[R] Piecewise Lasso Regression

[Lucas Santana dos Santos]

Dear Vito,

Thank you very much for the advice -  I will try it and see how the regression looks.

Best,

Lucas

On Jun 6, 2012, at 11:45 AM, Vito Muggeo (UniPa) wrote:

> dear Lucas,
> If you are interested in selecting the number of breakpoints here a possible remedy:
> 
> 1. Fit a segmented model with a large number of breakpoints via the arguments psi=NA and stop.if.error=FALSE in seg.control() (see the example below)
> 
> 2. extract the "model matrix" relevant to the variables of the breakpoints
> 
> 3. Use lars with the extracted model matrix
> 
> Notice that there are at least two issues that I did not mention:
> a) The criterion to be used to select the variables (i.e the breakpoints)
> b) When you remove variables (i.e. breakpoints) you are assuming that the ML estimate of the remaining breakpoints is the same (and this is not the case here, as point estimates of the breakpoints depend on the number of breakpoints themselves..)
> 
> 
> vito
> 
> ###Here a simple example
>     n=100
>     xx<-1:n/n
>     psi0<-seq(.2,.8,by=.2)
>     X<-sapply(c(0,psi0), function(x)pmax(xx-x,0))
>     b<-c(-.6,.7,.4,-1,.5)
>     mu<-drop(X%*%b)
>     yy<-mu+rnorm(n)*.01
>     plot(xx,mu);lines(xx, mu)
> 
> 
> 
> library(segmented)
> o<-lm(yy~xx)
> os<-segmented(o,seg.Z=~xx,psi=NA,control=seg.control(stop.if.error=FALSE,K=30, n.boot=0))
> plot(os, add=T, col=2) #have a look to results
> 
> #extract the new model matrix
> new.X<-model.matrix(os)[,os$nameUV$U]
> 
> #finally use lars on y~new.X
> 
> Hope this helps you,
> vito
> 
> Il 06/06/2012 15.48, Lucas Santana dos Santos ha scritto:
>> Hi Vito,
>> 
>> I am more interested in selecting the number of breakpoints. My data has some structure and I believe that fitting a piecewise regression would be of great benefit.
>> 
>> Thanks,
>> 
>> Lucas
>> 
>> On Jun 6, 2012, at 4:54 AM, Vito Muggeo (UniPa) wrote:
>> 
>>> dear lucas,
>>> yes you are right, segmented does not handle 'lars' objects.
>>> 
>>> Out of curisity, are you interested in selecting the number of breakpoints or in selecting additional covariates with linear parameters?
>>> 
>>> vito
>>> 
>>> 
>>> Il 06/06/2012 0.01, Lucas Santana dos Santos ha scritto:
>>>> Hi All,
>>>> 
>>>> 
>>>> I am trying to fit a piecewise lasso regression, but package Segmented does not work with Lars objects.
>>>> Does any know of any package or implementation of piecewise lasso regression?
>>>> 
>>>> Thanks,
>>>> 
>>>> Lucas
>>>> ______________________________________________
>>>> R-help at r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
>>>> and provide commented, minimal, self-contained, reproducible code.
>>>> 
>>> 
>>> --
>>> ====================================
>>> Vito M.R. Muggeo
>>> Dip.to Sc Statist e Matem `Vianelli'
>>> Università di Palermo
>>> viale delle Scienze, edificio 13
>>> 90128 Palermo - ITALY
>>> tel: 091 23895240
>>> fax: 091 485726
>>> http://dssm.unipa.it/vmuggeo
>>> ====================================
>> 
>> 
> 
> -- 
> ====================================
> Vito M.R. Muggeo
> Dip.to Sc Statist e Matem `Vianelli'
> Università di Palermo
> viale delle Scienze, edificio 13
> 90128 Palermo - ITALY
> tel: 091 23895240
> fax: 091 485726
> http://dssm.unipa.it/vmuggeo
> ====================================