Sensitivity To Findings
Of significant importance in Bayes net work is a measure of the independence between various nodes of the net. Using just the = 4 && typeof(BSPSPopupOnMouseOver) == 'function') BSPSPopupOnMouseOver(event);" class="BSSCPopup" onclick="BSSCPopup('X_PU_link_structure.htm');return false;">link structure and = 4 && typeof(BSPSPopupOnMouseOver) == 'function') BSPSPopupOnMouseOver(event);" class="BSSCPopup" onclick="BSSCPopup('X_PU_d_separation.htm');return false;">d-separation rules, you can determine which nodes are completely independent of which other ones (see Edit → Select Nodes → Info (D-) Connected), and how that changes as = 4 && typeof(BSPSPopupOnMouseOver) == 'function') BSPSPopupOnMouseOver(event);" class="BSSCPopup" onclick="BSSCPopup('X_PU_finding.htm');return false;">findings arrive. However, dependence is a matter of degree, and using Netica’s sensitivity functions you can efficiently determine how much a finding at one node will likely change the beliefs at another.
During diagnosis, you may wish to know which nodes will be the most informative in crystallizing the beliefs of the most probable fault nodes. Obviously, that will change as findings arrive, so it may need to be recomputed at each stage. In a net built for classification, you can determine which features are the most valuable for performing the classification (i.e. “feature selection”). In an information gathering environment, you can identify which are the most important questions to ask at each point (to provide information on the variables of interest), based on the answers to questions already received, so as to avoid asking unnecessary or irrelevant questions.
In real-world modeling, such as environmental modeling, you can determine which parts of the model most affect the variables of interest; thereby identifying which parts should be made the most carefully and accurately.
Select a node (called the "query node") and choose Network → Sensitivity to Findings from the menu. A report will be displayed in the = 4 && typeof(BSPSPopupOnMouseOver) == 'function') BSPSPopupOnMouseOver(event);" class="BSSCPopup" onclick="BSSCPopup('X_PU_Messages_window.htm');return false;">Messages Window displaying how much the beliefs, expected value, etc. of the query node would be influenced by a single finding at each of the other nodes (each is called a "varying node").
The first part of the report has a section for each varying node, showing how much it can effect the query node using several different sensitivity measures. The second part is a summary table which compares the sensitivities for each of the varying nodes.
If you want to limit the report to a few varying nodes, first select the query node, and then use ctrl-select to add the desired varying nodes to the selection. Then choose Network → Sensitivity to Findings. Currently this sensitivity analysis will only work for Bayes nets and not = 4 && typeof(BSPSPopupOnMouseOver) == 'function') BSPSPopupOnMouseOver(event);" class="BSSCPopup" onclick="BSSCPopup('X_PU_decision_nets.htm');return false;">decision networks (i.e. networks with decision nodes).
Here is an example use during diagnosis. Here is a description of the measures that Netica calculates.