Data Mining: Practical Machine Learning Tools and Techniques, Second Edition

with all probabilities conditioned on the class value. This is because the search

algorithm defaults to K2 with the maximum number of parents of a node set

to one. Reconfiguring this to three by clicking on K2in the configuration panel

yields the more interesting network in Figure 10.18(b). Clicking on a node

shows its probability distribution—Figure 10.18(c) is obtained by clicking on

the windynode in Figure 10.18(b).

Trees

Of the tree classifiers in Table 10.5 we have already seen how to use J4.8, which

reimplements C4.5 (Section 6.1). To see the options, click the line beside the

Choosebutton in Figure 10.4(b) to bring up the object editor in Figure 10.19.

You can build a binary tree instead of one with multiway branches. You can set

406 CHAPTER 10 | THE EXPLORER

(a)

(c)

(b)

Figure 10.18Visualizing a Bayesian network for the weather data (nominal version): (a)

default output, (b) a version with the maximum number of parents set to 3 in the search

algorithm, and (c) probability distribution table for the windynode in (b).