -7.66/17 0} f4 {(g3) -9.33/12 0} 0-1

I haven't bothered to test/use it for obvious reasons. The wild evaluation scores reminds me of Stockfish's nongranular scores. Maybe this is an inherited trait from Stockfish.

Prima wrote:I haven't bothered to test/use it for obvious reasons. The wild evaluation scores reminds me of Stockfish's nongranular scores. Maybe this is an inherited trait from Stockfish.

According to Jury the evaluation of a position somehow depends on the moves which lead to this position (that's something truly unique).

Well, engines tend to do that, specially when dealing with repetition of positions, that's why when analyzing a game, it's better to give the engines the full line, instead of giving them a FEN.

ThinkingALot wrote:

Prima wrote:I haven't bothered to test/use it for obvious reasons. The wild evaluation scores reminds me of Stockfish's nongranular scores. Maybe this is an inherited trait from Stockfish.

According to Jury the evaluation of a position somehow depends on the moves which lead to this position (that's something truly unique).

So in essence, Strelka 5's analysis [scores] are not reliable. Anyway, thanks for the info.

Prima wrote:So in essence, Strelka 5's analysis [scores] are not reliable. Anyway, thanks for the info.

If they were not reliable it wouldn't be so strong! And it's incredibly strong. So the scores are just unstable a bit:).

at least these scores look interesting. what i see in the moment it also plays very aggressive. e.g. moving pawns like hell a6-b5-b4 and in the same game h6-g5...
very entertaining....

strelka 5 is not only strong, it plays amazing games with amazing evaluations.

a mixture out of THE KING (johan de koning) and CSTAL (c.whittington) and Nimzo (chrilly donninger) in aggressive mode,
with evaluations few times bigger then the opponent program, so similar like mchess or CSTAL.

when the normal rybka-clones all have an evaluation that is LOWER than normal , strelka comes with speculative/full-moon-scores.

a very interesting program.


the way it organizes SPACE and TIME on the board reminds me very much on the king.

the opponent is often paralyzed. often the opponents are not coming above the 4th or even 3rd rank while strelka attacks like hell.

wow !!

congratulations. this is really cool and unsound chess.


there is something unique in strelka 5.

you must watch the games against it. amazing.

maybe a new approach ?!