FGRL - 16 Cores rating list - Lc0 0.25.1 63305 (Kayra4)

[fastgm]

New: 13. Lc0 0.25.1 63305 (Kayra4)

Code: Select all

   # Engine                            :    Elo  Error   Games    Points    (%)      W      D      L     D(%)   CFS
 -------------------------------------------------------------------------------------------------------------------
   1 Lc0 0.24.1 LS 14.2                :   3472     10    3000    2313.5   77.12   1721   1185     94   39.50    94
   2 Lc0 0.24.0 LS 14                  :   3461      9    3500    2677.0   76.49   1964   1426    110   40.74    95
   3 Lc0k 0.23.2 256x20-T40-1541       :   3451      8    4000    2979.0   74.47   2130   1698    172   42.45    57
   4 Lc0 0.24.1 LS 13                  :   3449     10    2750    2032.5   73.91   1431   1203    116   43.75    89
   5 Stockfish 200502 C0               :   3441      8    2500    1301.5   52.06    383   1837    280   73.48    96
   6 Fat Fritz 1.1                     :   3431      9    3750    2772.5   73.93   1998   1549    203   41.31    59
   7 Lc0 0.24.1 384x30-T60-3010        :   3430      9    3250    2408.5   74.11   1703   1411    136   43.42    67
   8 Lc0 0.22.0 T40B.4-160             :   3427      8    4250    3027.0   71.22   2062   1930    258   45.41    50
   9 Stockfish 11 C0                   :   3427      6    4250    2173.5   51.14    589   3169    492   74.56    73
  10 Lc0 0.22.0 42850                  :   3424      8    4250    3013.0   70.89   2057   1912    281   44.99    71
  11 Stockfish 11                      :   3421      7    4000    2157.5   53.94    777   2761    462   69.03    75
  12 Lc0 0.25.1 63305                  :   3417     10    2500    1823.5   72.94   1246   1155     99   46.20    57
  13 Lc0 0.25.1 63305 (Kayra4)         :   3415     10    2500    1820.0   72.80   1239   1162     99   46.48    90
  14 Lc0k 0.23.2 62172                 :   3406     10    2500    1827.0   73.08   1256   1142    102   45.68    52
  15 Lc0 LS-LS 12.2                    :   3405      8    3750    2680.5   71.48   1809   1743    198   46.48    64
  16 Lc0 0.22.0 J13B.2-188             :   3403      8    4250    2918.5   68.67   1884   2069    297   48.68    79
  17 Lc0 LS-LS 12.1                    :   3398      9    3000    2056.0   68.53   1299   1514    187   50.47    53
  18 Lc0k 0.23.2 384x30-T40-2036       :   3398      7    4000    2761.0   69.03   1859   1804    337   45.10    99
  19 Fat Fritz 1.0                     :   3385      8    3000    2012.0   67.07   1311   1402    287   46.73    85
  20 Lc0 0.22.0 32930                  :   3379      7    4250    2804.5   65.99   1764   2081    405   48.96    66
  21 Lc0 0.22.0 LS 12 preview          :   3377      9    2750    1876.0   68.22   1258   1236    256   44.95    56
  22 Stockfish 10                      :   3376      8    2750    1337.5   48.64    428   1819    503   66.15    71
  23 Lc0 0.23.2 61950                  :   3373      8    3750    2482.0   66.19   1550   1864    336   49.71    63
  24 Lc0 0.23.2 58613+                 :   3371      9    2500    1750.0   70.00   1200   1100    200   44.00    71
  25 Lc0 0.22.0 384x30-T40-1573        :   3367      9    2750    1846.0   67.13   1220   1252    278   45.53   100
  26 Lc0 0.22.0 11260                  :   3339      9    2750    1755.5   63.84   1044   1423    283   51.75    98
  27 Houdini 6.03                      :   3327      6    5750    2291.5   39.85    566   3451   1733   60.02    77
  28 Komodo 13.3                       :   3324      7    3750    1379.0   36.77    294   2170   1286   57.87   100
  29 Komodo 13.2.5                     :   3311      6    4500    1749.5   38.88    503   2493   1504   55.40   100
  30 Scorpio 3.07 NN-ens-net-20x256    :   3280      9    2500    1466.5   58.66    855   1223    422   48.92   100
  31 Ethereal 12.00                    :   3226      8    3750     947.5   25.27    167   1561   2022   41.63   100
  32 Xiphos 0.6                        :   3211      6    5750    1488.0   25.88    232   2512   3006   43.69    71
  33 Ethereal 11.75                    :   3208      7    4500    1196.0   26.58    210   1972   2318   43.82    94
  34 Fire 7.1                          :   3200      7    5750    1422.5   24.74    271   2303   3176   40.05   100
  35 RofChade 2.3                      :   3165     10    3250     627.0   19.29     81   1092   2077   33.60   100
  36 Booot 6.4                         :   3138     10    3750     644.0   17.17     73   1142   2535   30.45    72
  37 Laser 1.7                         :   3135      7    5750    1071.5   18.63    172   1799   3779   31.29    96
  38 Andscacs 0.95                     :   3125      8    5750    1026.5   17.85    178   1697   3875   29.51   100
  39 Deep Shredder 13                  :   3103      9    5000     818.0   16.36    114   1408   3478   28.16    83
  40 Booot 6.3.1                       :   3096      9    4750     767.0   16.15    118   1298   3334   27.33   ---

White advantage = 59.92 +/- 0.92
Draw rate (equal opponents) = 65.73 % +/- 0.25

Games        : 75.500 (finished)

White Wins   : 25.358 (33.6 %)
Black Wins   : 15.658 (20.7 %)
Draws        : 34.484 (45.7 %)

White Score  : 56.4 %
Black Score  : 43.6 %

http://www.fastgm.de

[pohl4711]

Hmmm. Disappointing. No progress.

[mehmet123]

Thanks Andreas for the testing of Kayra4.

I don' think the results of Kayra4 is worse. Test results are compatible with my results of Lc0 (Kayra4/Default) vs Stockfish. I had only put Kayra vs. Stockfish (except Lc0 Default) matches at my topics:

Look at this:
Lc0 0.25.1 63305(Default) vs Stockfish 11 C0 : 250 ( 16, 199, 35), % 46.2
Lc0 0.25.1 63305 (Kayra4) vs Stockfish 11 C0 : 250 ( 28, 188, 34), % 48.8

Lc0 (Kayra4) is + 18 elo better than Lc0 (Default) against Stockfish. (Elostat)

http://www.fastgm.de/h2h16-60.html

My last test shows an improvement of 18 elo:

Program Elo + - Games Score Av.Op. Draws
1 Lc0 63500 (Kayra4] : 2410 20 20 300 51.7 % 2399 73.3 %
2 Stockfish 110520 x64 : 2399 14 14 600 49.7 % 2401 74.0 %
3 Lc0 63500 (Default) : 2392 20 20 300 49.0 % 2399 74.7 %

Arena Gui, 1 min + 0.5 sec tc, Lc0 v.0.25.1, Balsa Opening Book (5 moves)
Stockfish: Core i7- 9750h (1 core), Contempt 0, Lc0:Nvdia Gtx 1650
Hash:512 Mb , No Tablebase, Ponder:Off

http://talkchess.com/forum3/viewtopic.php?f=6&t=73913

Lc0 Kayra4 had developed to get better results against Stockfish. The result of Kayra3 against weaker engines could be better than Kayra4, but it's not as good as Kayra4 against Stockfish. It's easy getting better results against weaker engines by changing some Kayra4 settings.
But this isn' t my goal. I think the main rival of Lc0 is Stockfish.

[pohl4711]

OK, I am curious, if a "reduced" version of Kayra4 setting will give the strongest lc0 (0.24.1 with LS 14.3 net) a Elo-boost vs. Stockfish in my NN-testings.

https://www.sp-cc.de/nn-vs-sf-testing.htm

Because lc0 0.24.1 does not support MLH (and 0.25.1 does not run with LS 14.3 net), the MLH-parameters can not be used:
MovesLeftMaxEffect=0.24
MovesLeftTreshhold=0.48
MovesLeftSlope=0.00032

And, because I am using my NBSC-Armageddon openings, any Contempt-setting (=drawscore) makes no sense at all. So, the drawscore-settings will not be used:
DrawScoreSideToMove=5
DrawScoreOpponent=5
DrawScoreWhite=12
DrawScoreBlack=-6

So, what remains of Kayra4 setting is:
CPuct=2.2
CPuctAtRoot=2.2
CPuctBase=16800
CPuctBaseAtRoot=16800
CPuctFactor=1.69
CPuctFactorAtRoot=1.69
FpuValue=0.36
PolicyTemperature=1.45
MaxConcurrentSearchers=2
SmartPruningFactor=1.55

I call it Kayra4(xp) and will try it with Lc0 0.24.1 LS 14.3 very soon...