(Compact) Symbolic regression in gnumeric?

[Fabian Deutsch <fabian deutsch gmx de>]

Hey,

as a part of my diploma thesis I developed a symbolic regression tool,
finding _compact formulas_ for datasets.
So what does this mean? E.g. it can find the formula for the surface of
a circle providing samples to the software.

Maybe it makes sense or there is interest in integrating this method as
an alternative to classic curve-fitting methods into gnumeric. AFAIK
neither LibreOffice nor Microsoft Office ship this kind of curve-fitting
technique.

I've written a prototype in vala which works fairly well. It can be
found at
http://gitorious.org/pigp/libmlgp

It includes a library doing all the work and a simple commandline
interface to run a symbolic regression. Please handle this software with
care, sometimes it does not know what it is doing :)

The docs/ folder provides more informations on how to build and how to
run a regression.

More on the topic of symbolic regression (Schmidt and Lipson published a
nice implementation in 2009 which raised my interest): 
http://ccsl.mae.cornell.edu/eureqa
http://www.hakank.org/eureqa/
My implementation differs from different ones, as it uses
multi-objective optimization (via NSGA-2) to find compact formulas,
classical symbolic regression is also fitting curves but by creating
arbitary long formulas.

Questions | Thoughts?

- fabian