R devel - Jan 2006 - e1071: using svm with sparse matrices (PR#8527)

Full_Name: Julien Gagneur
Version: 2.2.1
OS: Linux (Suse 9.3)
Submission from: (NULL) (194.94.44.4)


Using the SparseM library (SparseM_0.66)
and the e1071 library (e1071_1.5-12)


I fail using svm method with a sparse matrix. Here is a sample example.

I experienced the same problem under Windows.


> library(SparseM)
[1] "SparseM library loaded"
> library("e1071")
Loading required package: class
> data(iris)
> attach(iris)
> M=as.matrix(iris[,1:4])
> Msparse=as.matrix.csr(M)
> Species=iris[,5]
> mod=svm(Msparse,Species)
Error in svm.default(Msparse, Species) : object "nac" not found

First: this is not a bug, more a feature request.

Secondly, even if it was a bug, it is _not_ a bug in R, please read  
the posting rules for bugs. Now a member of R-core has to use  
valuable time to clean up after your bug report.

Correspondence such as this such really be sent to the package  
maintainers (and - perhaps - a cc to R-devel).

Having said all of this, of course it would be nice if svn supported  
sparse matrices.

/Kasper


On Jan 27, 2006, at 2:02 AM, julien.gagneur at embl.de wrote:
> Full_Name: Julien Gagneur
> Version: 2.2.1
> OS: Linux (Suse 9.3)
> Submission from: (NULL) (194.94.44.4)
>
>
> Using the SparseM library (SparseM_0.66)
> and the e1071 library (e1071_1.5-12)
>
>
> I fail using svm method with a sparse matrix. Here is a sample  
> example.
>
> I experienced the same problem under Windows.
>
>
>
>> library(SparseM)
> [1] "SparseM library loaded"
>> library("e1071")
> Loading required package: class
>> data(iris)
>> attach(iris)
>> M=as.matrix(iris[,1:4])
>> Msparse=as.matrix.csr(M)
>> Species=iris[,5]
>> mod=svm(Msparse,Species)
> Error in svm.default(Msparse, Species) : object "nac" not found
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel

From: Kasper Daniel Hansen
> 
> First: this is not a bug, more a feature request.
> 
> Secondly, even if it was a bug, it is _not_ a bug in R, please read  
> the posting rules for bugs. Now a member of R-core has to use  
> valuable time to clean up after your bug report.
> 
> Correspondence such as this such really be sent to the package  
> maintainers (and - perhaps - a cc to R-devel).
> 
> Having said all of this, of course it would be nice if svn supported  
> sparse matrices.
Libsvm, the `engine' underneath svm() in `e1071', uses a sparse
representation of the data.  I vaguely recall seeing Chih-Jen Lin's code
that uses the SparseM package to pass sparse data to svm()...  David would
know best, of course.

Andy

 
> /Kasper
> 
> 
> On Jan 27, 2006, at 2:02 AM, julien.gagneur at embl.de wrote:
> 
> > Full_Name: Julien Gagneur
> > Version: 2.2.1
> > OS: Linux (Suse 9.3)
> > Submission from: (NULL) (194.94.44.4)
> >
> >
> > Using the SparseM library (SparseM_0.66)
> > and the e1071 library (e1071_1.5-12)
> >
> >
> > I fail using svm method with a sparse matrix. Here is a sample  
> > example.
> >
> > I experienced the same problem under Windows.
> >
> >
> >
> >> library(SparseM)
> > [1] "SparseM library loaded"
> >> library("e1071")
> > Loading required package: class
> >> data(iris)
> >> attach(iris)
> >> M=as.matrix(iris[,1:4])
> >> Msparse=as.matrix.csr(M)
> >> Species=iris[,5]
> >> mod=svm(Msparse,Species)
> > Error in svm.default(Msparse, Species) : object "nac" not
found
> >
> > ______________________________________________
> > R-devel at r-project.org mailing list
> > https://stat.ethz.ch/mailman/listinfo/r-devel
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
> 
>

> 
> First: this is not a bug, more a feature request.
[not in R as correctly pointed out, so just for the records:]

It _is_ a bug (in e1071), since svm() is indeed supposed to support
sparse data. The bug was introduced in 1.5-9 I think when support for
correct na-handling was added. The bug will be fixed in 1.5-13.

Thanks for pointing this out!

David.
> 
> Secondly, even if it was a bug, it is _not_ a bug in R, please read  
> the posting rules for bugs. Now a member of R-core has to use  
> valuable time to clean up after your bug report.
> 
> Correspondence such as this such really be sent to the package  
> maintainers (and - perhaps - a cc to R-devel).
> 
> Having said all of this, of course it would be nice if svn supported  
> sparse matrices.
> Libsvm, the `engine' underneath svm() in `e1071', uses a sparse
> representation of the data.  I vaguely recall seeing Chih-Jen Lin's
code
> that uses the SparseM package to pass sparse data to svm()...  David would
> know best, of course.
> Andy
 
> /Kasper
> 
> 
> On Jan 27, 2006, at 2:02 AM, julien.gagneur at embl.de wrote:
> 
> > Full_Name: Julien Gagneur
> > Version: 2.2.1
> > OS: Linux (Suse 9.3)
> > Submission from: (NULL) (194.94.44.4)
> >
> >
> > Using the SparseM library (SparseM_0.66)
> > and the e1071 library (e1071_1.5-12)
> >
> >
> > I fail using svm method with a sparse matrix. Here is a sample  
> > example.
> >
> > I experienced the same problem under Windows.
> >
> >
> >
> >> library(SparseM)
> > [1] "SparseM library loaded"
> >> library("e1071")
> > Loading required package: class
> >> data(iris)
> >> attach(iris)
> >> M=as.matrix(iris[,1:4])
> >> Msparse=as.matrix.csr(M)
> >> Species=iris[,5]
> >> mod=svm(Msparse,Species)
> > Error in svm.default(Msparse, Species) : object "nac" not
found
> >