R help - Oct 2004 - Cluster Analysis: Density-Based Method

Hi people,

Does anybody know some Density-Based Method for clustering implemented in R?

Thanks,

Fernando Prass
		
_______________________________________________________

Fernando Prass wrote:
>Hi people,
>
>Does anybody know some Density-Based Method for clustering implemented in R?
>  
>
Have you looked at CRAN package mclust?
>Thanks,
>
>Fernando Prass
>
>		
>_______________________________________________________ 
>
>______________________________________________
>R-help at 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
>
>
>  
>

-- 

Kjetil Halvorsen.

Peace is the most effective weapon of mass construction.
               --  Mahdi Elmandjra

I'm no expert in this, but mclust is `density-based' because it
estimates
the density with a mixture of Gaussians.  If this is not what you want, you
should clarify what you mean by `density-based'.  Do you mean an algorithm
based on kernel estimator of the density?

Andy
> From: Fernando Prass
> 
> Yes, but mclust don't have a density-based algorithm. Mclust 
> have the algorithm
> BIC, that is a model-based method...
> 
> Fernando Prass
> 
>  --- Kjetil Brinchmann Halvorsen <kjetil at acelerate.com> escreveu: 
> > Fernando Prass wrote:
> > 
> > >Hi people,
> > >
> > >Does anybody know some Density-Based Method for clustering 
> implemented in R?
> > >  
> > >
> > Have you looked at CRAN package mclust?
> > 
> > >Thanks,
> > >
> > >Fernando Prass
> 
> 
> 
> 	
> 	
> 		
> _______________________________________________________ 
> 
> ______________________________________________
> R-help at 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
> 
>

> From: Martin Maechler
> 
> >>>>> "AndyL" == Liaw, Andy <andy_liaw at
merck.com>
> >>>>>     on Thu, 21 Oct 2004 09:18:54 -0400 writes:
> 
>     AndyL> I'm no expert in this, but mclust is `density-based'
>     AndyL> because it estimates the density with a mixture of
>     AndyL> Gaussians.  If this is not what you want, you should
>     AndyL> clarify what you mean by `density-based'.  Do you
>     AndyL> mean an algorithm based on kernel estimator of the density?
> 
> yes, kernel or other "nonparametric" density estimator, is what
> is usually meant in these contexts.
> [ Of course, many "nonparametric" estimators can be seen to live
>   in finite-dimensional spaces, so the difference to an explicit
>  "flexible" / "high dimensional" method isn't that
big.. ]
>
> Martin

Yes.  However, after reading
ftp://ftp.stat.rice.edu/pub/scottdw/TECH/ipra.ps (David Scott's `From
Kernels to Mixtures' published in Technometrics in 2000, I believe the Tukey
memorial issue) I thought the line between kernel densities and mixture
models is rather gray...

Best,
Andy
>     >> From: Fernando Prass
>     >> 
>     >> Yes, but mclust don't have a density-based algorithm.
Mclust
>     >> have the algorithm
>     >> BIC, that is a model-based method...
>     >> 
>     >> Fernando Prass
>     >> 
>     >> --- Kjetil Brinchmann Halvorsen <kjetil at
acelerate.com>
> escreveu: 
>     >> > Fernando Prass wrote:
>     >> > 
>     >> > >Hi people,
>     >> > >
>     >> > >Does anybody know some Density-Based Method for
clustering
>     >> implemented in R?
>     >> > >  
>     >> > >
>     >> > Have you looked at CRAN package mclust?
>     >> > 
>     >> > >Thanks,
>     >> > >
>     >> > >Fernando Prass
>     >> 
>     >> 
>     >> 
>     >> 
>     >> 
>     >> 
>     >> _______________________________________________________ 
>     >> 
>     >> ______________________________________________
>     >> R-help at 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
>     >> 
>     >> 
> 
>     AndyL> ______________________________________________
>     AndyL> R-help at stat.math.ethz.ch mailing list
>     AndyL> https://stat.ethz.ch/mailman/listinfo/r-help
>     AndyL> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
> 
> ______________________________________________
> R-help at 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
> 
>

Andy,

I can be wrong, I'm no expert too, but density estimation is different of
density-model. MClust is a model-basead method because use model statistics
from clustering data (more information in
ftp://ftp.u.washington.edu/public/mclust/tr415R.pdf).

I need some package that implement algorithms like OPTICIS, DBSCAN or
DENCLUE...

Fernando Prass

 
>  --- "Liaw, Andy" <andy_liaw at merck.com> escreveu: 
> > I'm no expert in this, but mclust is `density-based' because
it estimates
> > the density with a mixture of Gaussians.  If this is not what you
want, you
> > should clarify what you mean by `density-based'.  Do you mean an
algorithm
> > based on kernel estimator of the density?
> > 
> > Andy
> > 
> > > From: Fernando Prass
> > > 
> > > Yes, but mclust don't have a density-based algorithm. Mclust 
> > > have the algorithm
> > > BIC, that is a model-based method...
> > > 
> > > Fernando Prass
> > >

Dear Christian,

Do you have DBSCAN algorithm in R? Can you send me? I need it for my 
dissertation (ms degree).

Thanks,

Fernando

 --- Christian Hennig <fm3a004 at math.uni-hamburg.de> escreveu:
> Dear Fernando,
> 
> below you find a DBSCAN function I wrote for my own purposes.
> It comes with no warranty and without proper documentation, but I followed
> the notation of the original KDD-96 DBSCAN paper.
> For large data sets, it may be slow.
> 
> Best,
> Christian
> 
> On Thu, 21 Oct 2004, Fernando Prass wrote:
> 
> > No, kmeans is a partition method. I need a model-based method, like
DBSCAN
> or
> > DENCLUE algorithm...
> > 
> > Fernando Prass
> 
> distvector <- function(x,data){
>   ddata <- t(data)-x
>   dv <- apply(ddata^2,2,sum)
> }
> 
> # data may be nxp or distance matrix
> # eps is the dbscan distance cutoff parameter
> # MinPts is the minimum size of a cluster
> # scale: Should the data be scaled?
> # distances: has to be TRUE if data is a distance matrix
> # showplot: Should the computation process be visualized? 
> # countmode: dbscan gives messages when processing point no. (countmode)
> dbscan <- function(data,eps,MinPts=5, scale=FALSE, distances=FALSE,
>                    showplot=FALSE,
>                    countmode=c(1,2,3,5,10,100,1000,5000,10000,50000)){
>   data <- as.matrix(data)
>   n <- nrow(data)
>   if (scale) data <- scale(data)
>   unregpoints <- rep(0,n)
>   e2 <- eps^2
>   cv <- rep(0,n)
>   cn <- 0
>   i <- 1
>   for (i in 1:n){
>     if (i %in% countmode) cat("Processing point ", i," of
",n, ".\n")
>     unclass <- cv<1
>     if (cv[i]==0){
>       if (distances) seeds <- data[i,]<=eps
>       else{
>         seeds <- rep(FALSE,n)
>         seeds[unclass] <- distvector(data[i,],data[unclass,])<=e2
>       }
>       if (sum(seeds)+unregpoints[i]<MinPts) cv[i] <- (-1)
>       else{
>         cn <- cn+1
>         cv[i] <- cn
>         seeds[i] <- unclass[i] <- FALSE
>         unregpoints[seeds] <- unregpoints[seeds]+1
>         while (sum(seeds)>0){
>           if (showplot) plot(data,col=1+cv)
>           unclass[seeds] <- FALSE
>           cv[seeds] <- cn
>           ap <- (1:n)[seeds]
> #          print(ap)
>           seeds <- rep(FALSE,n)          
>           for (j in ap){
> #            if (showplot) plot(data,col=1+cv)
>             jseeds <- rep(FALSE,n)          
>             if (distances) jseeds[unclass] <- data[j,unclass]<=eps
>             else{
>               jseeds[unclass] <-
distvector(data[j,],data[unclass,])<=e2
>             }
>             unregpoints[jseeds] <- unregpoints[jseeds]+1
> #            if (cn==1)
> #              cat(j," sum seeds=",sum(seeds),"
unreg=",unregpoints[j],
> #                  "
newseeds=",sum(cv[jseeds]==0),"\n")
>             if (sum(jseeds)+unregpoints[j]>=MinPts){              
>               seeds[jseeds] <- cv[jseeds]==0
>               cv[jseeds & cv<0] <- cn
>             }
>           } # for j
>         } # while sum seeds>0
>       } # else (sum seeds + ... >= MinPts)
>     } # if cv==0
>   } # for i
>   if (sum(cv==(-1))>0){
>     noisenumber <- cn+1
>     cv[cv==(-1)] <- noisenumber
>   }
>   else
>     noisenumber <- FALSE
>   out <- list(classification=cv, noisenumber=noisenumber,
>               eps=eps, MinPts=MinPts, unregpoints=unregpoints)
>   out
> } # dbscan
> # classification: classification vector
> # noisenumber: number in the classification vector indicating noise points
> # unregpoints: ignore...
> 
> ***********************************************************************
> Christian Hennig
> Fachbereich Mathematik-SPST/ZMS, Universitaet Hamburg
> hennig at math.uni-hamburg.de, http://www.math.uni-hamburg.de/home/hennig/
> #######################################################################
> ich empfehle www.boag-online.de
> 
>