Re: [R] How to insert a certain model in SVM regarding to fixed kernels

From: Amir Safari <amsa36060_at_yahoo.com>
Date: Tue 16 Aug 2005 - 04:05:02 EST

 

Dear David, Dear Gabor , Dear All,
Many thanks for your reply and informative emails. Actually I think it is difficult to define for example a regression model within a SVM framework theoretically and experimentaly. What we have to do is that we work on input data to construct model befor entering data into SVM. This procedure ( building a certain model) can be viewed as a preprocessing of data and or model building. Doesn't it? Parametric models are basically rejected, because SVM learns only nonparametric ones. I would be very appreciated again if I have any correction or guidance. Have a fun,
Amir        

 <ggrothendieck@gmail.com> wrote:
On 8/12/05, Gabor Grothendieck wrote:
> David, Please correct me if I am wrong but I think svm partially works
> with dyn although I don't remember what the specific limitations were.
> Its possible that what works already is enough for Amir. For example,
>
> library(e1071)
> library(dyn)
> set.seed(1)
> y <- ts(rnorm(100))
> y.svm <- dyn$svm(y ~ lag(y))

The above statement should have been y.svm <- dyn$svm(y ~ lag(y,-1)) since we want to bring the previous value of y forward so that it is being used to predict y (rather than predicting y by bringing the future value of y backward). In R positive values for the lag move the series backward and negative values move it forward.

> yp <- predict(y.svm)
> ts.plot(y, yp, col = 1:2)
>
> On 8/12/05, David Meyer wrote:
> > Amir,
> >
> > >
> > > Suppose that we want to regress for example a certain autoregressive
> > > model using SVM. We have our data and also some fixed kernels in
> > > libSVM behinde e1071 in front. The question: Where can we insert our
> > > certain autoregressive model ? During creating data frame ?
> >
> > Yes, I think.
> >
> > > Or perhaps we can make a
> > > relationship between our variables ended to desired autoregressive
> > > model ?
> >
> > Gabor Grothendieck's `dyn` package provides support for the use of
> > general regression functions for time series analysis, and we are
> > currently struggling to integrate the e1071 interface into that
> > framework (but nothing is ready so far). Is it that kind of support you
> > have been looking for?
> >
> > Cheers,
> > David
> >
> > >
> > > Thanks a lot for your help.
> > > Amir Safari
> > >
> > >
> > >
> > >
> > > __________________________________________________
> > > Do You Yahoo!?

> > > http://mail.yahoo.com
> >
> >
> > --
> > Dr. David Meyer
> > Department of Information Systems and Operations
> >
> > Vienna University of Economics and Business Administration
> > Augasse 2-6, A-1090 Wien, Austria, Europe
> > Fax: +43-1-313 36x746
> > Tel: +43-1-313 36x4393
> > HP: http://wi.wu-wien.ac.at/~meyer/
> >
> > ______________________________________________
> > R-help@stat.math.ethz.ch mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-help
> > PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
> >
>


        [[alternative HTML version deleted]]



R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Tue Aug 16 04:18:17 2005

This archive was generated by hypermail 2.1.8 : Sun 23 Oct 2005 - 15:21:18 EST