Messages

1

General Discussion / arbitrary joint distribution for compiled network

« on: April 22, 2013, 08:03:49  »

Suppose I want to generate a joint distribution over some arbitrary set (or subset) of network model variables, e.g., Pr(X, Y, Z), where X, Y, Z are nodes/variables in the model.  Can I do this using the compiled model, e.g., using the simulation function?  Or some other way?

andersrm

2

HUGIN GUI Discussion / Using new Function Node

« on: February 23, 2013, 03:44:06  »

I would like to compute a utility based on the difference between two discrete chance interval nodes.  My understanding is that this CANNOT be done with real-valued function nodes at present since these cannot accept interval nodes as parents.  Any other suggestions on how to do this?  Other than representing the difference as a discrete chance interval node and assigning a discrete utility to each of its states..  Even if I could use a function node, and wanted to assign a utility node as a child, that utility node I suppose could only handle discrete values...

3

General Discussion / Interpreting Evidence

« on: January 14, 2013, 10:35:34  »

I have a Bayesian network containing a continuous variable x discretized into a number of possible states.  I obtain evidence/an observation that gives the average value of x. No other information is available about the evidence.  Is it reasonable and defensible to model the observation as likelihood (or soft) evidence?  I realize that without additional information about the evidence, there's still an infinite number of ways to interpret it.  Any thoughts and/or references to handling this issue (in Hugin or in general) would be appreciated.

andersm

4

General Discussion / Discretization Algorithm

« on: January 12, 2013, 22:47:36  »

Where can I obtain information/description of the discretization algorithm used by Hugin?

andersrm

5

General Discussion / Re: Calculation of average and variance for compiled nodes

« on: January 12, 2013, 20:57:08  »

Some follow-up questions:

1. With respect to calculating the mean, was I correct to assume that the mean for a node is always calculated by placing the probability mass at the center of the range?

2. How are the mean and variance calculations done for variables that have open-ended ranges, e.g., (0-5,5-Inf) ?

3. Is there any general guidance on interpreting mean and variance in model results?

andersrm

6

General Discussion / Bug in 64 bit V. 7.7 ???

« on: January 02, 2013, 22:27:08  »

I double click on a decision variable and a bug report dialogue comes up, requesting a report.  After filing the report, I cannot access the "states" tab in the variable dialogue.  Don't recall this happening on V. 7.6.

andersrm

7

HUGIN GUI Discussion / Re: Entering Likelihood as data

« on: November 21, 2012, 07:48:11  »

Yes, but as far as I can tell from looking at the online help manual, there are no obvious instructions for how you would use the case file format to enter likelihood data.  What am I missing?

andersrm

8

HUGIN GUI Discussion / Re: Most Likely Configuration

« on: November 21, 2012, 07:35:04  »

max propagate normal is grayed out.  And, I get an error when I use the wizard to go to Joint Analysis.  For "variables" I enter "No group", because I want to calculate the MPE over the entire network; but again I get an error, "invalid argument passed to a function". 

I have decision nodes in my network and I think the MPE function cannot be done to such a network.  One fix is to make the decision nodes chance nodes for purposes of this calculation.  Any reason one would recommend against that?

andersrm

9

HUGIN GUI Discussion / Most Likely Configuration

« on: November 18, 2012, 04:38:51  »

How does one determine most likely configuration (or most probable explanation) using the Hugin GUI?  Is it even possible?  It is mentioned in the Help page on Junction Tree, but there is no explanation that I could discover on how to do it.

andersrm

10

General Discussion / Re: Calculation of average and variance for compiled nodes

« on: November 09, 2012, 02:04:28  »

Somewhat.  I understand that (b-a)(b-a)/12 is the variance of a uniform random variable defined over the range (b-a).  So 750*750/12 is the added variance for one of the 2 intervals?  What about the second? 

11

General Discussion / Calculation of average and variance for compiled nodes

« on: November 07, 2012, 22:54:32  »

I have a compiled node output with interval nodes for a 2 state node, 0-750, 750-1500 with computed probabilities 0.9975, 0.0025.  I agree with the calculation 376.89 for the mean, but don't know how they came up with the variance of 48,287.  I got 1,402, assuming the midpoints of the ranges in the calculation.  How does Hugin calculate the variance?

12

General Discussion / Display of Hugin results for dynamic bayesian network

« on: October 15, 2012, 23:27:10  »

Anyone know of good examples of how to display results for a dynamic bayesian network for multiple variables over multiple time steps in a way that is clear, compelling, and succinct?

Richard

13

General Discussion / Re: Extracting the value of a state from a node

« on: October 12, 2012, 01:13:38  »

Frustrating.  I still can't get the file to upload.  Anders L Madsen, if there's another way to get the file to you/to the Forum, please let me know. 

Thanks!

 

14

General Discussion / Re: Extracting the value of a state from a node

« on: October 12, 2012, 01:06:23  »

I've attached an excerpt from the model.  Referencing my previous post:

1. the size variable (e.g., see size2) tracks population size;

2. I have not yet defined the threshold variable that will track the adults; and

3. the number variable (e.g., see number3) tracks the total number of animals in the reef population.  The harvest variable tracks the decision whether to harvest.  Note that currently harvest mortality is defined as a fixed proportion (0.1) of the population that is removed when the decision is to dredge and natural mortality is "low" (i.e., 0-0.5).  This needs to be changed to reflect the actual proportion of adults in the population.  Hence, the approach I've outlined.  I think you may be correct in that if I can define a variable as in step 2 above, I don't need the function variable, and thus the issue of the restriction on its usage does not occur.  I'm thinking "out loud" here...make sense, or am I missing something...?

Richard

P.S.  I"m having trouble getting the file to upload.  It's only 35K, but I'm getting a message saying the uploader is full.

15

General Discussion / Re: Extracting the value of a state from a node

« on: October 10, 2012, 00:01:31  »

I may be up against the limitations of the probabilistic graphical model vs. a more traditional systems dynamics model.

1. I would like to define a variable that computes the size of a population of oysters on a reef as part of dynamic probabilistic network.  At each time step, growth over the previous time interval is added to the previous size to equal the new size. 

2. I would also like a variable that computes the probability that the size exceeds a fixed size threshold (i.e., become adults). 

3. I would like to interpret this probability as a proportion of the population and use it as part of an expression in a third variable.  In this third variable, the number of animals on the reef is being tracked.  In this third variable, I would like to use an expression in which the the adults are harvested/removed from the population.   

Does this make sense?

Richard