Hi,
I am still making slow progress on my automatic parameter-optimization
tool. I have just released a user-friendly GUI that you might like to try:
http://remi.coulom.free.fr/QLR/
Source code and Windows binaries are provided. Source code should be
easy to compile on Ubuntu linux. The GUI is based on Qt, so, in theory,
it should compile and work for Mac OS, too.
I also made progress on theory. In a previous mail, I conjectured that
the asymptotic rate of convergence is O(1/n^{3/4}). It turns out this
was wrong. My new conjecture is now O(1/n^{3/5}). It may be wrong too,
but it is a lot less wrong that the previous one. It is still very good,
because it is better than O(1/sqrt(n)). This means that if the fonction
to be maximized is smooth, you can approach optimal performance faster
than you can measure it. I am still working on a paper with more details.
I am not working on Crazy Stone any more, so I don't have a strong
motivation to apply this algorithm to my own program. I hope you find it
useful. Don't hesitate to ask if you have any question. If you find a
situation where it seems that QLR does not behave well, please send me
your data file. I am interested in your feedback.
Rémi
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
