tidyRF: Tidy Random Forest

In tree.interpreter: Random Forest Prediction Decomposition and Feature Importance Measure

Description

Converts random forest objects from various libraries into a common structure, i.e. a “tidy” random forest, calculating absent auxiliary information on demand, in order to provide a uniform interface for other tree.interpreter functions. Note that the output is of a private format, and is subject to change.

Usage

tidyRF(rfobj, trainX, trainY)

Arguments

rfobj	A random forest object. Currently, supported libraries include randomForest and ranger.
trainX	A data frame. Train set features.
trainY	A data frame. Train set responses.

Value

A list of class tidyRF, with entries:

num.trees

An integer. Number of trees.

feature.names

A vector. Names of features.

num.classes

An integer. For classification trees, number of classes of the response. For regression trees, this field will always be 1.

class.names

A vector. Names of response classes.

inbag.counts

A list. For each tree, a vector of the number of times the observations are in-bag in the trees.

left.children

A list. For each tree, a vector of the left children of its nodes.

right.children

A list. For each tree, a vector of the right children of its nodes.

split.features

A list. For each tree, a vector of the indices of features used at its nodes. Indices start from 0. A value of 0 means the node is terminal. This does not cause ambiguity, because the root node will never be a child of other nodes.

split.values

A list. For each tree, a vector of the values of features used at its nodes.

node.sizes

A list. For each tree, a vector of the sizes of its nodes.

node.resp

A list. For each tree, a vector of the responses of its nodes.

delta.node.resp.left

A list. For each tree, a vector of the difference between the responses of the left children of its nodes and themselves.

delta.node.resp.right

A list. For each tree, a vector of the difference between the responses of the right children of its nodes and themselves.

Examples

library(ranger)
rfobj <- ranger(Species ~ ., iris, keep.inbag=TRUE)
tidy.RF <- tidyRF(rfobj, iris[, -5], iris[, 5])
str(tidy.RF, max.level=1)

tree.interpreter documentation built on March 26, 2020, 6:21 p.m.