conditionalPred: Evaluates interaction importance using conditional prediction

In karlkumbier/iRF2.0: Iterative Random Forests

Description

Computes the prediction accuracy of a fitted Random Forest using only the leaf nodes for which the specified interaction falls on the decision path. For classification, accuracy is measured by AUROC. For regression, accuracy is measured by decrease in variance.

Usage

  conditionalPred(rfobj, rd.forest, x, y, ints,
                  varnames.group=NULL,
                  n.cores=1,
  )

Arguments

rfobj	Fitted randomForest object
rd.forest	list as returned by readForest
x	numeric matrix of predictors
y	response vector
ints	a character vector specifying interactions, features separated by '_', as returned by iRF
n.perms	number of times to permute data matrix
varnames.grp	If features can be grouped based on their demographics or correlation patterns, use the group of features or “hyper-feature”s to conduct random intersection trees
n.cores	number of cores to parallelize over

Value

A numeric vector of the same length as ints giving the conditional prediction accuracy for each interaction

See Also

randomForest

karlkumbier/iRF2.0 documentation built on Sept. 9, 2021, 11:05 a.m.