untitled

14.3 Constructing Bayesian Networks 341

Summary

• The main use of BNs is for stochastic inference.


• BN inference is analogous to the process of logical inference and querying
  performed by rule engines.


• Bayes’ law is the foundation for BN inference.


• Evidence can be either hard observations with no uncertainty or uncertain
  observations specified by a probability distribution.


• Evidence can be given for any nodes, and any nodes can be queried.


• The nodes of a BN can be continuous random variables, but inference in
  this case is more complicated.


• BNs can be augmented with other kinds of nodes, and used for making
  decisions based on stochastic inference.

14.3 Constructing Bayesian Networks

We now consider the important question of how to construct BNs. While

there are many tools for performing inference in BNs, the methodology com-

monly employed for developing BNs is rudimentary. A typical methodology

looks something like this:

1. Select the important variables.


2. Specify the dependencies.


3. Specify the CPDs.


4. Evaluate.


5. Iterate over the steps above.
   This simple methodology will work for relatively small BNs, but it does
   not scale up to the larger BNs that are now being developed. The following
   are some of the development techniques that can be used as part of the pro-
   cess of constructing BNs, and each of them is discussed in more detail in its
   own section:


6. Requirements. Without clearly stated requirements, it is difficult to de-
   termine whether a BN has been successfully developed.