Cheng: A great lost to the community

[Frank Quisinsky]

Hi Michael,

nice to read that and if you are working "if you have time for it" on Nemo it's perfect because Nemo is also an interesting engine!!

I test a bit in SWCR2 with ponder (my actual rating list is without ponder). With ponder the strength in my test is clearly weaker as I can see in CEGT. So I am thinking Nemo have a "Ponder Problem". I am not searching in detail and delete my test-run at this time.

You can be sure that I will test your engine in FCT1 if a newer version is available. After replaying CEGT games I saw that Nemo is very interesting and also with an own face (playing style).

I hope that we lost in 2015 not many of such great engines. We lost Sjeng, others and that are bad news for me. We won x clones and lost completly own developments in the years after Igorit.

With Senpai, Smarthink, DisasterArea & others different programmers come back. That are good news. Furthermore, in 2014 a lot of engines comes with good improvements (ICE, Nirvanachess, Deuterium, Gaviota, Vajolet, Andscacs and many others).

I think after all the available clones since IPP that computer chess is in 2014 on a good way. Much programmers find after disappointments to clone topics again much fun.

2014 are a good year after 2010-2013 ...
2015 will be better ... I am sure!

Keep up your work in Nemo development!
Our hobby need each talented programmer at this times!

Best
Frank

Next good news comes by yourself!
Thanks ... I will wait of the next Nemo!

[velmarin]

Hi Michael,

nice to read that and if you are working "if you have time for it" on Nemo it's perfect because Nemo is also an interesting engine!!

I test a bit in SWCR2 with ponder (my actual rating list is without ponder). With ponder the strength in my test is clearly weaker as I can see in CEGT. So I am thinking Nemo have a "Ponder Problem". I am not searching in detail and delete my test-run at this time.

You can be sure that I will test your engine in FCT1 if a newer version is available. After replaying CEGT games I saw that Nemo is very interesting and also with an own face (playing style).

I hope that we lost in 2015 not many of such great engines. We lost Sjeng, others and that are bad news for me. We won x clones and lost completly own developments in the years after Igorit.

With Senpai, Smarthink, DisasterArea & others different programmers come back. That are good news. Furthermore, in 2014 a lot of engines comes with good improvements (ICE, Nirvanachess, Deuterium, Gaviota, Vajolet, Andscacs and many others).

I think after all the available clones since IPP that computer chess is in 2014 on a good way. Much programmers find after disappointments to clone topics again much fun.

2014 are a good year after 2010-2013 ...
2015 will be better ... I am sure!

Keep up your work in Nemo development!
Our hobby need each talented programmer at this times!

Best
Frank

Next good news comes by yourself!
Thanks ... I will wait of the next Nemo!

What has this to do with clones and Ippolit
That boredom of speech.

[Frank Quisinsky]

A lot!

[velmarin]

A lot!

Cheng` Martin is a big job and very original.
A great work of a very detailed computer programmer, and lover of the art of programming.
I suppose studying chess is another thing, there is much to do in response to Github Cheng.
Precisely for that Martin does not follow the current copy of ideas Ippolit.
VERY, VERI original.

But you talk much, but your list is full of clones, clones of clones, only you do not see.

[Frank Quisinsky]

Full of clones?

Here is my list!
Now which one is a clone?

Please with proofs ... I can see nothing after all information I have!

Code: Select all

 "FCT1 Rating List"
 Will be updated after ~ 1000 new games!
 ---------------------------------------
  
 Date         : January 15th, 2014 (02:00)
 FCT1         : 49700 games
 Version      : 1.16.500.s2
 Hardware     : 2x Intel Core i7-4770K *4.3Ghz, 45-minutes games
 Engines      : Maximum allowed = three versions of same engine!
 
 Calculation  : Ordo 0.981, "Hiarcs 14 WCSC w32" with 2.825 Elo / News 057 & News 062
 Parameter    : Ordo_0981.exe -a 2825 -A "Hiarcs 14 WCSC w32" -p 1.pgn -o rating.txt -W -s1000 -D -E -V
  
 
      Program                          Elo    +    -   Games   Score  Av.Op.  Draw
  01. Komodo 8 x64                     3108   16   16  1550    80.8%   2827   31.2%
  02. Stockfish 02.10.14 BMI2 x64 C    3105   19   19  1150    79.5%   2844   34.6%
  --. Stockfish 03.08.14 BMI2 x64      3104   16   16  1600    82.7%   2804   30.2%
  --. Komodo 7a x64                    3078   19   19  1250    79.6%   2816   31.9%
  --. Stockfish 5 SSE42 x64            3077   20   20  1000    79.6%   2818   35.4%
  03. Fire 4 x64                       3052   16   16  1650    79.3%   2791   30.1%  NEW, +102 Elo, runs as S1 (Special -01-)
  --. Komodo TCECr x64                 3048   20   20  1000    80.3%   2786   32.2%
  04. GullChess 3.0 BMI2 x64           3035   14   14  1850    73.6%   2832   37.9%
  05. Equinox 3.30 x64                 2988   16   16  1600    72.8%   2798   40.1%  NEW, runs as S2 (Special -02-)
  --. GullChess 2.8 Beta BMI2 x64      2985   18   18  1000    74.0%   2790   37.7%
  --. Fire 3.0 AVX x64                 2950   13   13  2050    65.5%   2825   41.2%
  06. Protector 1.7.0 x64              2905   15   15  1200    55.7%   2863   44.3%
  ---------------------------------------------------------------------------------
  Main-Test (engines 2900-2650 Elo)
  ---------------------------------------------------------------------------------
  07. Chiron 2.0 x64                   2892   11   11  2650    59.7%   2818   39.8%
  08. Critter 0.90 SSE4 x64            2887   14   14  1400    55.8%   2843   43.1%
  --. Protector 1.6.0 x64              2881   14   14  1600    58.7%   2816   44.4%
  09. Hannibal 1.4b x64                2862   11   11  2650    55.9%   2819   42.3%
  10. Texel 1.04 x64                   2849   11   11  2350    54.6%   2817   40.4%
  --. Protector 1.5.0 JA x64           2845   16   16  1000    56.5%   2797   45.7%
  11. Nirvanachess 2.0a x64            2836   24   24   500    62.1%   2745   43.8%  NEW, +145 Elo
  12. Senpai 1.0 SSE42 x64             2830   11   11  2650    51.9%   2820   39.8%
  13. Hiarcs 14 WCSC w32               2825   11   11  2650    51.2%   2820   42.5%
  14. Shredder 12 x64                  2792   12   12  2150    44.9%   2838   40.0%
  --. Texel 1.03 x64                   2787   16   16  1000    48.7%   2799   42.8%
  15. Junior 12.5.03 x64               2779   15   15  1400    42.2%   2846   38.4%
  --. Junior 13.3.00 x64               2778   13   13  1500    45.9%   2813   40.9%
  --. Junior 13.8.04 Yokohama x64      2775   13   13  1650    43.4%   2833   40.8%
  16. Spike 1.4 Leiden w32             2771   13   13  1600    41.6%   2843   40.3%
  17. DiscoCheck 5.2.1 x64             2762   15   15  1250    39.2%   2853   36.2%
  18. iCE 2.0 v2240 POP x64            2754   10   10  2850    49.5%   2763   43.2%
  19. Quazar 0.4 x64                   2753   12   12  2150    40.0%   2839   41.1%
  20. SmarThink 1.70 SSE3 x64          2752   12   12  2200    42.1%   2820   36.5%
  21. Spark 1.0 x64                    2750    9    9  3650    47.6%   2774   40.1%
  22. Deuterium 14.3.34.130 POP x64    2748   22   22   600    44.6%   2794   43.2%  NEW, + 29 Elo
  23. Zappa Mexico II x64              2744   13   13  1700    37.6%   2849   40.6%
  --. SmarThink 1.60 x64               2740   17   17  1100    38.0%   2841   36.2%
  24. Fizbo 1.3.1 x64                  2737   21   21   600    43.2%   2795   39.7%  NEW, + 57 Elo
  25. Vajolet2 1.45 POP x64            2725    9    9  3150    45.3%   2768   41.5%
  --. Vajolet2 1.28 POP x64            2722   15   15  1450    36.4%   2837   38.7%
  --. Deuterium 14.2.33.276 x64        2719   10   10  3050    44.3%   2771   39.9%
  26. Gaviota 1.0 AVX x64              2719    9    9  3650    43.7%   2775   36.6%
  27. Tornado 5.0 SSE4 x64             2714   12   12  2250    43.4%   2773   37.2%
  28. Andscacs 0.70 POP x64            2700   22   22   600    38.3%   2796   41.7%  NEW, +105 Elo
  --. SmarThink 1.50 SSE3 x64          2700   17   17  1000    36.9%   2804   37.1%
  --. Nirvanachess 1.7 x64             2691   11   11  2500    41.4%   2766   38.4%
  --. Fizbo 1.2 x64                    2680   12   12  2100    41.8%   2749   39.0%
  29. Arasan 17.4 POP x64              2677   13   13  1600    46.7%   2706   41.4%
  --. Nirvanachess 1.6 x64             2673   15   15  1350    29.6%   2847   33.6%
  30. Cheng4 0.36c x64                 2668   13   13  1600    45.4%   2706   41.5%
  31. EXchess 7.51b x64                2663   23   23   500    38.0%   2753   37.6%  NEW, + 21 Elo
  32. Rodent 1.6 Build 6 POP x64       2662   22   22   600    33.5%   2798   36.0%  NEW, + 33 Elo
  33. DisasterArea 1.54 x64            2661   23   23   500    37.7%   2754   43.0%  NEW
  34. Crafty 24.1 SSE42 x64            2643   13   13  1600    42.2%   2707   35.8%
  --. EXchess 7.31b x64                2642   16   16  1100    45.0%   2686   39.0%
  ---------------------------------------------------------------------------------
  35. Glaurung 2.2 JA x64              2633   17   17  1000    47.1%   2653   40.9%
  --. Crafty 24.0 SSE42 x64            2632   18   18  1000    28.4%   2807   34.4%
  --. Rodent 1.4 POP Build 2 x64       2629   17   17  1000    46.6%   2653   40.4%
  36. Atlas 3.70em x64                 2627   17   17  1050    44.6%   2672   38.7%
  37. OctoChess r5190 SSE4 x64         2623   16   16  1000    45.9%   2654   45.5%
  38. Rhetoric 1.4.1 x64               2598   13   13  2000    31.9%   2752   35.8%
  --. Andscacs 0.64 POP x64            2595   17   17  1000    41.9%   2655   39.8%
  39. Godel 3.4.9 x64                  2584   17   17  1000    40.4%   2656   39.3%
  40. Djinn 1.021 POP x64              2515   18   18  1000    31.2%   2659   36.6%
  41. ProDeo 1.87 w32                  2508   17   17  1000    30.4%   2659   31.3%
  
	  
 Move average                : 174.81 / 87.40
 White advantage             : 40.05
 Draw rate (equal opponents) : 46.44%
 
 White Wins   :  17585 (35.4%)
 Black Wins   :  12733 (25.6%)
 Draws        :  19382 (39.0%)
 Unfinished   :      0

 White Perf.  :   54.9%
 Black Perf.  :   45.1%

 ECO A        =   9064 Games (18.2%)
 ECO B        =  11637 Games (23.4%)
 ECO C        =  10545 Games (21.2%)
 ECO D        =   9716 Games (19.5%)
 ECO E        =   8738 Games (17.6%)

[Adam Hair]

Full of clones?

Here is my list!
Now which one is a clone?

Probably Gaviota. I do not trust the author. He is known to use the alias "Mike Loan".
Mike Loan -> Mi Kloan -> My Clone

[mar]

Probably Gaviota. I do not trust the author. He is known to use the alias "Mike Loan".
Mike Loan -> Mi Kloan -> My Clone

Hmm

[velmarin]

Hi Martin, from the code above Cheng,
  engine has come to have a PSQ very different than the usual static evaluation.
I wonder if more measured not be better for the engine.

[mar]

Hi Martin, from the code above Cheng,
  engine has come to have a PSQ very different than the usual static evaluation.
I wonder if more measured not be better for the engine.

You mean the version on GitHub? I've already updated the sources.
I've already sent the package to Emil so I guess it's a matter of 1-2 days, depending on how often he reads e-mails
Yes the values seem a bit odd but I tuned the whole eval using Peter's method (solely using selfplay) and it seems to work.
psq can also offset material values (but so can mobility), so I guess there is no universal "right" range for psq.
And I dare to say psq don't have to be symmetric either.
In the case of cheng it's possible that some missing knowledge was simply baked into psqs because cheng is a dumb beancounter.
I'm pretty sure that if you take psq from engine A and put it in engine B it won't work (unless you move whole eval).

[velmarin]

I must confess that I did the test and put the PSQ Cheng 3
Cheng 4 clearly won the duel.

If I imagine that any change upsets the balance.
The engine Github is better than Cheng 0.38.
Congrats.