tests in diagnosis tool

[samoht]

Hello,

I haves several test T1, T2, ... and T1 measure not only one but two variables. Since each variable contributes to value of information I need for each variable a node. But the test consists of these two variables since if I execute the test automatically both variables are measured. So its really only one test and not one test for each variable.
How can I model it that on one side T1 is suggested as one test and on the other side this test consists of two variables?

Second Problem: for some of my test types it makes sense to execute it more than one time. E.g. a sensor (=Test) measures if a event has happened. You only has an estmate about the time the event will happen. So maybe you make one test and recognize that the event hasn't happened yet, but now you wamt to execute the test again after some time. The problem is the diagnose tool don't offer a test any more once it's been executed. So how can i model it with the diagnose tool?

best regards

[samoht]

Hello,

perfectly, I joined as a new user, post some questions and after 4 weeks still no reply at all.

Regards

[marek [BayesFusion]]

Sorry for the delay in answering -- it is entirely my fault. I was forwarded information about your query but did not answer under the pressure of other things.

I haves several test T1, T2, ... and T1 measure not only one but two variables. Since each variable contributes to value of information I need for each variable a node. But the test consists of these two variables since if I execute the test automatically both variables are measured. So its really only one test and not one test for each variable.
How can I model it that on one side T1 is suggested as one test and on the other side this test consists of two variables?

Perhaps I'm missing something obvious in your query but it seems to me that just having the test variable being a child of the two variables will do the job. The network will correctly compute the VOI (mutual entropy) bewteen the test variable and whatever variables you are pursuing. Have you tried this? If I'm misunderstanding your problem entirely, please try to explain it again.

Second Problem: for some of my test types it makes sense to execute it more than one time. E.g. a sensor (=Test) measures if a event has happened. You only has an estmate about the time the event will happen. So maybe you make one test and recognize that the event hasn't happened yet, but now you wamt to execute the test again after some time. The problem is the diagnose tool don't offer a test any more once it's been executed. So how can i model it with the diagnose tool?

The only solution I envision here for static BNs is having multiple instances of the test variable, each corresponding to the measurement that you plan to perform.
I hope this helps.
Cheers,

Marek

[samoht]

Hi Marek,

thank you again for your reply.
ok lets say I have a failure (=Hypothesis) which can be measured by the sensor types (=tests) A, B, C, ... Lets say each time i get a measurement of Sensortype A, it return me the varables temperature (in degree) and Pressure (in pascal). Pressure as well as temperature contribute to my VoI. But I cannot make an independent test for temperature, i always got a pair pressure/temperature with a new measurement. You suggested, if i understood right, to create a test node (=of type observation) for the sensor measurement which is dependent from a node pressure and a node temperature (=both of type auxiliary). The nodes Temperature and Pressure have intervals for temperature and pressure as node states. But what kind of states should this sensormeasurement node have, it alway consists of the two variables? And to compute the VoI I have P(Pressure|Failure) and P(Temperature|Failure) but not P(Sensormeasurement|Failure).

The only solution I envision here for static BNs is having multiple instances of the test variable, each corresponding to the measurement that you plan to perform.

But that does mean that I have to know in advance how often at maximum I want to execute a test and then e.g create n nodes for TestA?
So if the test should be able to be executed almost in continuous time is it better to use a dynamic BN? But I think a dynamic BN makes only sense if I have intertemporal dependencies like P(Testresult_at_t1|Testresult_at_t0) or not?

Best Regards Thomas

Sorry for the delay in answering -- it is entirely my fault. I was forwarded information about your query but did not answer under the pressure of other things.

I haves several test T1, T2, ... and T1 measure not only one but two variables. Since each variable contributes to value of information I need for each variable a node. But the test consists of these two variables since if I execute the test automatically both variables are measured. So its really only one test and not one test for each variable.
How can I model it that on one side T1 is suggested as one test and on the other side this test consists of two variables?

Perhaps I'm missing something obvious in your query but it seems to me that just having the test variable being a child of the two variables will do the job. The network will correctly compute the VOI (mutual entropy) bewteen the test variable and whatever variables you are pursuing. Have you tried this? If I'm misunderstanding your problem entirely, please try to explain it again.

Second Problem: for some of my test types it makes sense to execute it more than one time. E.g. a sensor (=Test) measures if a event has happened. You only has an estmate about the time the event will happen. So maybe you make one test and recognize that the event hasn't happened yet, but now you wamt to execute the test again after some time. The problem is the diagnose tool don't offer a test any more once it's been executed. So how can i model it with the diagnose tool?

The only solution I envision here for static BNs is having multiple instances of the test variable, each corresponding to the measurement that you plan to perform.
I hope this helps.
Cheers,

Marek