NBTree in R

November 26, 2014

NBTree is a hybrid between Naive Bayes and decision trees. The model, roughly speaking, is a tree with feature tests separating branches and Naive Bayes models on its leaves. Recently, I stumbled upon the need to implement a data analysis method employing this model, but I couldn’t find a pure R package containing NBTree.

I found out that the Java library Weka contains this hybrid model, and Weka can be plugged into R via rWeka. In spite of seeming victory, I soon realized that NBTree didn’t come right out of the box and I needed to undergo some additional steps to get it working. Here I have documented the steps just in case anyone else is in need of this model.

Step 1: Installing and loading R packages

Open up a R prompt and install RWeka along with its dependencies. You will have to have a working Java environment installed.



Step 2: Install add-on from Weka package manager

Weka comes with a package system and manager. We will have to use this package manager to install the NBTree add-on as shown below.

WPM("package-info", "repository", "naiveBayesTree")
WPM("install-package", "naiveBayesTree")

Step 3: Test that it was installed properly

You should see similar output as outlined below.

        If set, classifier is run in debug mode and may output
        additional info to the console
        If set, classifier capabilities are not checked before
        classifier is built (use with caution).

Step 4: Register classifier and bind to variable name

NBTree <- make_Weka_classifier("weka/classifiers/trees/NBTree")

Step 5: Test on Iris data set

Your R environment may be pre-loaded with the Iris data set. If such is the case, you can quickly test if the method is working.

fitted.model <- NBTree(Species ~ ., data=iris)

Sepal.Width <= 3.35
|   Sepal.Width <= 2.95
|   |   Petal.Length <= 4.75: NB 3
|   |   Petal.Length > 4.75: NB 4
|   Sepal.Width > 2.95
|   |   Sepal.Length <= 5.25: NB 6
|   |   Sepal.Length > 5.25: NB 7
Sepal.Width > 3.35
|   Sepal.Length <= 5.9: NB 9
|   Sepal.Length > 5.9: NB 10

Leaf number: 3 Naive Bayes Classifier

Installing other Weka add-ons

Following these instructions you can perhaps install other add-ons successfully as well. A listing of all add-ons can be queried using the package manager.

WPM("list-packages", "available")