Structure learning in Bayesian networks

In the previous sections, we considered that we already know the network structure and we tried to estimate the parameters of the network using the data. However, it is quite possible that we might neither know the network structure nor have the domain knowledge to construct the network. Hence, in this section, we will discuss constructing the model structure when the data is given.

Constructing the model from the data is a difficult problem. Let's take an example of tossing two coins and representing the outcome of the first with the variable, X, and the second with the variable, Y. We know that if the coins are fair, these two random variables should be independent of each other. However, to get this independence condition just from the data, we need to have all these outcomes equal number of times in the data that we will rarely see in real life.

So, in general, we need to make some assumptions about the dependencies. The assumptions that we make...