Inference on hierarchical data

arovinel
Posts: 4
Joined: Mon Feb 01, 2016 5:38 pm
Reputation: 0

Inference on hierarchical data

Postby arovinel » Wed Feb 01, 2017 2:49 am

Hi there,
I'm utilizing a multiscale approach to identify failure mechanism in materials. The required networks for each scale are previously constructed from the data.
The idea is this:
at each hierarchy's level, the related BN is queried, and n-instance of the same set of m-variables is evaluated to determine the set exhibiting the highest posterior probability (PP). After this process is completed, if any PP exceeds a given threshold value (PP_th) , the next scale comes into place. Then, other p-instances of another set of q-variables, belonging to the next hierarchical level need to be processed and the q_i data exhibiting the highest PP. The same scheme is utilized until the lowest hierarchy level is reached.
Right now, the implementation is as follow:
    evaluate n x m set of variables from network_1, then send PP results to an external software which find the highest posterior
    If PP_th is exceed
      gather data at the next hierarchical level and send them to network_2
      evaluate the PP of the p x q set of variables and send the PP result to an external software which find the highest posterior
    else: do something else
this process is repeated until the lowest scale is reached.

Now my question is, can the aforementioned process be implemented in one BN?
If I think to a relational database (such as mySQL) this would be easy, because knowing the ids of the data (row identifiers in bayesialab), will allow retrieving data at the subsequent scale. In Bayesialab, where data are interpreted by rows, I can't find a way to process it in one single query, without increasing the number of variables (in my case would be 4*m*12*q) and multiply the number of required networks (n*m).

I know the process is convoluted, and I hope that my explanation is clear enough. If not I can create a small graphical example or I can post some picture representing my dataset.

Do you have nay suggestion?

Thanks in advance
Andrea Rovinelli
Purdue Unviersity

Mark
Posts: 8
Joined: Tue Jun 09, 2015 9:31 am
Reputation: 1

Re: Inference on hierarchical data

Postby Mark » Fri Feb 10, 2017 4:50 pm

Hi Andrea,

After thinking about your problem, you should try the API BayesiaEngine.
With it you could create several networks and make them interact with each others.
http://www.bayesia.com/bayesia-engine-api
And this is the javadoc:
http://library.bayesia.com/conflu/doc/
You can take a look at a simple example here:
http://library.bayesia.com/display/Blab ... 0d78f9f013

Don't hesitate to tell me if you have more questions.

Brownson
Posts: 1
Joined: Sat Nov 25, 2017 9:36 am
Reputation: 0

Re: Inference on hierarchical data

Postby Brownson » Tue Nov 28, 2017 8:39 am

Thanks Mark, I'm giving that a try now. I'll let you know how I get on.

Social Media

       

Return to “Analysis”

Who is online

Users browsing this forum: No registered users and 2 guests