Network instantiation which maximizes a state probability

[mh]

Hello,

Is there any tool / feature in Hugin for evaluating in which states the other network nodes (or a subset of them) than a certain node of interest should be in order to have the highest (or lowest) probability of a certain state of of this node of interest, and further what is then this maximum (or minimum) probability? As an example, let's assume the node of interest is describing whether there is a fire or not. Then I would like to examine in which states the other variables affecting the possibility of fire should be in order to have the highest probability of fire, and what that highest fire probability would be. For a small network of nodes with relatively few states this is of course possible by running the model and then manually trying each instantiation combination but for a larger network this becomes tedious.

Regards,
mh

[Frank Jensen]

The distribution of a node in a Bayesian network is completely determined by the nodes in the "Markov blanket" of the node (http://en.wikipedia.org/wiki/Markov_blanket).  This set is often small.

There is no special tool in Hugin to do the analysis you seek.

[Anders L Madsen]

If you want to evaluate the impact of observing a node on the distribution of a specific node, then you may use value of information. The HUGIN GUI has a Value of Information dialog that computes the pair-wise mutual information between a target node and each node in a set of selected nodes.

Also, the Evidence Sensitivity dialog allows you to determine the minimum and maximum probability observing a node will produce for a specific target node. This analysis can be performed using a set of selected nodes.

Both types of analysis will help you to determine the impact of individual nodes on the distribution of a target node.