Author: Vasik Rajlich
Date: 01:20:17 12/15/05
Go up one level in this thread
On December 14, 2005 at 19:20:46, enrico carrisco wrote: >On December 14, 2005 at 14:32:59, Vasik Rajlich wrote: > >>On December 14, 2005 at 14:14:36, enrico carrisco wrote: >> >>>On December 14, 2005 at 05:30:59, Vincent Lejeune wrote: >>> >>>>On December 14, 2005 at 05:22:16, Gian-Carlo Pascutto wrote: >>>> >>>>>On December 14, 2005 at 05:13:48, Joachim Rang wrote: >>>>> >>>>>>On December 14, 2005 at 05:06:41, Mimic wrote: >>>>>> >>>>>>>============================================================== >>>>>>>Rybka 1.0 Beta 32-bit +282-130=388 59.50% 476.0/800 +66 >>>>>>>-------------------------------------------------------------- >>>>>>>Fruit 2.2.1 +130-282=388 40.50% 324.0/800 -66 >>>>>>>============================================================== >>>>>>> >>>>>>> Games Rybka >>>>>>>======================== >>>>>>> 100 +31 >>>>>>> 200 ----- +33 >>>>>>> 300 +50 >>>>>>> 400 ----- +50 >>>>>>> 500 +53 >>>>>>> 600 ----- +60 >>>>>>> 700 +63 >>>>>>> 800 ----- +66 >>>>>>>======================== >>>>>>> >>>>>>>Pentium M Dothan 2.52Ghz >>>>>>>32MB Hash >>>>>>>No EGTB >>>>>>> >>>>>>>Mimic >>>>>> >>>>>>has Rybka some hidden learning? The steady increase of the rating difference is >>>>>>surprising. >>>>> >>>>>The rating difference didn't increase, just the plus score. (And it wasn't >>>>>steady, either) >>>>> >>>>>-- >>>>>GCP >>>> >>>> >>>>No, +66 is the rating difference, not the score difference. >>>> >>>>Surprisingly it doubled between 200 and 800 games... >>>>Who know it will be after 1200 games ... >>> >>>Rybka scores SIGNIFICANTLY better (15-20%+) vs. a given opponent (Fruit 2.2.1 in >>>this case) in an engine-engine match than it will in a tournament style match >>>with MORE than two engines (i.e. engines being routinely loaded and unloaded.) >>> >>>Try it yourself. >>> >>>Regards, >>> >>>-elc. >> >>There is no learning or opponent recognition in Rybka, so this observation is >>likely an anomaly. >> >>The only other explanation is that the opponents are "learning" :) >> >>Vas > >Hello Vasik. > >That's very interesting then... I will do some more tests and see if this >continues (as well as check the results of others that test but do not post >here.) > >So far I have 4 tournaments (at least 3 engines inlcuding Rybka) with 126 games >each tournament and 5 engine vs. engine matches (also 126 games each) and >Rybka's percentage against the same opponents in the tournament are better (by >at least 15%) when it is engine vs. engine only. > >Hopefully continuing the matches will break the anomaly. :) > >Regards, > >-elc. That would be quite an anomaly. Can you also check to make sure that there isn't some sort of weird load-time problem? Thanks, Vas
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.