Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: CEGT testing of CM10

Author: Heinz van Kempen

Date: 00:30:15 01/24/06

Go up one level in this thread


On January 24, 2006 at 01:00:32, Maurizio De Leo wrote:

>I think one of the best things that CEGT could do to improve the testing of CM10
>settings is to actually play more games with the default instead of trying even
>more setting. Right now the error bars are too high and testing more and more
>setting will not help, becaue they can`t be differentiated. If the error bars of
>the default could be reduced to 12-15, everything would become much more clear.
>We could even find that the error bar for some of the best settings (and Spock
>seems pretty good) don`t overlap with that of default.
>
>Maurizio
>
>
>
>On January 23, 2006 at 15:56:04, Wilhelm Hudetz wrote:
>
>>Hi All,
>>
>>CEGT Blitz after 514 games:
>>
>>    Program                          Elo    +   -   Games   Score   Av.Op.
>>Draws
>>
>>
>>  1 CM10th Mr.Spock                : 2711   29  24   514    53.2 %   2688   >  2 CM9000 Xperience               : 2703   32  32   304    50.0 %   2703   %
>>  3 CM10th Magic II                : 2699   49  32   198    50.5 %   2696   >  4 CM10th R10                     : 2698   32  25   432    52.5 %   2680   >  5 CM10th Berean 5.54             : 2697   51  36   184    51.1 %   2689   >  6 CM10th Xperience               : 2697   48  32   207    51.2 %   2688   >  7 CM10th Steadfast               : 2696   34  39   251    42.2 %   2751   >  8 CM9000 Slayer 2B               : 2695   31  43   266    49.6 %   2698   >  9 CM9000 R1                      : 2692   45  37   233    51.1 %   2685
>> 10 CM9000 Gladiator               : 2687   44  44   172    50.0 %   2687
>> 11 CM10th Jabba the Hutt          : 2687   23  18   863    51.4 %   2677
>> 12 CM10th Master Yoda             : 2685   27  21   619    52.3 %   2668
>> 13 CM10th R2D2 II                 : 2682   34  27   381    52.2 %   2666
>> 14 CM9000 Apex                    : 2681   35  46   218    48.6 %   2691
>> 15 CM10th Imperator               : 2680   34  45   230    49.1 %   2686
>> 16 CM10th R2D2                    : 2676   38  28   337    50.4 %   2673
>> 17 CM10th Default                 : 2668   39  43   218    44.3 %   2708
>> 18 CM10th Behemoth                : 2668   36  48   185    46.2 %   2694


Hi Maurizio,

good point. I would like to encourage CEGT testers to pay more attention to
CM10th Default in Blitz.

The amount of 40/40 games played with default setting is higher. 710 games are
played with the default setting. Here is the rating and those from other
settings with more games and ranked fairly good:

 39 CM10th Imperator               : 2686   23  23   604    52.0 %   2672   33.8
%
 40 CM10th Xperience               : 2685   16  16  1188    47.6 %   2702   37.5
%
 41 CM10th Behemoth                : 2685   27  27   337    48.7 %   2694   46.9
%
 42 CM10th Cell                    : 2684   14  14  1410    47.9 %   2699   38.4
%
 46 CM10th Pestilence              : 2679   16  16  1179    48.5 %   2690   38.5
%
 47 CM10th Milan 2.3               : 2679   27  27   394    48.5 %   2690   38.1
%
 48 CM10th Default                 : 2679   21  21   710    51.9 %   2666   34.5
%

So the error bars for the default setting are +-21, what is still a bit too much
in my opinion. None of the settings with really many games does differ a lot
from default. So maybe Mr. Spock might be the first one. CM settings fans will
never lose hope to finally detect a very good one :-).

Blitz is still a bit neglected and was started only some months ago in CEGT. So
here some games for the default setting will surely soon added.

Best Regards
Heinz



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.