Stronger Rybka 4 settings - FRC list updated

Questions and comments related to CCRL testing study
Post Reply
Ray
Posts: 22607
Joined: Sun Dec 18, 2005 6:33 pm
Sign-up code: 10159
Location: NZ

Stronger Rybka 4 settings - FRC list updated

Post by Ray »

Shaun tested a Rybka setting at standard chess blitz. That setting, Rybka 4 C13510 64-bit, currently shows some +23 ELO over default settings, here:

http://www.computerchess.org.uk/ccrl/40 ... librate=no


I decided to test this setting at FRC as well, and I got a similar result, slightly better in fact, +29 ELO above default, and also the draw rate drops from 25% to 17%:

http://www.computerchess.org.uk/ccrl/40 ... librate=no

What is particularly stunning, is that Rybka 4 C13510 64-bit is +138 ELO above Rybka 3 64-bit, what a massive improvement
Ray
Posts: 22607
Joined: Sun Dec 18, 2005 6:33 pm
Sign-up code: 10159
Location: NZ

Re: Stronger Rybka 4 settings - FRC list updated

Post by Ray »

I decided to run the FRC database through ELOSTAT as well. Our ratings lists use BayesELO

ELOSTAT has a slightly more conservative view of things:

Rybka 4 x64 to Rybka 4 C13510 64-bit improvement = +16 ELO (BayesELO says +29 ELO)

Otherwise ELOSTAT is similar
Rybka 3 x64 - to Rybka 4 x64 improvement = +116 ELO
Rybka 4 x64 to Rybka 4 C13510 64-bit improvement = +16 ELO
Rybka 3 x64 - to Rybka 4 C13510 64-bit improvement = +132 ELO

BayesELO (the official list says)
Rybka 3 x64 - to Rybka 4 x64 improvement = +109 ELO
Rybka 4 x64 to Rybka 4 C13510 64-bit improvement = +29 ELO
Rybka 3 x64 - to Rybka 4 C13510 64-bit improvement = +138 ELO
zezius
Posts: 2
Joined: Mon Jul 26, 2010 11:07 pm
Sign-up code: 10159

Re: Stronger Rybka 4 settings - FRC list updated

Post by zezius »

It´s better here also.

thank you.
Post Reply