Author: Uri Blass
Date: 03:00:46 12/18/05
Go up one level in this thread
On December 18, 2005 at 05:11:16, Uri Blass wrote: >On December 18, 2005 at 04:14:19, John J. J. Smith wrote: > >>On December 18, 2005 at 03:51:33, Uri Blass wrote: >> >>>My idea is the following: >>> >>>decide that some weak program has rank 0. >>>Choose hardware and time control. >>> >>>play the Noomen match between every 2 programs(this match has 100 games). >>> >>>Every program that score more than 60% against the weak program that you choose >>>in the Noomen match is at least rank 1. >>> >>>If they cannot score more than 60% against programs with rank 1 they have >>>exactly rank 1. >>> >>>You can see that I defined the programs with rank 1 and the programs with rank >>>higher than 1. >>> >>>Suppose I defined the meaning of programs with rank n and programs with higher >>>ranking than n. >>>programs with rank n+1 are programs that scored more than 60% against at least 1 >>>program with rank n but did not score more than 60% against all the programs >>>with rank that is higher than n. >>> >>>Not that this definition is meaningful only if the ratio is transitive and we >>>need that to have that if A gets more than 60% against B and B gets more than >>>60% against C then C cannot get more than 60% against A. >> >>I think the potential problem is in the transitive assumption. If A beats B and >>B beats C it doesn't always follow that A beats C. Look at how well Ruffian is >>doing against Rybka. > > >I did not claim that if A beats B and B beats C than A beats C but only about >results of more than 60% in 100 games. > >Note also that Rybka scores more than 60% against Ruffian > >http://kd.lab.nig.ac.jp/chess/cegt/pairwise-results-all.shtml > >+4 -11 =8 for Ruffian means 8 out of 23 for Ruffian that is less than 40% > >Even if we take Fritz9 and TogaII 1.0 we do not get more than 60% for the weaker >program. > >Fritz9 scored +19 -34 =27 against Toga1.0 > >It is 33.5/80 that is more than 40% > >Note that we also cannot find many programs that get more than 60% against >TogaII 1.0 > >Rybka32 bit got 32/53 against it that is slightly more than 60% but certainly >Fritz9 is not going to get more than 60% against Rybka and even if it could get >it then it is still not enough when Fritz9 scores slightly more than 40% against >TogaII 1.0. > >Fritz8 bilbao got +21 -13 =10 against TogaII 1.0 and it means 26.5/44 that is >slightly more than 60% but it is not 100 games and it is only slightly more than >60% and it may change with more games. > >Uri I can add about the CEGT statistics that the choice of the opening is different in different matches. I guess that this is the explanation for the fact that deep shredder9 scored significantly better against the 64 bit version of shredder9. My idea is to use fixed positions. Uri
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.