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.


Messages - MB

Pages: [1]
1
General Discussion / Re: reference for Max-Propagation algorithm
« on: July 17, 2010, 01:08:45  »
Can you please confirm the specific reference for the algorithm Hugin uses?

Furthermore, is it possible to obtain the max configuration for a subset of nodes in the BN?

2
General Discussion / reference for Max-Propagation algorithm
« on: July 01, 2010, 20:49:11  »
Can you please provide a reference for the the max-propagation algorithm used by Hugin?
Thank you.

3
General Discussion / LIMID versus UID
« on: June 10, 2010, 02:57:29  »
How does a Limited Memory Influence Diagram differ from an Unconstrained Influence Diagram?

Is there a benefit of working with one over another?  

Thank you.

4
General Discussion / sampling algorithms?
« on: May 05, 2010, 01:59:10  »
Why doesn't Hugin offer sampling-based algorithms for performing inference?  Is there a particular reason only exact inference algorithms (which can be associated with intractably large memory demands) are implemented?
Thank you!

5
I am using a BN containing both continuous and discrete nodes.  I also have a version of the same BN using all discrete nodes (the Gaussian nodes are discretized).   I am using the 64-bit version of Hugin.

When I run the model with both continuous and discrete nodes, I receive an out of memory when memory demands reach ~10.5GB (watching the memory demands in the task manager).  When I use the the BN with strictly discrete nodes, I do not receive an out-of-memory error until Hugin has accessed over 50GB of memory (including virtual memory).

I have decided to make use of continuous nodes to obtain increases in computational efficiency (because we know that the junction tree algorithm is exponential in the size of the largest clique when we use discrete nodes and but cubic when we use continuous nodes).   However, when I saw the reduced performance associated with the hybrid BN, I was puzzled.  Is there any reason I should be seeing the result I described above?  If not, is there any common mistake I may be making?

Thank you for your assistance.

6
HUGIN GUI Discussion / Re: Table import guidance
« on: September 30, 2009, 19:35:07  »
Thank you for your quick reply. 
I will re-consider using the table generator by reducing samples. However, I am uncertain about the sampling scheme used.  Does Hugin sample from the actual parent distribution (i.e. does it re-normalize to account for the actual shape of the distribution over a particular state when a specific distribution is selected) or does it assume the distribution is uniform within each state?

I am using a 64-bit version of Hugin (for Windows), so unfortunately I don't think I can an API/Excel to input tables.

Thank you for your guidance on the form on the input files.  I was using tab/space rather than comma delimiter and that seemed to cause a problem.  Now that I have written the files as comma deliminated, I am able to import them into Hugin (though admittedly, it takes some time for the files to import due to their large size).

7
HUGIN GUI Discussion / Table import guidance
« on: September 29, 2009, 19:47:37  »
I have generated CPTs using an external program because the build-in table generators take a very long time.  I would like to import to CPTs into Hugin. 
I have saved the files as .txt files however, when I attempt to import the files into Hugin, I receive the following error: "Unable to import table from file. The file contains 0 lines (should be 10) with number of values (big number)"  I am certain the file is not empty and that it contains the correct number of values. 
Is there any guidance available on the format the .txt files should take?  I have tried to export a table and use the same format for the imported file, but with no success.
Any guidance would be greatly appreciated.
Thank you.
-MB

8
General Discussion / Is it possible to allocate more memory?
« on: September 22, 2009, 01:10:30  »
I am working with a 64-bit version of Hugin.  I have been running a few BNs recently and they have been taking a very long time to compile (even after I have found and saved an optimal elimination order).  However, when I look at the Windows task manager, very little of my RAM is being used.  Is there a way  for Hugin to utilize more RAM to reduce computation time in exchange for the calculations using more memory?
Thank you.

9
General Discussion / Propagation
« on: May 13, 2009, 21:35:13  »
Is there are way to have Hugin only perform propagation after evidence has been entered in the BN (instead of computing the unconditional case automatically)?  This is useful when a network becomes computationally simpler/more efficient when certain nodes have been observed. 
Thank you.

Pages: [1]