Andscacs - New version 0.88

[cdani]

Thanks! I will follow your test. On it the improved pondering should add something more, and also of course the 4 pieces Syzygy bases.

I added them, but I personaly doubt that 4 pc tbs add a single elo in gameplay.
They are fine in Analysis but gameplay ...?

Ingo

I tested with 5 pieces and seems to add like 10 elo. But as you told well, 4 pieces is nothing that complicated for an engine to solve by itself in most positions. Maybe someone has tested 4 pieces and has an idea of the elo added.

[cdani]

I added an "andscacsbx.exe" to the zip file, that has maximum threads raised to 128.

[IWB]

Thanks! I will follow your test. On it the improved pondering should add something more, and also of course the 4 pieces Syzygy bases.

I added them, but I personaly doubt that 4 pc tbs add a single elo in gameplay.
They are fine in Analysis but gameplay ...?

Ingo

I tested with 5 pieces and seems to add like 10 elo. But as you told well, 4 pieces is nothing that complicated for an engine to solve by itself in most positions. Maybe someone has tested 4 pieces and has an idea of the elo added.

I remember that I forgot to add the 4pc bases to a test about 2 years ago and had to repeat the run with 4pc syz (or Nalimov, I forgot that detail) ... without any significant change.

I could swap to 5 pc tbs but I want the engine to know at least something by themself (and it is less simultanious traffic to the harddrives which is a problem when pondering with 5pc. With 4 pc this is basicaly in Windows cache after a few hits (like a Ram Disk ) and the Hd is silent)

Ingo

[cdani]

The test that is running Ingo:

http://www.inwoba.de/Andscacs088.html

[Marek Soszynski]

The test that is running Ingo:

http://www.inwoba.de/Andscacs088.html

So far, below Critter on single core.

[Ajedrecista]

Hello:

The test that is running Ingo:

http://www.inwoba.de/Andscacs088.html

The TBD (to be determined) engine is a mystery. Ingo has include TBD engines sometimes.

Spoiler alert! Here I go with my try of guess TBD's rating:

Ingo's test on Chiron 1.5 (year 2012 if I am not wrong):

Code: Select all

Chiron15

Chiron 1.5 - Houdini 3 STD (3095)               33.0    -       117.0             22.00%              Perf=2876
Chiron 1.5 - Komodo 5 (3011)                    50.5    -       99.5              33.67%              Perf=2894
Chiron 1.5 - Critter 1.4a (2990)                45.0    -       105.0             30.00%              Perf=2843
Chiron 1.5 - Stockfish 2.2.2 JA (2970)          54.0    -       96.0              36.00%              Perf=2871
Chiron 1.5 - Deep Rybka 4.1 (2965)              52.0    -       98.0              34.67%              Perf=2855
Chiron 1.5 - Naum 4.2 (2840)                    74.0    -       76.0              49.33%              Perf=2836
Chiron 1.5 - HIARCS 14 WCSC 32b (2824)          82.0    -       68.0              54.67%              Perf=2856
Chiron 1.5 - Gull 1.2 (2804)                    93.0    -       57.0              62.00%              Perf=2889
Chiron 1.5 - Deep Shredder 12 (2800)            81.0    -       69.0              54.00%              Perf=2827
Chiron 1.5 - Hannibal 1.2 (2800)                84.5    -       65.5              56.33%              Perf=2844
Chiron 1.5 - Deep Sjeng c't 2010 32b (2794)     89.5    -       60.5              59.67%              Perf=2862
Chiron 1.5 - Spike 1.4 32b (2783)               92.0    -       58.0              61.33%              Perf=2863
Chiron 1.5 - spark-1.0 (2772)                   92.5    -       57.5              61.67%              Perf=2854
Chiron 1.5 - Protector 1.4.0 (2764)             91.0    -       59.0              60.67%              Perf=2839
Chiron 1.5 - Deep Junior 13.3 (2753)            88.0    -       62.0              58.67%              Perf=2813
Chiron 1.5 - Quazar 0.4 (2740)                  103.5   -       46.5              69.00%              Perf=2878
Chiron 1.5 - Zappa Mexico II (2709)             93.5    -       56.5              62.33%              Perf=2796
Chiron 1.5 - MinkoChess 1.3 (2698)              98.0    -       52.0              65.33%              Perf=2808
                                                1397.0  -       1303.0            51.74%              Perf=2851



2700 out of 2700 games played

The average rating of the 18 opponents was 2839.6 Elo more less. So 2839.6 + 400·log10(1397/1303) ~ 2839.6 + 12.1 = 2851.7 Elo, which is very similar to 2851 Elo.

Crunching the current numbers of Andscacs 0.88 test:

Code: Select all

Andscacs088

Andscacs 0.88 - Komodo 10.1 (3246)              3.5     -       15.5              18.42%              Perf=2988
Andscacs 0.88 - Stockfish 7 (3227)              3.0     -       16.0              15.79%              Perf=2937
Andscacs 0.88 - Houdini 4 (3109)                7.0     -       12.0              36.84%              Perf=3016
Andscacs 0.88 - Gull 3 (3059)                   8.0     -       10.0              44.44%              Perf=3021
Andscacs 0.88 - Ginkgo 1.8 (3038)               7.5     -       10.5              41.67%              Perf=2980
Andscacs 0.88 - Jonny 8.00 (3025)               9.0     -       8.0               52.94%              Perf=3045
Andscacs 0.88 - Equinox 3.30 (3000)             6.0     -       12.0              33.33%              Perf=2880
Andscacs 0.88 - Critter 1.6a (2993)             9.5     -       8.5               52.78%              Perf=3012
Andscacs 0.88 - Fritz 15 (2987)                 8.5     -       10.5              44.74%              Perf=2951
Andscacs 0.88 - Fizbo 1.8 (2985)                5.0     -       13.0              27.78%              Perf=2820
Andscacs 0.88 - Hannibal 1.7 (2961)             5.5     -       12.5              30.56%              Perf=2819
Andscacs 0.88 - Protector 1.9.0 (2923)          9.0     -       8.0               52.94%              Perf=2943
Andscacs 0.88 - Chiron 3.01 (2925)              7.5     -       10.5              41.67%              Perf=2867
Andscacs 0.88 - Nirvanachess 2.3 (2919)         8.5     -       9.5               47.22%              Perf=2900
Andscacs 0.88 - Texel 1.06 (2908)               7.5     -       10.5              41.67%              Perf=2850
Andscacs 0.88 - TBD                             5.0     -       13.0              27.78%              
                                              110.0     -       180.0             37.93%       	    Perf=2942



290 out of 3520 games played 
Level: 5 Minutes/Game + 3 sec/Move

Code: Select all

TBD's individual stats (5.0 - 13.0) have been taken into account in the last line stats (110.0 - 180.0).

400·log10(110/180) ~ -85.6 Elo.
Average rating (16 opponents) = <r_16> = Perf - 400·log10(110/180) ~ 2942 - (-85.6) = 3027.6 Elo.
Average rating (15 opponents) = <r_15> = 45305/15 ~ 3020.3 Elo.
Rating (TBD) = 16·<r_16> - 15·<r_15> ~ 3135.8 Elo ~ 3136 Elo.

As I have doubts about the roundings of performance ratings, I will use some bounds in the calculation of <r_16>:

Code: Select all

 Perf      Rating (TBD)
2941.5         3128
2942           3136
2943           3152

It would be good if these ratings are more less stable with more games played. Otherwise, this method is unsound.

Am I very far of the true TBD's rating, Ingo?

Regards from Spain.

Ajedrecista.

[pferd]

Thanks for this new release. I will give it a try later today.

Do you still pursue the plan of making a Linux binary? It would be a great addition to my Linux chess tournaments.


Edit: I think the TBD engine in Ingo's test might very well be the new Shredder. At least the score corresponds to the one from Clemens tournaments earlier this year

[Dokterchen]

Hi Daniel,

I got two crashes in the first 50 games. Conditions: ponder off, 4 core, 6-men syzygy.

[Event "Liga01"]
[Date "2016.10.29"]
[Round "3"]
[White "Gull 3"]
[Black "Andscacs 0.88"]
[Result "1-0"]
[ECO "A01"]
[Opening "Nimzowitsch-Larsen Angriff"]
[Time "13:05:26"]
[Variation "Moderne Variante, 1...e5 2.Bb2 Nc6 3.e3"]
[TimeControl "180+1"]
[Termination "time forfeit"]
[PlyCount "119"]
[WhiteType "program"]
[BlackType "program"]

1. b3 e5 2. Bb2 Nc6 3. e3 Nf6 4. Bb5 Bd6 5. Na3 {FOT02} Bb4 {+0.05/20 5} 6.
Ne2 {-0.12/14 1} O-O {+0.15/22 13} 7. O-O {-0.14/16 1} d5 {+0.14/22 4} 8.
c4 {-0.23/15 1} Ne7 {-0.01/20 6} 9. Nc2 {-0.13/17 3} Bd6 {+0.08/22 6} 10.
c5 {-0.13/16 0} Bxc5 {+0.11/21 4} 11. Bxe5 {-0.13/16 0} Ng6 {+0.13/23 5}
12. Bb2 {-0.07/18 12} a6 {+0.12/21 4} 13. Bd3 {-0.17/18 5} Re8 {+0.06/21 9}
14. Ng3 {-0.15/19 7} Bd6 {+0.12/22 10} 15. Rc1 {-0.03/17 2} c5 {+0.04/19 3}
16. f4 {-0.10/17 6} b5 {+0.13/20 5} 17. Nh5 {-0.09/18 7} Be7 {+0.01/20 4}
18. Ne1 {-0.09/18 1} c4 {+0.14/20 5} 19. Bb1 {-0.07/18 3} Rb8 {+0.18/18 2}
20. Nc2 {-0.08/18 9} Bg4 {+0.16/20 2} 21. Nxf6+ {-0.13/19 1} Bxf6 {+0.24/20
2} 22. Bxf6 {-0.13/18 0} Bxd1 {+0.22/20 4} 23. Bxd8 {-0.12/20 6} Bxc2
{+0.26/20 7} 24. Bxc2 {-0.13/20 9} Rbxd8 {+0.22/19 2} 25. Kf2 {-0.08/18 0}
Ne7 {+0.25/19 4} 26. g4 {-0.05/17 1} Nc6 {+0.42/19 4} 27. h4 {-0.04/18 4}
Rd6 {+0.22/17 2} 28. h5 {-0.06/19 4} Re7 {+0.15/17 2} 29. Kf3 {0.00/18 7}
Na5 {+0.20/18 2} 30. Rfd1 {+0.08/17 4} Rd8 {+0.19/18 4} 31. bxc4 {+0.17/19
16} Nxc4 {+0.01/24 4} 32. Bb3 {+0.28/18 4} Rde8 {+0.01/22 2} 33. Rb1
{+0.27/18 4} Rb7 {-0.01/24 2} 34. g5 {+0.32/17 1} Kf8 {0.00/21 2} 35. Rbc1
{+0.27/19 4} Rbe7 {+0.01/24 3} 36. Rc2 {+0.29/20 7} Re4 {+0.01/20 2} 37.
Rdc1 {+0.24/19 8} R4e6 {-0.01/23 1} 38. Kg4 {+0.24/20 5} h6 {+0.01/21 3}
39. a4 {+0.25/19 2} hxg5 {-0.01/21 2} 40. Kxg5 {+0.25/17 0} Rd6 {-0.01/22
3} 41. axb5 {+0.22/17 4} axb5 {-0.01/24 2} 42. Ra2 {+0.22/16 0} Rc6
{-0.01/24 1} 43. Ra7 {+0.33/19 2} Rc5 {-0.01/23 2} 44. Kh4 {+0.33/17 0}
Rcc8 {-0.01/23 2} 45. Rc2 {+0.17/20 9} d4 {0.00/24 2} 46. exd4 {+0.14/18 2}
Re4 {0.00/24 1} 47. Rd7 {+0.14/18 0} Rxf4+ {0.00/25 1} 48. Kg3 {+0.14/18 0}
Rf5 {0.00/27 2} 49. Kg4 {0.00/18 3} Ke8 {0.00/25 2} 50. Kxf5 {0.00/21 2}
Kxd7 {0.00/25 0} 51. Ra2 {0.00/23 2} Nd6+ {+0.01/24 1} 52. Ke5 {+0.02/20 2}
f6+ {0.00/24 2} 53. Kd5 {+0.03/23 3} Nf5 {0.00/25 3} 54. Ra7+ {+0.03/22 0}
Rc7 {0.00/28 1} 55. Rxc7+ {+0.03/22 0} Kxc7 {0.00/29 0} 56. Bd1 {+0.03/22
0} b4 {0.00/27 1} 57. Kc5 {+0.14/21 0} Ng3 {0.00/27 1} 58. Kxb4 {+0.12/23
1} Kd6 {-0.01/28 1} 59. Bf3 {+0.03/25 6} f5 {-0.01/32 2} 60. Kc4 {+0.03/24
0} ... {Schwarz überschreitet die Zeit} 1-0

[Event "Liga01"]
[Date "2016.10.29"]
[Round "5"]
[White "Komodo 9.4.2"]
[Black "Andscacs 0.88"]
[Result "1-0"]
[ECO "A03"]
[Opening "Bird Eröffnung"]
[Time "14:50:22"]
[Variation "1...d5 2.Nf3 Nf6 3.g3 g6: 6.d3 c5"]
[TimeControl "180+1"]
[Termination "time forfeit"]
[PlyCount "67"]
[WhiteType "program"]
[BlackType "program"]

1. f4 d5 2. Nf3 Nf6 3. g3 g6 4. Bg2 Bg7 5. O-O O-O 6. d3 {FOT03} c5
{+0.16/19 9} 7. Nc3 {-0.07/21 16} d4 {+0.07/22 14} 8. Na4 {+0.02/22 2} Na6
{+0.16/20 5} 9. c4 {0.00/21 3} Rb8 {+0.12/21 12} 10. Rb1 {+0.06/23 11} Qd6
{+0.22/18 4} 11. Bd2 {0.00/23 6} Bf5 {+0.20/21 11} 12. b3 {+0.08/19 1} Ng4
{+0.09/20 10} 13. h3 {+0.27/21 4} Nf6 {-0.12/21 9} 14. Re1 {+0.36/22 29}
Rfd8 {+0.06/19 3} 15. a3 {+0.44/20 4} Bd7 {-0.02/19 2} 16. b4 {+0.39/22 4}
Nh5 {+0.21/18 4} 17. Kh2 {+0.37/20 2} b6 {+0.05/19 5} 18. e3 {+0.22/21 3}
dxe3 {+0.19/19 2} 19. Rxe3 {+0.31/20 2} Bh6 {+0.01/17 2} 20. Re1 {+0.35/21
5} Qxd3 {+0.40/19 7} 21. Nc3 {+0.41/20 3} Bf5 {+0.12/17 4} 22. Rc1
{+0.68/18 3} Be6 {+0.01/17 2} 23. Qa4 {+1.02/20 3} cxb4 {+0.01/16 4} 24.
axb4 {+0.81/22 2} Qxc4 {-0.44/19 3} 25. Re4 {+1.21/24 13} Qd3 {-0.65/22 4}
26. Be1 {+1.40/24 7} Nc7 {-0.53/21 4} 27. Ne5 {+1.40/24 3} Qa6 {-0.62/22 4}
28. Qxa6 {+1.41/22 1} Nxa6 {-0.50/23 4} 29. Nc6 {+1.44/23 4} Bf5 {-0.72/21
3} 30. Nxb8 {+1.52/23 4} Bxe4 {-0.69/20 2} 31. Nxa6 {+1.56/25 2} Bxg2
{-0.87/23 1} 32. Kxg2 {+1.49/25 1} e5 {-1.16/22 2} 33. Rd1 {+1.50/26 2}
Rxd1 {-1.29/24 3} 34. Nxd1 {+1.49/25 3} ... {Schwarz überschreitet die
Zeit} 1-0

KR
Torsten

[cdani]

Hi Daniel,

I got two crashes in the first 50 games. Conditions: ponder off, 4 core, 6-men syzygy.

Hi. Thanks for saying. I'm trying to reproduce the crash. Can you send me the files that end in .dmp and should be where the engine is? With this maybe I can find something more or faster: cdani at yahoo.com

[cdani]

After a lot of tries, I give up for the moment trying to make work well Syzygy. I changed the offcial version so it does not have it:
http://andscacs.com/andscacs088.zip
It appears as 0.88 r1.

If anyone is interested on the old one with Syzygy, you can grab it here:
http://andscacs.com/andscacs088s.zip

The Syzygy code seems to crash even on correct positions. I suppose is related to the changes I had to do to make it work on Visual Studio, but as the code of Syzygy is very complicated, I have no idea how to solve the problems. Well, maybe in the next version...