[Computer-go] Informal study

Don Dailey dailey.don at gmail.com
Sat Nov 13 05:46:58 PST 2010


Just for fun I just took a snapshot of the main page and the bayeselo page
and I'm comparing the ELO ratings on each page for a few active programs
near the top:


*Program              BayesElo  Incremental*
*------------------  --------  ------------*
*MyGo1.0_np              2775         2672  (not active)*
*Valkyria3.5.16_4cx      2723         2625*
*Fuego-1166-8c           2722         2601*
*ToGo-207_1c_b           2558         2484*
*Valkyria3.5.9_P4Bx      2560         2448*
*iced_goanna             2295         2199*
*N11_L3Chkif3.3          2248         2147*


We should, over time,  see these gradually converge.  Note that we are
approximately 100 ELO off in every case.

I have lazy,0 and lazy,1 playing,  but not simultaneously.    From time to
time I am synchronizing their ratings with bayeselo.   I also noticed when
doing this that their ratings jumped.   So these 2 programs will be feeding
the ELO economy with inflation points to bring it up to the bayeselo
standard.

This will probably take a long time - but I don't view that as a bad thing.
   I'm not even sure this additional program will do the job because the
anchor by itself was not strong enough to prevent this deflation.    The one
active anchor probably prevents it from drifting more than 100 ELO.

It occurred to me that I could also copy over the bayeselo ratings of
players who have played less than N games,  with the understanding that
their ratings with few games are pretty unreliable and their performance
ratings (as measured by bayeselo) are more reliable.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20101113/3a1f0d4f/attachment.html>


More information about the Computer-go mailing list