ordinalForest (version 2.0)

ordinalForest-package: Ordinal Forests: Prediction and Variable Ranking with Ordinal Target Variables

Description

Ordinal forests (OF) are a method for ordinal regression with high-dimensional and low-dimensional data that is able to predict the values of the ordinal target variable for new observations based on a training dataset. Using a (permutation-based) variable importance measure it is moreover possible to rank the covariates with respect to their importances in the prediction of the values of the ordinal target variable. OF will be presented in an upcoming technical report by Hornung et al..

Details

For a detailed description of OF see: ordfor

The main functions are: ordfor (construction of OF) and predict.ordfor (prediction of the target variable values of new observations).

NOTE: ordinalForest uses R code and C++ code from the R package ranger for the involved regression forests. ordinalForest does, however, not depend on ranger or import ranger, because it was necessary to copy the C++ code and parts of the R code from ranger to ordinalForest instead. The reason for this is that ranger's C++ code had to be altered in order to calculate a suitable permutation variable importance measure for ordinal forests.

Examples

# NOT RUN {
# Illustration of the key functionalities of the package:
##########################################################

# Load example dataset:

data(hearth)

# Inspect the data:
table(hearth$Class)
dim(hearth)

head(hearth) 


# Split into training dataset and test dataset:

set.seed(123)
trainind <- sort(sample(1:nrow(hearth), size=floor(nrow(hearth)*(2/3))))
testind <- setdiff(1:nrow(hearth), trainind)

datatrain <- hearth[trainind,]
datatest <- hearth[testind,]


# Construct OF using the training dataset:

ordforres <- ordfor(depvar="Class", data=datatrain, nsets=1000, ntreeperdiv=100, 
  ntreefinal=5000, perfmeasure = "equal")
ordforres

# Study variable importance values:
sort(ordforres$varimp, decreasing=TRUE)

# Take a closer look at the top variables:
boxplot(datatrain$oldpeak ~ datatrain$Class, horizontal=TRUE)
fisher.test(table(datatrain$exang, datatrain$Class))

# Predict values of the ordinal target variable in the test dataset:

preds <- predict(ordforres, newdata=datatest)
preds

# Compare predicted values with true values:
table(data.frame(true_values=datatest$Class, predictions=preds$ypred))
# }
# NOT RUN {
# }