>At the moment, Pebbles is creating a huge drift. Brian - CGOS requires us to 
>use
>new names on the server each time we change our bots. It computes the strength
>using all games (heavilly biased with the results of the first 100 games)?

This is basically my first working version. Its rating dropped from 1500 to 800
while I worked out time control, position repetition, crash bugs and so on.
Pebbles is basically just getting those points back.

The solution to drift is to keep Fatman-1 running.

I recall reading about an Elo system that had better adaptation to players whose
rating changes. It was called Glicko-2. Here is a link: 
http://en.wikipedia.org/wiki/Glicko_rating_system


>Pebbles is probably closer to 2000 ELO than 1000 ELO based on the games 
>against my bots.

Closer, yes, but I don't think it could be that high, as?Pebbles has a 
significant minus
in recent games against bots rated over 1800.

Its Bayes Elo is 1519, which also does not suggest that it is headed for 2000.

Also, it searches only 20K light (mostly) playouts per turn. (Old computer.)


>I did spot a bug in hb797-50k in its first resignation to Pebbles. My count 
>was hb797-50k +1.5

Happy to help. :-)


Brian

_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/

Reply via email to