theoreticalecology / reprtree

Representative trees from ensembles

reprtree is a package to implement the concept of representative trees from ensembles of tree-based machines introduced by Banerjee, et al, 2012.

reprtree currently implements the d2 metric (closeness based on prediction) from Banerjee, et al, using either euclidean distance (for numeric predictions) or the percentage mismatch (for multi-class classification). The package implements representative trees for random forests currently, and will be extended to bagged trees.

Plotting representative trees is based on the tree and plotrix packages. Other nicer visualizations will be developed.

Reference: Banerjee M., Ding Y., Noone A.M (2012). Identifying Representative Trees from Ensembles. Statistics in Medicine.