[R] erro in SVM (packsge "e1071")

[jmoreira@fe.up.pt]

With lm it works.
With ppr and nnet it gives exactly the same error as in svm.
I did:
lm(Duracao ~ ., data= orig.data[1:874,])
ppr(Duracao ~ ., nterms=3, data=orig.data[1:874,], 
       na.action=na.omit, max.terms=6, optlevel=3, 
       sm.method='supsmu', bass=5, span=3/10)
nnet(Duracao ~ ., size = 3, data=orig.data[1:874,], 
    na.action=na.omit, rang = 2, decay = 5^(2*0), 
    maxit = 4000)

Error in get(ctr, mode = "function", envir = parent.frame())(levels(x),  : 
        Orthogonal polynomials cannot be represented accurately enough for 236 
degrees of freedom

Joao

Citando David Meyer <david.meyer at wu-wien.ac.at>:

> So the error occurs during a call to model.matrix() from svm() because
> of the polynomial contrasts--do you get the same error using, e.g.,
> lm()?
> 
> best,
> David
> 
> > The way I call SVM is:
> > 
> > i <- (-2)
> > j <- 4
> > learner='svm'
> > learner.pars=list(Duracao ~ ., data=orig.data,
> >    scale=c(FALSE, TRUE, TRUE, FALSE, FALSE, FALSE, TRUE,
> >     FALSE, FALSE, FALSE, FALSE, FALSE, FALSE,
> >     FALSE, FALSE, FALSE),
> >    type='nu-regression', kernel='linear',
> >    cost=2^(2*i), nu=j/10)
> > learner.pars$data <- orig.data[begin.test.pos:(test.pos-1),]
> > 
> > model <- do.call(learner,learner.pars)
> > 
> > The variables begin.test.pos and test.pos are windexes for orig.data
> > and are working well. in this case begin.test.pos = 1 and test.pos =
> > 875.
> > 
> > orig.data is a data.frame where the second, third and seventh
> > parameters are numeric. The first parameter  is a date and all the
> > others are factors (some of them ordered). The ordered factors are:
> > Dia Semana (week day), DiaAno (day of the year), DiaMes (day of the
> > month), SemanaAno (week of the year) and SemanaMes (week of the
> > month). The first two lines of the orig.data data.frame are:
> >          Data       InicioViagem Duracao     DiaSemana  TipoDia
> >          EpocaEscolar 
> > DiasDesdeUltPagamento DiaAno DiaMes FluxoEntrada FluxoSaida
> > 13 2004-01-01        25056    3220        quinta-feira   feriado 
> > normal                     9                        1             1 
> > normal           fsp4
> > 9  2004-01-01        28554    2866         quinta-feira   feriado 
> > normal                     9                        1             1 
> > normal           fsp4
> >             Modelo                  Motorista SemanaAno SemanaMes
> >             Servico
> > 13    Mercedes_O530_N     10701         1             1 
> > 10597
> > 9      Mercedes_O530_N     11292         1             1 
> > 10597
> > 
> > I am using sliding window with 30 days (around 900 records) for
> > training. The error is in the svm function. May be because SVM uses
> > other functions, but it happens when I run svm.
> > 
> > Thanks a lot for the help
> > 
> > Joao
> > _______________________________________________
> > FEUP - Engineering Faculty, Porto University
> > Engineering and Industrial Management group
> > Tel.: +351 22 508 1639
> > Fax: +351 22 508 1538
> > 
> > ----- Original Message ----- 
> > From: "David Meyer" <david.meyer at wu-wien.ac.at>
> > To: <jmoreira at fe.up.pt>
> > Cc: <r-help at stat.math.ethz.ch>
> > Sent: Sunday, December 19, 2004 1:23 PM
> > Subject: Re: [R] erro in SVM (packsge "e1071")
> > 
> > 
> > > Joao:
> > >
> > > The reported error message is not from e1071.
> > > How *exactly* did you call svm()?
> > >
> > > As to the documentation of the nu parameter: yes, this is an
> > > omission, of course, nu is used in nu-regression as well; thanks for
> > > pointing this out.
> > >
> > > best,
> > > David
> > >
> > > ---------
> > >
> > > Hello,
> > >
> > > I am using SVM under e1071 package for nu-regression with 18
> > > parameters. The
> > > variables are ordered factors, factors, date or numeric datatypes. I
> > > use the
> > > linear kernel.
> > > It gives the following error that I cannot solve. I tryed debug,
> > > browser and
> > > all that stuff, but no way.
> > > The error is:
> > >
> > > Error in get(ctr, mode = "function", envir =
> > > parent.frame())(levels(x),:
> > >        Orthogonal polynomials cannot be represented accurately
> > >        enough
> > > for 236
> > > degrees of freedom
> > >
> > > I use the nu parameter. However, reading ?svm help it says
> > > "parameter needed
> > > for 'nu-classification' and 'one-classification'". Does not say
> > > anything about
> > > nu-regression. It is an omission in the ?svm help page? Or am I
> > > notundestanding something?
> > >
> > > I believe it has something to do with the calculus of the
> > > eigenvalues. Anyway
> > > how can I overpass this problem? Increasing the training data (is
> > > around 900
> > > records)?
> > >
> > > Thanks for any help
> > >
> > > Joao
> > >
> > >
> > >
> > >
> > > -- 
> > > Dr. David Meyer
> > > Department of Information Systems
> > >
> > > 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/
> > > 
> > 
> > 
> 
> 
> -- 
> Dr. David Meyer
> Department of Information Systems
> 
> 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/
>