This is very cool. As of 261 games played, I find it very difficult to
guess whether the mogo curve is beginning to dramatically flatten, or
will continue to rise steeply.
I have a few questions.
I can't see the cross table, I guess you haven't put it up yet?
How do you decide the pairings? I think more even pairings would be
more informative, and lead to faster convergence, but I suppose you
don't just want bots playing adjacent bots?
Are you going to ask for volunteers to run this script? I think it
is more likely to give interesting results than the SETI thing.
At 01:31 19/01/2008, Don wrote:
The new scalability study is in progress. It will be very slow going,
only a few games a day can be played but we are trying to get more
computers utilized.
I will update the data a few times a day for all to see. This includes
a crosstable and ratings graphs. The games will be made available for
anyone who wants them.
Although it's not on the graph itself, Gnugo-3.7.11 level 10 is set to
be 1800.0 ELO. The bayeselo program is used to calculate ratings.
Results can be found here:
http://cgos.boardspace.net/study/index.html
bayeslo assumes all players are equal strength until significant
evidence proves otherwise, and with only a few games played each, the
graph will look truly strange, but this will correct itself in time.
Each data point in the x axis represent a doubling in power. There are
13 doublings represented and mogo was set to approximate FatMan in
strength at each point - at least at the lower level but only with a
very crude and rough estimate. Since I don't know how well mogo
scales compared to FatMan, we can only wait and see how this works out
at the top levels.
- Don
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
--
This email has been verified as Virus free.
Virus Protection and more available at http://www.plus.net
_______________________________________________
computer-go mailing list
[email protected]
http://www.computer-go.org/mailman/listinfo/computer-go/
