Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Topics - lauchlin

Pages: [1]
Java / OOBN file format and continuous nodes
« on: August 14, 2012, 07:22:04  »
Hi, two questions for today.

Firstly, is there a plan to enable the reading of .oobn files?  I have an OOBN with a number of associated .oobn files that I have created using the GUI. At the moment I am flattening the network and saving it as a .net file so I can read it using the  DefaultClassParseListener() class but ideally I'd just like to read in the .oobn files.

Secondly, is there any work being done on enabling true continuous nodes in OOBNs rather than being restricted to interval nodes?



HUGIN GUI Discussion / Creating DBN using the GUI
« on: August 14, 2012, 07:17:14  »

I am currently working on some OO DBNs and at the moment am using the Java api to programmatically using a sliding window approach to roll out an n timestep DBN.  My colleague mentioned that you can use the GUI to do this but according to this post it is not possible.  Seeing as this post is from 2009 is it still the case that DBNs can only be done programmatically?


I'm in the process of converting a Netica model in to a OOBN and have been re-writing the Netica formulas to Hugin expressions. The problem I am running in to is that Netica and Hugin handle Normal distributions differently.  With Netica if you have a continuous nature node that has been discretized in to two intervals (e.g. 0-2 and 2-15) you can use Netica's NormalDist function to define a distribution across the bounded range.  Hugin on the other hand expects interval nodes to have a range of +/- infinity. To get around this I have created two extra ranges -inf-0 and 15-+inf and then created a child node that squashes the distribution using Distribution(0,1) and Distribution(1,0) back in to the two ranges effectively dumping everything less than 0 in to the 0-2 range and everything more than 15 in to the 2-15 range.  This ends up with the correct mean but the variance is out which is no good.  In the end it is a hack that only partially works and what I would really like to know is, what is the best way to use Normal distributions with interval nodes in the case that you have a node that represents a real world system that is bounded?

Pages: [1]