GiniImportanceForest: computes inbag and OOB Gini importance averaged over all...

In markusloecher/rfVarImpOOB: Unbiased Variable Importance for Random Forests

Description

workhorse function of this package

Usage

GiniImportanceForest(RF, data, ylabel = "Survived", 


    zeroLeaf = TRUE, agg = c("mean", "median", "none")[1], 


    score = c("PMDI21", "MDI", "MDA", "MIA")[1], Predictor = mean, 


    correctBias = c(inbag = TRUE, outbag = TRUE), ImpTypes = 0:5, 


    verbose = 0)

Arguments

RF	object returned by call to randomForest()
data	data which was used to train the RF. NOTE: assumes setting of inbag=TRUE while training
ylabel	name of dependent variable
zeroLeaf	if TRUE discard the information gain due to splits resulting in n=1
agg	method of aggregating importance scores across trees. If "none" return the raw arrays (for debugging)
score	scoring method:MDI=mean decrease impurity (Gini),MDA=mean decrease accuracy (permutation),MIA=mean increase accuracy
Predictor	function to estimate node prediction, such as Mode or mean or median. Alternatively, pass an array of numbers as replacement for the yHat column of tree
correctBias	multiply by n/(n-1) for sample variance correction!
ImpTypes	which scores should be computed
verbose	level of verbosity

Value

matrix with variable importance scores and their stdevs

Author(s)

Markus Loecher <[e-mail]>

Examples

data("titanic_train", package = "rfVarImpOOB",  envir = environment())


set.seed(123)


ranRows=sample(nrow(titanic_train), 300)


data=titanic_train[ranRows,]





RF = randomForest::randomForest(formula = Survived ~ Sex + Pclass + PassengerId,


                                data=data,


                                ntree=5,importance=TRUE,


                                mtry=3,keep.inbag=TRUE, 


                                nodesize = 20)


data$Survived = as.numeric(data$Survived)-1


VI_Titanic = GiniImportanceForest(RF, data,ylab="Survived")

markusloecher/rfVarImpOOB documentation built on July 5, 2020, 6:50 p.m.