In the help for rpart it says, "This differs from the tree function mainly in its handling of surrogate variables." And it says that an rpart object is a superset of a tree object. Both cite Brieman et al. 1984. Both call external code which looks like martian poetry to me. I've seen posts in the archives where BDR, and other knowledgeable folks, have said that rpart() is to be prefered over tree() Is there a simple reason why? They use the same fundamental algorithm. Are there differences in processing time? Bells and whistles? TIA, DRC
Difference between "tree" and "rpart"
2 messages · Dr Carbon, Brian Ripley
rpart does much more at C level, including pruning and cross-validation so can be much faster. It is also user-extensible. tree was actually written to track down bugs in the then S implementation, and so is much closer to the functionality in S. It is not where I would have started from. It is really only available for R to support MASS and PRNN (my books).
On Wed, 4 May 2005, Dr Carbon wrote:
In the help for rpart it says, "This differs from the tree function mainly in its handling of surrogate variables." And it says that an rpart object is a superset of a tree object. Both cite Brieman et al. 1984. Both call external code which looks like martian poetry to me. I've seen posts in the archives where BDR, and other knowledgeable folks, have said that rpart() is to be prefered over tree() Is there a simple reason why? They use the same fundamental algorithm. Are there differences in processing time? Bells and whistles?
Brian D. Ripley, ripley at 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 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595