In this section, we will cover some basic specialized probabilistic graph models that are very useful in different machine learning applications.
In Chapter 2, Practical Approach to Real-World Supervised Learning, we discussed the Naïve Bayes network, which makes the simplified assumption that all variables are independent of each other and only have dependency on the target or the class variable. This is the simplest Bayesian network derived or assumed from the dataset. As we saw in the previous sections, learning complex structures and parameters in Bayesian networks can be difficult or sometimes intractable. The tree augmented network or TAN (References [9]) can be considered somewhere in the middle, introducing constraints on how the trees are connected. TAN puts a constraint on features or variable relationships. A feature can have only one other feature as parent in addition to the target variable, as illustrated in the following figure:
Figure...