PPtreeExt is an R package with extensons to the Projection Pursuit Tree (PPtree) algorithm to improve its performance in multi-class settings and under nonlinear separations. The PPtree classifier finds separations between classes based on linear combinations of variables by optimizing a projection pursuit index. One of its limitations is a rigid structure: the depth of a PPtree object is at most -1, where is the number of classes, with each class assigned to a single terminal node. The proposed modifications enhance predictive performance in multi-class contexts, particularly in situations involving outliers or asymmetries. The objective is to increase the classifier’s flexibility to handle more complex scenarios, while retaining interpretability. The package includes an interactive web application to explore the behavior of the original and modified PPtree classifiers under a variety of scenarios. This interactive tool played a key role in identifying limitations of the original algorithm and informing the design of the proposed modifications.
Installation
Install version from CRAN (not available yet!):
install.packages("PPTreeExt")Install the development version from GitHub:
# install.packages("devtools")
devtools::install_github("natydasilva/PPtreeExt")Simple Example
n <- nrow(iris)
tot <- c(1:n)
n.train <- round(n*0.9)
train <- sample(tot,n.train)
test <- tot[-train]
Tree.result <- PPTreeclass_MOD(formula = Species~.,data = iris[train,],PPmethod = "LDA")
Tree.result
=============================================================
Projection Pursuit Classification Tree result
=============================================================
1) root
2) proj1*X < cut1
4) proj2*X < cut2
6)* proj3*X < cut3 -> "virginica"
7) proj3*X >= cut3
8) proj4*X < cut4
10) proj5*X < cut5
12)* proj6*X < cut6 -> "versicolor"
13) proj6*X >= cut6
14)* proj7*X < cut7 -> "versicolor"
15) proj7*X >= cut7
16) proj8*X < cut8
18)* proj9*X < cut9 -> "virginica"
19) proj9*X >= cut9
20) proj10*X < cut10
22)* proj11*X < cut11 -> "versicolor"
23) proj11*X >= cut11
24) proj12*X < cut12
26)* proj13*X < cut13 -> "versicolor"
27)* proj13*X >= cut13 -> "versicolor"
25) proj12*X >= cut12
28)* proj14*X < cut14 -> "virginica"
29) proj14*X >= cut14
30)* proj15*X < cut15 -> "virginica"
31)* proj15*X >= cut15 -> "virginica"
21)* proj10*X >= cut10 -> "versicolor"
17)* proj8*X >= cut8 -> "versicolor"
11)* proj5*X >= cut5 -> "versicolor"
9) proj4*X >= cut4
32)* proj16*X < cut16 -> "setosa"
33) proj16*X >= cut16
34) proj17*X < cut17
36)* proj18*X < cut18 -> "versicolor"
37) proj18*X >= cut18
38)* proj19*X < cut19 -> "setosa"
39)* proj19*X >= cut19 -> "setosa"
35)* proj17*X >= cut17 -> "setosa"
5)* proj2*X >= cut2 -> "virginica"
3)* proj1*X >= cut1 -> "virginica"
Error rates of various cutoff values
-------------------------------------------------------------
[1] NA
PPclassify_MOD(Tree.result,test.data = iris[test,1:4], true.class = iris[test,5])
$predict.error
[1] 2
$predict.class
[1] "setosa" "setosa" "setosa" "setosa" "setosa" "setosa"
[7] "setosa" "setosa" "virginica" "versicolor" "versicolor" "setosa"
[13] "virginica" "virginica" "virginica"