Hi all,
I am modelling Hazard. Hazard is determined by the intensity of an event and the probability that the event will happen. Intensity and probability have been discretized such that each node has 3 states: low, moderate and high. The hazard is classified in the same classes (low, moderate, high) according to the following matrix:
Here is how i modelled it in GeNIe . However I am not sure how to populate the CPT as some of the probabilities of states are 0 instead of 1. Any suggestions?
The graphics do not show for some reason, so here are the direct links:
hazard matrix: https://photos.google.com/share/AF1QipM ... 1mVU5FV0tR
geNIe model: https://photos.google.com/share/AF1QipM ... 1mVU5FV0tR
How to deal with impossible states in conditional probabilities?

 Site Admin
 Posts: 271
 Joined: Tue Dec 11, 2007 4:24 pm
Re: How to deal with impossible states in conditional probabilities?
When answering your query, I would like to stress the idea that Bayesian networks are not about probabilities, they are about values. Any node in a Bayesian network is a variable, with well defined domain, whether it consists of a set of categories, discrete, or continuous values. Probabilities are used to describe the interactions among these variables. I would most of all try to think about variables and how these interact with one another. Intensity sounds like a variable to me but probability of risk sound rather like a probability distribution over various intensities, so you wodul deal with just one variable with values (intensities) and probabilities over them.
This having said, you may do something unconventional if this is the way researchers or practitioners think about risk, i.e., specify it in terms of two variables (intensity and risk), then the variable hazard looks to me like representing what you have in the first figure. When the intensity is high, the probability of high hazard is 1.0, when the intensity is medium and the return period low, probability looks like 0.5 to me (from the figure), etc. I hope this helps.
Marek
This having said, you may do something unconventional if this is the way researchers or practitioners think about risk, i.e., specify it in terms of two variables (intensity and risk), then the variable hazard looks to me like representing what you have in the first figure. When the intensity is high, the probability of high hazard is 1.0, when the intensity is medium and the return period low, probability looks like 0.5 to me (from the figure), etc. I hope this helps.
Marek