gini_index: compute Gini impurity for binary values only

In rfVarImpOOB: Unbiased Variable Importance for Random Forests

compute Gini impurity for binary values only

Description

simple function to compute simple or penalized Gini impurity

The "penalty" compares the class probabilities pHat with a reference estimate pEst

which would typically serve as a prediction (e.g. in a tree node).

Usage

gini_index(pHat, pEst = NULL, k = 2, kind = 1, w = 2)

Arguments

pHat	probabilities from the current data,
pEst	estimated class probabilities (typically from an earlier inbag estimation). Only pass if you intend to compute the "validation-penalized Gini"
k	exponent of penalty term: abs(pHat-pEst)^k
kind	kind of penalty
w	weights, default is 2 if you pass just a single probability instead of the vector (p,1-p)

Value

simple or penalized Gini impurity

Author(s)

Markus Loecher <[e-mail]>

Examples

#Test binary case:





gini_index(0.5,0.5,kind=1)


gini_index(0.9,0.1,kind=1)


gini_index(0.1,0.9,kind=1)





gini_index(0.5,0.5,kind=2)


gini_index(0.9,0.1,kind=2)


gini_index(0.1,0.9,kind=2)








gini_index(0.5,0.5,kind=3)


gini_index(0.9,0.1,kind=3)


gini_index(0.1,0.9,kind=3)

rfVarImpOOB documentation built on July 1, 2022, 5:05 p.m.