R help - Jul 2017 - R-Package for Recursive Partitioning without Classification or Regression

Hello,

I have a question related to recursive partitioning, but I cannot find
an answer, likely because I don't know how to properly word my Google
search query.


All recursive partitioning examples, which I can find, are used for
either classification or regression trees like

   library(tree)
   data(iris)
   tree(Species ~ Sepal.Width + Petal.Width, data = iris)

which implies building a model. However, I would like to split data
like clustering similar to decision tree methods, because I have
nothing to predict.


My question is: Is there a package, which I can use to partition my
data without classification or regression so that it resembles
clustering methods?


Thanks and regards,

Dirk

Tom:

Recursive partitioning requires a response variable because splitting
criteria are based on node purity/homogeneity (by various criteria) of
the responses.

So why did you not just search on "clustering" (e.g. at rseek.org)?
And in particular, have you looked at the CRAN task view on clustering
here:

https://cran.r-project.org/web/views/Cluster.html

If nothing there satisfies your needs, you may need to clarify your query.

Cheers,
Bert


Bert Gunter

"The trouble with having an open mind is that people keep coming along
and sticking things into it."
-- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )


On Fri, Jul 28, 2017 at 5:56 AM, Tom D. Harray <tomdharray at gmail.com>
wrote:
> Hello,
>
> I have a question related to recursive partitioning, but I cannot find
> an answer, likely because I don't know how to properly word my Google
> search query.
>
>
> All recursive partitioning examples, which I can find, are used for
> either classification or regression trees like
>
>    library(tree)
>    data(iris)
>    tree(Species ~ Sepal.Width + Petal.Width, data = iris)
>
> which implies building a model. However, I would like to split data
> like clustering similar to decision tree methods, because I have
> nothing to predict.
>
>
> My question is: Is there a package, which I can use to partition my
> data without classification or regression so that it resembles
> clustering methods?
>
>
> Thanks and regards,
>
> Dirk
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.

On Fri, 28 Jul 2017, Tom D. Harray wrote:
> Hello,
>
> I have a question related to recursive partitioning, but I cannot find
> an answer, likely because I don't know how to properly word my Google
> search query.
I think you are looking for "divisive hierarchical clustering" which
is
the more commonly used term for clustering based on recursive 
partitioning. The classical implementation for this would be diana() from 
the "cluster" package. The "Cluster" task view at 
https://CRAN.R-project.org/view=Cluster also lists "isopam" and 
http://search.r-project.org/cgi-bin/namazu.cgi?query=divisive+clustering&idxname=functions&idxname=views
also gives a few further leads.
> All recursive partitioning examples, which I can find, are used for
> either classification or regression trees like
>
>   library(tree)
>   data(iris)
>   tree(Species ~ Sepal.Width + Petal.Width, data = iris)
>
> which implies building a model. However, I would like to split data
> like clustering similar to decision tree methods, because I have
> nothing to predict.
But do you want to split along observed variables or not? If not, then 
you're in a unsupervised clustering situation (see my comment above). But 
if you want to split along covariates, then this is supervised and you're 
possibly looking for a multivariate regression tree?

Hope that helps,
Z
> My question is: Is there a package, which I can use to partition my
> data without classification or regression so that it resembles
> clustering methods?
>
>
> Thanks and regards,
>
> Dirk
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.
>

I think that diana() or mona() in package cluster are what you are looking for.
The computational burdens of polythetic divisive clustering methods as they are
called are substantial so the available methods use shortcuts that should
provide good, but not necessarily optimal results.

-------------------------------------
David L Carlson
Department of Anthropology
Texas A&M University
College Station, TX 77840-4352

-----Original Message-----
From: R-help [mailto:r-help-bounces at r-project.org] On Behalf Of Tom D. Harray
Sent: Friday, July 28, 2017 7:56 AM
To: r-help at r-project.org
Subject: [R] R-Package for Recursive Partitioning without Classification or
Regression

Hello,

I have a question related to recursive partitioning, but I cannot find
an answer, likely because I don't know how to properly word my Google
search query.


All recursive partitioning examples, which I can find, are used for
either classification or regression trees like

   library(tree)
   data(iris)
   tree(Species ~ Sepal.Width + Petal.Width, data = iris)

which implies building a model. However, I would like to split data
like clustering similar to decision tree methods, because I have
nothing to predict.


My question is: Is there a package, which I can use to partition my
data without classification or regression so that it resembles
clustering methods?


Thanks and regards,

Dirk

______________________________________________
R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.

Hello Tom,
what you describe seems like using hierarchical clustering and
plotting a dendrogram to me, even though only in the "visual" way, not
in the math behind the algorithms.

Are you looking for something like this?
https://rpubs.com/gaston/dendrograms

Just please note, that this is very far from rpart in term of how the
algorithms work (already mentioned by Bert Gunter)

Best,
Michal

2017-07-28 14:56 GMT+02:00 Tom D. Harray <tomdharray at
gmail.com>:
> Hello,
>
> I have a question related to recursive partitioning, but I cannot find
> an answer, likely because I don't know how to properly word my Google
> search query.
>
>
> All recursive partitioning examples, which I can find, are used for
> either classification or regression trees like
>
>    library(tree)
>    data(iris)
>    tree(Species ~ Sepal.Width + Petal.Width, data = iris)
>
> which implies building a model. However, I would like to split data
> like clustering similar to decision tree methods, because I have
> nothing to predict.
>
>
> My question is: Is there a package, which I can use to partition my
> data without classification or regression so that it resembles
> clustering methods?
>
>
> Thanks and regards,
>
> Dirk
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.