Le 6 avril 2012 13:37, João André <[email protected]> a écrit :
> Dear All,
>
> Hello. My name is João André and I'm a Portuguese phd student at Oxford
> Brookes University. My subject is risk management of bridges during their
> construction phase.
> I've developed a structural robustness index (which basically weights the
> damage accumulation within the structure) which is a random variable but can
> only take values between [0,1]. The particularity is that the cdf of this
> robustness index is a mixture distribution with non-zero probabilistic
> content at R=0 and R=1 and a continuous function within.
> The classical polynomial approximation of the response surface fails to give
> good results because of these discontinuities. I'm using a community
> developed reliability software named OpenTurns (www.openturns.org, developed
> by EDF, EADS and PhimecaSoft) to do my analysis and they have implemented
> libsvm. However, I believe that within scikits new developments of SVR have
> been made that could be more efficient and accurate to solve such a problem
> as mine. I kindly ask your opinion about this and I thank you in advance for
> your time and attention.

The SVR class is the scikit is just a wrapper on the libsvm
implementation so the results will be the same. There is a dense
implementation that might be a bit more efficient that the default
libsvm by avoiding memory copy but that is not a game changer.

For discontinuous target signals you might want to give random forest
regression and extremly randomized regression trees a try though:

  http://scikit-learn.org/dev/modules/ensemble.html

-- 
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel

------------------------------------------------------------------------------
For Developers, A Lot Can Happen In A Second.
Boundary is the first to Know...and Tell You.
Monitor Your Applications in Ultra-Fine Resolution. Try it FREE!
http://p.sf.net/sfu/Boundary-d2dvs2
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to