
The Discrete Case: Multinomial Bayesian Networks 11
1.4 Estimating the Parameters: Conditional Probability
Tables
For the hypothetical survey described in this chapter, we have assumed to
know both the DAG and the parameters of the local distributions defining the
BN. In this scenario, BNs are used as expert systems, because they formalise
the knowledge possessed by one or more experts in the relevant fields. However,
in most cases the parameters of the local distributions will be estimated (or
learned) from an observed sample. Typically, the data will be stored in a text
file we can import with read.table,
> survey <- read.table("survey.txt", header =