Networks for continuous models
Franco Taroni, Alex Biedermann, Silvia Bozza, Paolo Garbolino, Colin G. G. Aitken · 2014
The variation of a random variable with a normal distribution is such that it may be represented by a probability density function which is unimodal, symmetric and bell shaped. Propagation in discrete networks is exact, as particular variables within the network become instantiated, and probabilities can be updated accordingly. Discrete distributions typically supported by Bayesian network software, such as Hugin, include the binomial, geometric, negative binomial and Poisson distributions. It is possible to represent a continuous entity as a discrete variable with states representing intervals for the continuous entity. When attempting to construct graphs to model a real-life system, it is often the case that both discrete and continuous variables will be involved within the system. Hence, it is important to consider the case of mixed graphs, in which both discrete and continuous nodes are of interest.