s-news
[Top] [All Lists]

Re: splitting criteria for regression trees

To: Jeff Hammel <jeff.hammel@cognigencorp.com>
Subject: Re: splitting criteria for regression trees
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Thu, 13 Dec 2001 18:55:48 +0000 (GMT)
Cc: "s-news (E-mail)" <s-news@lists.biostat.wustl.edu>
In-reply-to: <3C18D39E.5AB58381@cognigencorp.com>
On Thu, 13 Dec 2001, Jeff Hammel wrote:

> I'm doing some tree-based modeling in both S-plus
> version 6.0 for UNIX and S-plus 2000 in Windows
> 2000.  My understanding is that the node splitting
> is based on a reduction in deviance, which in the
> case of regression trees is equivalent to using
> the reduction in sum of squared deviations from
> the means.

true, although there is also an approach via impurity
indices, which is really what rpart implements (and
the defaulut Gini index for classification trees is one
example).

> Are there other choices available for
> the splitting criterion for regression trees
> either in the built-in S-plus tree functions or
> those in the rpart library?  If not, has anyone

Not visible, at least. (Surely it would be on the help page if there
were?)

> implemented his/her own splitting criteria?

The latest rpart allows users to plug their own criteria in. I had
understood Terry Therneau had sent the code to statlib fairly recently,
but I couldn't see it there.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


<Prev in Thread] Current Thread [Next in Thread>