[R] Oblique decision tree using mvpart

[Achim Zeileis]

On Wed, 3 Sep 2008, Clusty wrote:

> First, thanks for your answer.
> I find the description of mvpart package quite short:
>
> "Wrapper function for fitting and plotting rpart models "

I agree that this could be improved.

> Then, I thought that mvpart was the package which seems closest to what I
> would like to do (i.e. oblique decision tree also named multivariate
> decision tree). However, as you pointed out, it is the answer which is
> multivariate in the mvpart package.

Terminology for different flavors of decision trees and recursive 
partitioning varies a lot between different communities (machine learning, 
artificial intelligence, statistical learning, ...) and even within 
communities.

>> Thus, this is for multivariate *responses*, not for multivariate splits.
>
> So, to your knowledge, there is no R package for multivariate splits (i.e.
> oblique decision trees)?

No, to the best of my knowledge, the list in the Machine Learning task 
view is up to date.

Also, I think that "oblique decision trees" is specific to the algorithm 
underlying the OC1 implementation.
   http://www.cs.jhu.edu/~salzberg/announce-oc1.html
The term trees with "functional splits" or "multiway splits" is more 
common, I guess. Binaries for the original implementation of the CRUISE 
algorithm are available at
   http://www.stat.wisc.edu/~loh/cruise.html
We played around with interfacing the latter in R, but unfortunately that 
is not really finished, yet.

hth,
Z