sensitivity analysis

[mc]

I am a GeNIe new user. I built a network and now I would like to do the sensitivity analysis using the sensitivity tool. I tried it but I didn't understand how it works.
I found the Helps information too general instead the reference paper too difficult (sorry, I am not a statistician ).
Could someone help me?

[marek [BayesFusion]]

I am a GeNIe new user. I built a network and now I would like to do the sensitivity analysis using the sensitivity tool. I tried it but I didn't understand how it works.
I found the Helps information too general instead the reference paper too difficult (sorry, I am not a statistician ).
Could someone help me?

The numbers that you will find in the tables for the sensitivity mode mean essentially gradients (i.e., first derivatives) of the function of the posterior probability of the targets over the parameter in question. If there are no targets, then the algorithm takes the average, I believe. I hope this helps. Cheers,

Marek

[mc]

The numbers that you will find in the tables for the sensitivity mode mean essentially gradients (i.e., first derivatives) of the function of the posterior probability of the targets over the parameter in question. If there are no targets, then the algorithm takes the average, I believe. I hope this helps. Cheers,

Marek[/quote]

Therefore the sensitivity tool is kind of one-way sensitivity analysis were each parameter is variated separately, is it correct?
For 'tha tables for sensitivity mode' you mean Max Min Avg in the balloon containing information about Sensitivity?
Tanks.
mc

[marek [BayesFusion]]

Therefore the sensitivity tool is kind of one-way sensitivity analysis were each parameter is variated separately, is it correct?
For 'tha tables for sensitivity mode' you mean Max Min Avg in the balloon containing information about Sensitivity?

This is correct -- it is one-way sensitivity analysis and we are displaying a maximum or average and all taken over the target nodes.
Cheers,

Marek

[mc]

What is exactly a 'target node'? Is simply the variable you are interested in or is necessary that it has specific properties (e.g. could be performed a diagnosis test on this variable)?

[marek [BayesFusion]]

What is exactly a 'target node'? Is simply the variable you are interested in or is necessary that it has specific properties (e.g. could be performed a diagnosis test on this variable)?

Well, it depends where it is -- have used this name somewhat inconsistently . In case of sensitivity analysis, we mean the nodes that are marked as target (they get a "target" icon on them once you set them as targets). There is another target, one that is accessible through diagnostic extensions (main tab on the properties dialog). These targets (a.k.a. "faults") are used in computation of value of information and in the diagnosis dialogs. I hope this helps.
Cheers,

Marek