Christoph Helma wrote:>
> Hallo,
>
> I have a question concerning SVM regression in R. I intend to use SVMs for
feature selection (and knowledge discovery). For this purpose I will need to
extract the weights that are associated with my features. I understand from a
previous thread on SVM classification, that predictive models can be derived
from SVs, coefficiants and rhos, but it is unclear for me how to transfer this
information to the regression problem. Can anyone help in this respect (I am
*not* an SVM expert)?
That's pretty simple.
The ``decision'' (predictor) function for regression is as follows:
f(x) = \sum_{i=1}^{l} alpha_i * K(x_i, x) - rho
where `alpha_i' are the coefficients of the SVs, `x_i' are the SVs
themselves, and `l' the number of SVs.
Note that `rho' must be *substracted* because libsvm returns -b for some
reasion.
Best,
David.
>
> Thanks,
> Christoph
> --
> :: christoph helma
> :: computational toxicologist
> :: university freiburg
> :: georges koehler allee 079, d-79110 freiburg/br
> :: phone ++49-761-203-8013, fax -8007
> :: helma at informatik.uni-freiburg.de
> :: http://www.informatik.uni-freiburg.de/~helma/
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> http://www.stat.math.ethz.ch/mailman/listinfo/r-help
--
Mag. David Meyer Wiedner Hauptstrasse 8-10
Vienna University of Technology A-1040 Vienna/AUSTRIA
Department of Tel.: (+431) 58801/10772
Statistics and Probability Theory Fax.: (+431) 58801/10798