Ratings from all 169.033 FCP Tourney games ...

[Frank Quisinsky]

Please have a look in differents between Komodo 14 (without NN) and the last Dragon 3 NN (Komodo) ...
Or Stockfish 11 (without NN) with the last Version 15 NN ...

With longer time controls and many opponents an other view as "Blitz".

Komodo team are speanking from 300 Elo differents (with and without NN).
In reality it is lesser as 200 Elo with longer time controls, many opponents and an equal opening book like FEOBOS.

At the moment I made some test with the time control 40 moves in 40 minutes on other systems.
And again, SF or Dragon lost 10-15 Elo to the others.
Maybe later I will public the results with more games.

Code: Select all

   # Player                      :      Elo  Games  Score%   won  draw  lost  Points  Draw%  Error   OppAvg   OppE   OppD
   1 Stockfish 311221 NN dev     :  3485.92   1740    80.7  1072   666     2  1405.0   38.3  14.15  3206.00  10.28   58.0
   2 Stockfish 15 NN             :  3482.44    690    77.4   379   310     1   534.0   44.9  21.35  3243.39  11.83   42.0
   3 Lc0 0.28.2 611062 GPU       :  3474.34   1200    80.5   733   466     1   966.0   38.8  16.58  3197.11   9.50   40.0
   4 Dragon 3 NN (Komodo)        :  3473.66   1380    79.3   815   559     6  1094.5   40.5  14.94  3213.37  11.68   46.0
   5 Stockfish 151121 NN dev     :  3466.36   1200    81.9   767   431     2   982.5   35.9  17.83  3174.51  10.14   40.0
   6 Stockfish 110121 NN dev     :  3462.64   2000    88.0  1519   480     1  1759.0   24.0  16.37  3081.85   8.59   40.0
   7 ShashChess 20.2 NN          :  3460.35   1200    78.1   676   523     1   937.5   43.6  15.95  3213.69   9.81   40.0
   8 Dragon 2.6 NN (Komodo)      :  3459.79   2055    77.6  1149   893    13  1595.5   43.5  12.36  3216.09  11.13   73.9
   9 ShashChess 21.1 NN          :  3458.22    690    75.0   347   341     2   517.5   49.4  20.41  3243.92  11.85   42.0
  10 Dragon 2.5.1 NN (Komodo)    :  3450.83   1200    80.6   741   452     7   967.0   37.7  17.17  3174.90  10.16   40.0
  11 Dragon 1 NN (Komodo)        :  3416.64   2000    85.0  1404   590     6  1699.0   29.5  14.86  3083.00   8.63   40.0
  12 Fire 8 MC.3 NNSf            :  3381.04   2565    69.1  1072  1403    90  1773.5   54.7   9.96  3220.85  11.61   90.9
  13 Koivisto 8.6 NN             :  3377.40    690    66.2   255   404    31   457.0   58.6  17.78  3245.67  11.90   42.0
  14 SlowChess Blitz 2.83 NN     :  3370.97   1575    66.8   599   906    70  1052.0   57.5  12.37  3234.38  11.18   58.0
  15 Koivisto 7.13 NN            :  3366.20   1200    68.4   502   638    60   821.0   53.2  14.07  3216.05   9.86   40.0
  16 Berserk 8.5.1 NN            :  3366.04   1755    67.8   698   985    72  1190.5   56.1  11.71  3219.76  10.87   63.9
  17 Stockfish 11                :  3365.72   2000    83.7  1353   643     4  1674.5   32.1  14.27  3050.25   9.09   40.0
  18 rofChade 2.321 NN dev       :  3360.56    690    64.3   247   393    50   443.5   57.0  18.11  3246.04  11.90   42.0
  19 Revenge 2.0 NN              :  3359.88   2085    66.9   805  1181    99  1395.5   56.6  10.50  3221.23  11.21   74.9
  20 Berserk 9 NN dev            :  3357.80   1200    67.5   481   657    62   809.5   54.8  13.65  3216.26   9.87   40.0
  21 SlowChess Blitz 2.8 NN      :  3356.75   1710    68.1   691   946    73  1164.0   55.3  11.91  3206.65  10.65   57.0
  22 Andscacs 0.1 NNSf dev       :  3353.76   1200    76.3   646   538    16   915.0   44.8  15.03  3132.57   9.85   40.0
  23 Koivisto 7.5 NN             :  3346.85   1200    70.5   545   601    54   845.5   50.1  14.60  3177.50  10.22   40.0
  24 Seer 2.5.0 NN               :  3341.59   1500    65.7   582   807   111   985.5   53.8  12.14  3214.69  11.79   50.0
  25 Ethereal 13.25 NN           :  3341.36   1710    66.3   666   937   107  1134.5   54.8  11.68  3206.92  10.65   57.0
  26 rofChade 2.317 NN dev       :  3340.98   1200    65.5   451   670    79   786.0   55.8  13.47  3216.68   9.87   40.0
  27 RubiChess 20220223 NN       :  3340.17    690    61.8   219   415    56   426.5   60.1  17.25  3246.48  11.92   42.0
  28 Koivisto 7.9 NN             :  3335.38   1230    66.5   475   685    70   817.5   55.7  14.19  3201.76   9.56   41.0
  29 Berserk 7 NN                :  3334.39   1200    69.1   508   642    50   829.0   53.5  13.99  3177.81  10.24   40.0
  30 RubiChess 2021 NN           :  3318.91   1740    63.7   615   988   137  1109.0   56.8  11.37  3208.88  10.32   58.0
  31 RubiChess 2.2 NN            :  3299.93   1200    65.1   441   681    78   781.5   56.8  13.86  3178.67  10.24   40.0
  32 Komodo 14.0                 :  3294.66   2000    77.7  1150   809    41  1554.5   40.5  12.89  3052.02   9.13   40.0
  33 Arasan 23.3 NN              :  3288.78    690    55.4   167   430    93   382.0   62.3  17.15  3247.60  11.92   42.0
  34 rofChade 2.313 NN dev       :  3287.44   1230    60.8   410   675   145   747.5   54.9  13.02  3202.93   9.59   41.0
  35 Houdini 6.03 Pro            :  3285.12   2000    76.8  1104   865    31  1536.5   43.3  11.70  3052.26   9.16   40.0
  36 Revenge 1.0 NN              :  3281.44   1200    62.9   404   702    94   755.0   58.5  13.64  3179.13  10.25   40.0
  37 Seer 2.4.0 NN               :  3281.09   2505    61.1   830  1400   275  1530.0   55.9   9.04  3192.47  11.34   88.9
  38 Igel 3.0.10 NN              :  3270.74   3075    58.7   872  1863   340  1803.5   60.6   8.28  3201.51  11.70  107.8
  39 Seer 2.4.0 NN dev           :  3270.47   1200    61.6   397   684   119   739.0   57.0  13.36  3179.41  10.25   40.0
  40 SlowChess Blitz 2.5 NN      :  3269.05   2000    71.9  1014   849   137  1438.5   42.5  11.21  3086.69   8.72   40.0
  41 Ethereal 12.75              :  3266.22   2000    71.6   969   927   104  1432.5   46.4  11.72  3086.76   8.71   40.0
  42 Ethereal 13.07              :  3265.41   1725    59.9   538   992   195  1034.0   57.5  11.00  3186.40  11.17   62.9
  43 Caissa 0.5 NNSf             :  3264.34    690    52.2   132   457   101   360.5   66.2  17.18  3248.13  11.92   42.0
  44 Fire 8.2                    :  3260.65   1200    65.9   449   683    68   790.5   56.9  13.06  3134.90   9.90   40.0
  45 Halogen 10.23 NN dev        :  3254.84   1200    65.2   448   668    84   782.0   55.7  13.37  3135.04   9.89   40.0
  46 Nemorino 6.09 NN dev        :  3252.05   3075    56.4   848  1771   456  1733.5   57.6   7.83  3201.69  11.71  107.8
  47 rofChade 2.310 NN dev       :  3251.93   1200    59.3   370   683   147   711.5   56.9  13.11  3179.87  10.26   40.0
  48 Lc0 0.28.2 752187 CPU       :  3248.96   2055    54.0   457  1305   293  1109.5   63.5   9.64  3219.15  11.14   73.9
  49 Clover 3.1 NN               :  3248.07    690    50.1   138   416   136   346.0   60.3  17.64  3248.49  11.91   42.0
  50 Arasan 23.2 NN              :  3246.13   2280    57.7   644  1344   292  1316.0   58.9   9.30  3185.72  10.75   76.0
  51 Rodent 1.0 NNSf             :  3245.20   1740    54.7   396  1111   233   951.5   63.9  10.10  3210.15  10.35   58.0
  52 Coiled 1.1 NNSf             :  3244.52   1200    53.5   267   750   183   642.0   62.5  12.72  3219.09   9.89   40.0
  53 Halogen 10.23.11 NN dev     :  3239.60    690    49.1   124   429   137   338.5   62.2  16.51  3248.67  11.93   42.0
  54 Tucano 10.00 NN             :  3232.82   2085    53.9   482  1283   320  1123.5   61.5   9.82  3202.95  11.44   74.9
  55 Booot 7.0 NN dev            :  3231.11   2085    53.6   490  1257   338  1118.5   60.3   9.82  3202.97  11.44   74.9
  56 Wasp 5.53 NN dev            :  3229.66   1410    51.0   293   853   264   719.5   60.5  11.89  3223.80  11.76   47.0
  57 Minic 3.18 NN               :  3229.11   2085    53.4   489  1250   346  1114.0   60.0  10.15  3203.00  11.44   74.9
  58 Ethereal 12.25              :  3225.05   2000    70.7   909  1009    82  1413.5   50.5  11.30  3053.76   9.17   40.0
  59 Arasan 23.0.1 NN            :  3224.51   1200    55.8   309   722   169   670.0   60.2  13.09  3180.56  10.26   40.0
  60 Lc0 0.28.0 744204 CPU       :  3208.87   1200    53.8   311   670   219   646.0   55.8  13.15  3180.95  10.26   40.0
  61 Wasp 5.50 NN                :  3207.10    630    46.0   101   378   151   290.0   60.0  17.28  3239.63  11.77   40.6
  62 Nemorino 6.04 NN dev        :  3206.42   2000    64.9   808   980   212  1298.0   49.0  10.68  3088.25   8.73   40.0
  63 Wasp 5.30 NN dev            :  3202.11   1200    58.5   356   692   152   702.0   57.7  12.76  3136.36   9.90   40.0
  64 Igel 2.9.0 NN               :  3200.10   2000    64.2   767  1032   201  1283.0   51.6  10.42  3088.41   8.74   40.0
  65 Booot 6.5                   :  3197.03   1710    48.7   343   981   386   833.5   57.4  10.87  3209.45  10.67   57.0
  66 SlowChess Blitz 2.2         :  3194.08   2000    67.2   843  1002   155  1344.0   50.1  10.64  3054.54   9.19   40.0
  67 RubiChess 1.9 NN            :  3193.62   2000    63.4   764  1007   229  1267.5   50.4  10.79  3088.57   8.73   40.0
  68 Fritz 18 NN (Ginkgo)        :  3192.03    690    43.0   103   387   200   296.5   56.1  17.13  3249.70  11.92   42.0
  69 Fire 7.1                    :  3191.91   2000    67.0   818  1042   140  1339.0   52.1  10.64  3054.59   9.19   40.0
  70 Wasp 5.26 NN dev            :  3186.97   1200    46.0   216   673   311   552.5   56.1  12.76  3220.53   9.89   40.0
  71 Toga IV 1.1 NNRe            :  3185.87    690    42.2    94   394   202   291.0   57.1  16.75  3249.84  11.93   42.0
  72 Xiphos 0.6                  :  3175.76   7075    56.1  2013  3910  1152  3968.0   55.3   5.48  3128.70  10.24  139.4
  73 Pedone 3.0 NN               :  3175.53   2000    61.2   705  1037   258  1223.5   51.9  10.25  3089.03   8.74   40.0
  74 Wasp 5.20 NN                :  3166.80   1230    45.5   208   703   319   559.5   57.2  12.24  3205.87   9.61   41.0
  75 Booot 6.4                   :  3166.63   4000    62.0  1375  2212   413  2481.0   55.3   7.32  3072.24   9.04   63.9
  76 Velvet 3.3.0 NN             :  3166.19    690    39.7    84   380   226   274.0   55.1  16.11  3250.27  11.94   42.0
  77 rofChade 2.3                :  3165.08   5200    59.9  1743  2748   709  3117.0   52.8   6.35  3087.28   9.30   88.9
  78 Minic 3.17 NN               :  3162.66   1710    44.4   269   981   460   759.5   57.4  10.80  3210.05  10.67   57.0
  79 Clover 3.0 NN               :  3162.63   1740    47.4   347   954   439   824.0   54.8  10.42  3185.74  10.59   58.0
  80 Rebel 14.2 NN               :  3158.16   1290    50.9   298   718   274   657.0   55.7  11.98  3152.17  10.14   43.0
  81 Lc0 0.26.3 703810 CPU       :  3158.00   2064    59.5   700  1056   308  1228.0   51.2   9.76  3085.36   8.78   41.3
  82 Wasp 5.00 NN                :  3156.92   1200    47.1   202   727   271   565.5   60.6  12.51  3182.25  10.27   40.0
  83 Weiss 2.0                   :  3150.72   3075    43.8   501  1693   881  1347.5   55.1   8.03  3202.68  11.70  107.8
  84 Defenchess 2.3 dev          :  3142.04   5075    48.4  1109  2692  1274  2455.0   53.0   6.44  3158.27  10.58  123.3
  85 Combusken 2.0.0             :  3140.89   2085    42.4   320  1129   636   884.5   54.1   9.70  3204.27  11.44   74.9
  86 Gogobello 3.0 NNSf          :  3137.05   1230    41.7   141   744   345   513.0   60.5  12.24  3206.60   9.61   41.0
  87 Laser 1.7                   :  3133.29   7075    50.8  1686  3823  1566  3597.5   54.0   5.45  3129.48  10.24  139.4
  88 Schooner 2.2 XB             :  3131.94   7075    50.7  1633  3905  1537  3585.5   55.2   5.41  3129.50  10.24  139.4
  89 Zahak 8.6 dev               :  3131.68   1200    43.9   205   643   352   526.5   53.6  12.46  3182.88  10.28   40.0
  90 Fritz 18 (Ginkgo)           :  3128.04   2760    42.2   424  1484   852  1166.0   53.8   8.77  3193.06  11.20   92.0
  91 Bit-Genie 9.19 dev          :  3127.99   1290    37.5   155   658   477   484.0   51.0  12.81  3230.81  10.11   43.0
  92 Zahak 9.0                   :  3127.60   2595    41.4   383  1382   830  1074.0   53.3   8.75  3199.65  11.40   91.9
  93 Velvet 3.2.0 NN             :  3127.50   1740    42.9   270   954   516   747.0   54.8  10.77  3186.34  10.59   58.0
  94 Shredder 13                 :  3125.00   7075    49.8  1668  3713  1694  3524.5   52.5   5.28  3129.63  10.24  139.4
  95 Fritz 17 (Ginkgo)           :  3123.22   4000    56.6  1158  2213   629  2264.5   55.3   7.38  3073.32   9.04   63.9
  96 Rebel 14.1 NN               :  3123.11   1200    37.9   146   618   436   455.0   51.5  13.46  3222.12   9.87   40.0
  97 Clover 2.4                  :  3119.89   1710    39.1   194   950   566   669.0   55.6  11.00  3210.81  10.67   57.0
  98 Defenchess 2.2              :  3117.78   2000    58.0   586  1146   268  1159.0   57.3  10.22  3056.45   9.20   40.0
  99 RubiChess 1.7.3             :  3114.02   2000    57.5   611  1077   312  1149.5   53.9  10.46  3056.54   9.19   40.0
 100 Marvin 5.2.0 NN             :  3111.09   3075    39.0   373  1653  1049  1199.5   53.8   8.12  3203.07  11.70  107.8
 101 Hiarcs 15.0                 :  3107.93   2085    38.4   260  1080   745   800.0   51.8  10.30  3204.74  11.43   74.9
 102 Chiron 5                    :  3107.39   3075    38.6   418  1537  1120  1186.5   50.0   8.46  3203.10  11.70  107.8
 103 Andscacs 0.95               :  3106.50   2690    50.4   642  1425   623  1354.5   53.0   9.00  3106.70   9.96   67.1
 104 DanaSah 9.0 NN              :  3104.11   2565    35.6   258  1311   996   913.5   51.1   9.50  3224.09  11.61   90.9
 105 LG Evolution 3.15.4 NN      :  3102.05   1290    43.6   175   775   340   562.5   60.1  11.92  3153.47  10.14   43.0
 106 Fizbo 2.0                   :  3099.11   7075    46.6  1507  3581  1987  3297.5   50.6   5.17  3130.10  10.24  139.4
 107 Andscacs 0.95.123 dev       :  3096.12   4250    42.9   724  2198  1328  1823.0   51.7   7.03  3157.23  10.00   95.9
 108 Dark Toga 1.1 NN            :  3090.68   1710    35.6   191   836   683   609.0   48.9  11.15  3211.32  10.66   57.0
 109 Stash 32.6 dev              :  3088.81   1290    41.9   189   703   398   540.5   54.5  12.62  3153.78  10.12   43.0
 110 Stash 32                    :  3083.30   1230    35.1   116   631   483   431.5   51.3  12.75  3207.91   9.60   41.0
 111 Halogen 10 NN               :  3072.40   1710    33.5   153   839   718   572.5   49.1  11.34  3211.64  10.66   57.0
 112 GullChess 3.0 Sy            :  3068.92   3800    40.4   601  1866  1333  1534.0   49.1   7.62  3151.77   9.77   81.9
 113 Black Marlin 5.0 NN         :  3065.45    690    27.8    35   314   341   192.0   45.5  19.23  3252.46  11.87   42.0
 114 Wasp 4.50                   :  3065.23   3230    41.7   501  1692  1037  1347.0   52.4   7.48  3136.17   9.16   66.6
 115 Arasan 22.0                 :  3059.31   2000    50.5   490  1038   472  1009.0   51.9  10.02  3057.91   9.20   40.0
 116 Winter 0.9                  :  3057.20   3800    38.9   590  1780  1430  1480.0   46.8   7.68  3152.02   9.77   81.9
 117 GullChess 3.0               :  3056.22   2000    50.0   451  1100   449  1001.0   55.0  10.17  3057.99   9.20   40.0
 118 Arasan 22.2                 :  3052.54   2000    45.5   370  1078   552   909.0   53.9   9.97  3092.10   8.75   40.0
 119 Stash 31.16 dev             :  3047.32   1200    33.4    96   610   494   401.0   50.8  13.84  3184.99  10.24   40.0
 120 Orion 0.8 NNSf              :  3043.15   3774    38.1   461  1956  1357  1439.0   51.8   7.51  3144.57   9.70   80.4
 121 Koivisto 4.19               :  3038.18   2000    43.6   339  1066   595   872.0   53.3  10.15  3092.46   8.75   40.0
 122 Caissa 0.4 NNSf             :  3036.95   1200    36.8   109   666   425   442.0   55.5  12.96  3140.49   9.90   40.0
 123 Drofa 3.3.0                 :  3024.62   1230    34.8   131   594   505   428.0   48.3  12.83  3145.69   9.96   41.0
 124 Nirvanachess 2.5            :  3024.57   3200    37.7   395  1623  1182  1206.5   50.7   7.93  3127.58   9.40   64.6
 125 Combusken 1.4.0             :  3023.42   3200    37.6   445  1514  1241  1202.0   47.3   8.57  3127.61   9.38   64.6
 126 Mantissa 3.3.0 NN           :  3022.51   1200    35.0   119   603   478   420.5   50.3  12.91  3140.85   9.90   40.0
 127 Marvin 5.0.0 NN             :  3020.77   2000    41.4   326  1003   671   827.5   50.1  10.02  3092.90   8.75   40.0
 128 Black Marlin 4.0 NN         :  3015.31   1740    29.7   122   788   830   516.0   45.3  11.74  3188.28  10.57   58.0
 129 Fritz 16 (Rybka)            :  3014.79   3200    40.7   512  1582  1106  1303.0   49.4   8.46  3089.78   9.53   69.0
 130 Chiron 4                    :  3011.81   4000    42.3   653  2076  1271  1691.0   51.9   7.35  3076.11   9.04   63.9
 131 Pedone 2.0                  :  3010.77   2000    44.2   362  1043   595   883.5   52.1  10.12  3059.12   9.20   40.0
 132 Vajolet2 2.8                :  3009.62   5200    40.0   767  2629  1804  2081.5   50.6   6.20  3091.16   9.30   88.9
 133 Expositor 2WQ23 NN          :  3009.50   1200    33.5   119   565   516   401.5   47.1  13.59  3141.17   9.88   40.0
 134 Demolito 2021-07-09         :  3008.77   1200    29.0    93   510   597   348.0   42.5  13.62  3185.95  10.25   40.0
 135 Winter 0.8                  :  3006.29   2000    43.6   391   962   647   872.0   48.1  10.25  3059.23   9.19   40.0
 136 Wasp 4.00                   :  3005.70   2000    43.5   342  1057   601   870.5   52.9  10.66  3059.25   9.18   40.0
 137 Counter 4.0                 :  3001.49   1200    32.5   109   562   529   390.0   46.8  13.69  3141.38   9.88   40.0
 138 Halogen 9 NN                :  3000.34   2000    38.8   284   984   732   776.0   49.2  10.32  3093.41   8.74   40.0
 139 Seer 1.2.1 NN               :  2998.57   2064    39.0   296  1019   749   805.5   49.4   9.92  3089.84   8.78   41.3
 140 Critter 1.6a                :  2994.63   4000    40.1   648  1912  1440  1604.0   47.8   7.13  3076.54   9.05   63.9
 141 Igel 2.5.0                  :  2992.38   2000    41.8   314  1045   641   836.5   52.3  10.13  3059.58   9.20   40.0
 142 Equinox 3.30                :  2992.19   2000    41.8   305  1062   633   836.0   53.1  10.41  3059.59   9.19   40.0
 143 Nirvanachess 2.4            :  2988.65   2064    41.7   317  1086   661   860.0   52.6  10.22  3057.15   9.21   42.2
 144 Demolito 2020-12-24         :  2987.65   2000    37.2   272   945   783   744.5   47.3   9.94  3093.72   8.75   40.0
 145 Mr Bob 1.1.0                :  2981.53   1200    30.2    75   574   551   362.0   47.8  13.75  3141.87   9.88   40.0
 146 Texel 1.08a18               :  2975.33   1200    29.5    85   537   578   353.5   44.8  13.89  3142.03   9.88   40.0
 147 chess22k 1.14 JAVA          :  2975.32   2064    36.1   245  1002   817   746.0   48.5  10.20  3090.49   8.77   41.3
 148 Nemorino 5.00               :  2974.96   2000    39.6   323   939   738   792.5   47.0  10.59  3060.02   9.19   40.0
 149 Texel 1.08a13               :  2969.01   2000    35.0   246   906   848   699.0   45.3  10.73  3094.19   8.73   40.0
 150 Demolito 2020-05-14         :  2965.76   2000    38.5   311   917   772   769.5   45.9  10.10  3060.25   9.20   40.0
 151 Hannibal 1.7                :  2960.69   4000    35.9   485  1902  1613  1436.0   47.5   7.08  3077.39   9.05   63.9
 152 iCE 4.0 v853                :  2957.91   4000    35.6   504  1837  1659  1422.5   45.9   7.32  3077.46   9.04   63.9
 153 Protector 1.9.0             :  2957.19   4064    35.7   479  1940  1645  1449.0   47.7   7.39  3075.97   9.05   64.0
 154 Texel 1.07                  :  2951.78   2000    36.8   263   944   793   735.0   47.2  10.81  3060.60   9.18   40.0
 155 Minic 2.33                  :  2949.53   2000    36.5   269   921   810   729.5   46.0  10.32  3060.65   9.19   40.0
 156 Senpai 2.0                  :  2930.06   2000    34.1   220   925   855   682.5   46.3  10.81  3061.14   9.18   40.0
 157 Combusken 1.2.0             :  2916.71   2000    32.5   204   894   902   651.0   44.7  10.81  3061.47   9.18   40.0
 158 pirarucu 3.3.5 JAVA         :  2910.85   2064    28.6   165   851  1048   590.5   41.2  10.70  3092.30   8.76   41.3
 159 SmarThink 1.98              :  2905.64   2064    31.5   243   815  1006   650.5   39.5  10.41  3059.48   9.20   42.2
 160 Topple 0.8.0                :  2902.62   2000    27.4   147   800  1053   547.0   40.0  11.38  3095.85   8.71   40.0
 161 Monolith 2.01               :  2877.08   2064    28.3   127   914  1023   584.0   44.3  10.99  3060.28   9.19   42.2
 162 Rodent IV 0.22              :  2871.09   2000    27.4   129   839  1032   548.5   42.0  11.33  3062.61   9.17   40.0

White advantage = 61.04 +/- 0.56
Draw rate (equal opponents) = 66.41 % +/- 0.16

[Frank Quisinsky]

Will be checked again with run-5, next month:
Last version from Stockfish and Komodo (without NN) added in my KI-Ratings with two others.
Later ...

Updates for run-5:
114. 01. Stockfish 310720 dev ... available since Jul. 31st, 2020, replaces Stockfish 010422 NN dev from run-4 ... Downgrade (last version without NN)
115. 02. Komodo 14.1 ... available since Nov. 02nd, 2020, replaces Dragon 2.6.1 NN (Komodo) from run-4 ... Downgrade (last version without NN)
116. 03. Nemorino 6.00 NNSF ... available since Jan. 21st, 2021, replaces Nemorino 6.09 NN dev from run-4 ... Downgrade (last release with NNSf)
117. 04. Pedone 3.1 NN ... available since Apr. 25th, 2021, replaces Revenge 2.0 NN from run-4 ... Downgrade (last version before Revenge 1.0)

Best
Frank

[dkappe]

Frank,

thanks for your work. In evaluating elo figures you have to keep in mind time controls, openings, etc. Balanced vs unbalanced openings (like uho) differ by a factor of 2, so +20 elo uho would be +10 elo balanced. Similar calculations can be made for time controls.

What interests me is if one engine outperforms expectations, so an engine is -20 elo with uho but +5 elo balanced. So far I haven’t seen evidence of this, but you run lots of statistics on your games so may be better able to answer this.

[Frank Quisinsky]

You know nothing is perfect Dietrich!

The FEOBOS idea from the year 2016 (ten engines working in teamwork ... find out the balanced positions 3 moves after forming the ECO codes) is old. In NN times very strong engines like Dragon, Stockfish and others evaluate around 8% of the old FEOBOS position not as 100% balanced.

A good example is a lot of "Grünfeld Defence" lines!

In different cases engines like Dragon or Stockfish lost Elo with different balanced complicated positions.
A good example is a lot of "Scandinavian Defense" lines!

I do not want to put the work around FEOBOS in the wastebasket.
I am testing a new way around all the produced games with FEOBOS opening book.

The topic is 1:0, 0:1 games in combination with books stats (without produced draw games). Here I made interesting observations. I can reduce the draw quote and length of games with my FEOBOS based games with such a new book concept drastically. Stronger engines like Dragon and Komodo will get 10-20 Elo more with this new idea.

All the problems seem to be solved:
- lesser draw games
- lesser move length of games without resign-mode
- advantages for stronger engines in rating systems if clear draw-lines are clearly reduced.

Today I added the book in beta 2 in CSS Forum (with *.pgn database) and a description.

CSS Forum:
https://forum.computerschach.de/cgi-bin ... ?tid=13078

To your question:
Never I test the uho idea.
I try more to optimate the idea with balanced lines.

But in my opinion:
“uho” opening: Stockfish (3500 Elo) vs. Wasp (3230 Elo) …
With a balanced line the probably for 2:0 is higher as with an unbalanced line. Wasp has in one of the two games a bigger chance for a draw.

For a rating system not a nice situation for Stockfish if uho lines are used.
Strongest engines lost Elo vs. clearly weaker engines.

Again ...
The same problem I have today with FEOBOS??!
If different of the lines are inside, NN engines today have the opinion that the line is not balanced, I produced the same result uho produced.

The reason I am working on a new book concept!
With the material I have possible! With 40 moves in 40 minutes (160 minutes games) the results of my new book idea are great. Only bad is that from the 500 Eco codes around 30 are lost.

I am thinking with FEOBOS 20.1 from end of the year 2016 strongest Engines lost 10-20 Elo.
I think with uho openings strongest engines lost to others around 10-20 Elo in rating systems.

Best
Frank

[Frank Quisinsky]

And to 200 or 300 Elo different with NN files:

If Komodo 14.1 (without NN) have to play vs. TOP-46 (a lot of opponents with NN) the different with longer time controls is around 200 Elo.
If Komodo 14.1 (without NN) have to play vs. TOP-46 (the opponents without NN files) the different to Dragon 3 (Komodo) is around 260 Elo.

And this is the main problem!!
Clearly weaker engines have a bigger draw chance with NN.

In my opinion we can't mix older with newer results in NN times with rating systems.
A big chaos will be the final result!!

In times today, with so many NN engines, we have to start new rating system.

The next problem is:
With more time the weaker NN engines have clearly higher draw quotes.

Some different things are new for me, I can see in stats.

I will start in around 3-4 months a new KI-Rating list with my new book concept.
That make sense as to waste time in my current KI-Rating system.

Very hard ... in 3-4 months 120.000 80min games in the trash!
But I can use all this games for a clearly better and new opening book system.



Best
Frank

[Frank Quisinsky]

I forgot:
What I wanted to say is:

For very clear ratings it makes sense to let only NN based engines play against each other.
With a very special book, avoids ECO-codes that tend to force a draw but with balanced positions!

Thats my opinion after all the work around the FCP Tourneys!

Working on it ... but for the new book idea I need more games.
The reason that I will work the next 3 months on my old KI-Rating system (before I start new things).

[Graham Banks]

Impressive work and analysis, Frank.
Pretty amazing really.

[lkaufman]

You know nothing is perfect Dietrich!

The FEOBOS idea from the year 2016 (ten engines working in teamwork ... find out the balanced positions 3 moves after forming the ECO codes) is old. In NN times very strong engines like Dragon, Stockfish and others evaluate around 8% of the old FEOBOS position not as 100% balanced.

A good example is a lot of "Grünfeld Defence" lines!

In different cases engines like Dragon or Stockfish lost Elo with different balanced complicated positions.
A good example is a lot of "Scandinavian Defense" lines!

I do not want to put the work around FEOBOS in the wastebasket.
I am testing a new way around all the produced games with FEOBOS opening book.

The topic is 1:0, 0:1 games in combination with books stats (without produced draw games). Here I made interesting observations. I can reduce the draw quote and length of games with my FEOBOS based games with such a new book concept drastically. Stronger engines like Dragon and Komodo will get 10-20 Elo more with this new idea.

All the problems seem to be solved:
- lesser draw games
- lesser move length of games without resign-mode
- advantages for stronger engines in rating systems if clear draw-lines are clearly reduced.

Today I added the book in beta 2 in CSS Forum (with *.pgn database) and a description.

CSS Forum:
https://forum.computerschach.de/cgi-bin ... ?tid=13078

To your question:
Never I test the uho idea.
I try more to optimate the idea with balanced lines.

But in my opinion:
“uho” opening: Stockfish (3500 Elo) vs. Wasp (3230 Elo) …
With a balanced line the probably for 2:0 is higher as with an unbalanced line. Wasp has in one of the two games a bigger chance for a draw.

For a rating system not a nice situation for Stockfish if uho lines are used.
Strongest engines lost Elo vs. clearly weaker engines.

Again ...
The same problem I have today with FEOBOS??!
If different of the lines are inside, NN engines today have the opinion that the line is not balanced, I produced the same result uho produced.

The reason I am working on a new book concept!
With the material I have possible! With 40 moves in 40 minutes (160 minutes games) the results of my new book idea are great. Only bad is that from the 500 Eco codes around 30 are lost.

I am thinking with FEOBOS 20.1 from end of the year 2016 strongest Engines lost 10-20 Elo.
I think with uho openings strongest engines lost to others around 10-20 Elo in rating systems.

Best
Frank

In principle, the idea of "balanced, but complicated and not drawish" openings is very appealing. I assume that "balanced" doesn't mean equal chances for White and Black, but rather than White's advantage is roughly in line with his normal 56% score in chess, is that correct? But the big question is whether it is possible to create a big book with such lines that will dramatically reduce draws between top engines, or only marginally. So for example, if Dragon 3 plays against SF 15 at some time control you like, maybe with a normal book you might get 98% draws (just a guess); with the new one, would that just drop to something like 97% or would we see a more meaningful drop? So far only unbalanced opening books have produced acceptable draw rates between the best engines at Rapid time controls with 4 or more threads. I would like to see that happen with more balanced books, but I am skeptical.

[Frank Quisinsky]

Good evening Larry,

that is exactly the idea ... to collect best of 41.614 balanced FEOBOS positions for eng-eng testing.
In the year 2016 ten engines analyzed with 1min per position on 10 cores.

Average of eval can be see here:
https://www.amateurschach.de/common/feo ... ttings.png
From end of 2016.

As is well known, our engines are with NN clearly stronger.
After some current engine analyzes (random samples) I estimate that 8% of 41.614 FEOBOS positions are perhaps not balanced.

Shortly:
FCP Tourney 2020, 2021, 2022 & FCP Tourney-KI = 170.000 games without resign mode.
I am using the same FEOBOS book.

28.331 won games white (max. 80 moves until mate)
16.275 won games black (max. 85 moves until mate)
= 44.606 games

This weekend I will try to select the A00-E99 positions from this 44.606 games from FEOBOS main database (41.614 positions).
Stefan Pohl wrote today that around 1.600 lines 2 times in the database. 1.600 balanced lines are more as enough.

Yes, I made such a test for around two weeks.
SF 12.04.2022 NN dev vs. Dragon 2.6 NN (Komodo) with an opening book of 28.000 of my 1:0, 0:1 games.
88% draws with 40 moves in 8 minutes, 100 games. I made this test with some other eng-eng combination, and the results are really good (clearly lesser draws and lesser move average). I produced in the last months thousands of games with the idea to use only 1:0, 0:1 games.

If ready, I will set a short information in TalkChess.
So, you can download the database for your own test if you like.

Interesting are test-set with this material.
Perhaps Stefan Pohl will work on such a test-set.
I have more interest to work on an opening book for Shredder GUI.

Best
Frank

[lkaufman]

Good evening Larry,

that is exactly the idea ... to collect best of 41.614 balanced FEOBOS positions for eng-eng testing.
In the year 2016 ten engines analyzed with 1min per position on 10 cores.

Average of eval can be see here:
https://www.amateurschach.de/common/feo ... ttings.png
From end of 2016.

As is well known, our engines are with NN clearly stronger.
After some current engine analyzes (random samples) I estimate that 8% of 41.614 FEOBOS positions are perhaps not balanced.

Shortly:
FCP Tourney 2020, 2021, 2022 & FCP Tourney-KI = 170.000 games without resign mode.
I am using the same FEOBOS book.

28.331 won games white (max. 80 moves until mate)
16.275 won games black (max. 85 moves until mate)
= 44.606 games

This weekend I will try to select the A00-E99 positions from this 44.606 games from FEOBOS main database (41.614 positions).
Stefan Pohl wrote today that around 1.600 lines 2 times in the database. 1.600 balanced lines are more as enough.

Yes, I made such a test for around two weeks.
SF 12.04.2022 NN dev vs. Dragon 2.6 NN (Komodo) with an opening book of 28.000 of my 1:0, 0:1 games.
88% draws with 40 moves in 8 minutes, 100 games. I made this test with some other eng-eng combination, and the results are really good (clearly lesser draws and lesser move average). I produced in the last months thousands of games with the idea to use only 1:0, 0:1 games.

If ready, I will set a short information in TalkChess.
So, you can download the database for your own test if you like.

Interesting are test-set with this material.
Perhaps Stefan Pohl will work on such a test-set.
I have more interest to work on an opening book for Shredder GUI.

Best
Frank

The 88% draw rate for Dragon vs SF with your lowdraw book sounds pretty good, but what was the draw rate for the same match with the normal (unpruned) book? Presumably it was higher, but how much higher? As I'm sure you know, an experiment like this needs a control group for comparison! I suspect that if you reran the lowdraw book test using 4 cpus, Dragon 3 vs SF, and a longer tc like 40 moves in 15 min (CCRL Rapid), the 88% draw rate would rise to the upper 90s. But I could be wrong.