learning decision trees with one's own scoring functins
Please do study the packages you mention a great deal more carefully before posting such negative remarks about them. In particular, rpart is already fully user-extensible (and comes with a worked example), and both packages are supplied in source code on CRAN.
On Fri, 26 Aug 2005, zhihua li wrote:
Hi netters, I want to learn a decision tree from a series of instances (learning data). The packages tree or rpart can do this quite well, but the scoring functions (splitting criteria) are fixed in these packages, like gini or something. However, I'm going to use another scoring function. At first I wanna modify the R code of tree or rpart and put my own scoring function in. But it seems that tree and rpart perform the splitting procedure by calling external C functions, which I have no access to. So do I have to write R code from scratch to build the tree with my own scoring functions? It's a really tough task. Or r there other R packages that can do similar things with more flexible and extensible code? Thanks a lot!
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