ABSTRACT
In this chapter, the authors discuss one of the main challenges encountered when building a Bayesian network (BN) and attempting to complete the node probability tables (NPT) in the BN. This is that the number of probability values needed from experts can be unfeasibly large, despite the best efforts to structure the model properly. The authors aim to use functions of varying sorts and give guidance on which functions work best in different situations. They show that for labeled nodes comparative expressions can be used, and also show that the authors can also use a range of Boolean functions including OR, AND, and special functions such as NoisyOR. They also show how to implement the notion of a weighted average. The authors describe a range of functions that can be used for ranked nodes. They also describe the challenges of eliciting from experts probabilities and functions that generate NPTs.
