Messages

1

Bayesian Networks and Influence Diagrams / What is new in the 2nd edition

« on: April 25, 2013, 19:57:04  »

I see that Springer is advertising a 2nd edition of the book. The downloadable front matter does not have a "What's New" or 2nd edition preface. Could you post some information as to what is new in the book? (I already have the first edition)


Thanks

2

HUGIN GUI Discussion / Re: Entering Likelihood as data

« on: July 19, 2012, 16:34:35  »

Thank you. I never realized that there were two different formats.

3

HUGIN GUI Discussion / Re: Entering Marginal Distributions as Evidence

« on: July 19, 2012, 16:25:40  »

Thanks Martin for this and your other replies.

To summarize my interest:
1. from time to time I get individual level (panel) data from market surveys/ studies, which after some form of centering, can be used to update/generate a bayes net. (I have to do all my testing by hand to control for "stratification"/"selection bias", via  Cochran-Mantel-Haenzel tests or a Quade's matched nonparamteric correlation)

2. When I get marginals back on a different study, there is an immediate interest in tuning the BN to match those marginals and get some information about the structure of that study (particularly what caused the primary outcome). This is where the soft evidence and IPFP gets used. At the moment I am doing this by hand, which is of course a pain.

3. Some of the variables never show up in individual form, but just show up as confidence intervals, means and standard deviations and so on for each strata. I can convert those to likelihood weights, using the usual soft discretization tricks (e.g. mixture weights for known kernals). Which leads to further need for likelihood evidence. 

4

HUGIN GUI Discussion / Entering Likelihood as data

« on: July 13, 2012, 17:12:23  »

The data files section of the Hugin manual seems to imply that data can only be exact:

     <Data> ::= <Value> | N/A | <Empty>
     <Value> ::= <State index> | <Label> | <Real number> | true | false

Is there any way to enter any of the various forms of  soft and likelihood/statistical evidence other than faking it with the "case count" field?

For example: I observe an observation that has where one of the variables is observed inexactly as being in state 1 with probability p₁ and state 2 with probability p₂, or if I am modeling something like genotypes and want to enter data based on phenotypes (e.g. observing a phenotype eliminates certain genotypes). I realize that the one dodge is to add another layer of observation or virtual nodes. I am trying to avoid that...

5

HUGIN GUI Discussion / Soft evidence vs likelihood evidence

« on: July 13, 2012, 15:21:05  »

The current interface (V7.6) allows the entry of likelihood based evidence, which is combined with the current state in the usual bayesian matter. Is it possible to do so-called "soft evidence" directly, setting a nodes distribution to a fixed state?

6

HUGIN GUI Discussion / Re: Entering Marginal Distributions as Evidence

« on: July 13, 2012, 15:18:01  »

Hello Anders,

Actually, one can fake IPFP by repeatedly entering likelihood evidence on the nodes of interest. For example, if I want to look at the Asia net where both bronchitis and "lung cancer" are set to 15%/85%, I would need to enter the likelihood for bronchitis then lung given bronchitis then bronchitis then lung, etc. till I get close enough to my target marginals.

This is of interest to me so that I can take a generic network which describes one population and retarget it towards another, related population, and was, as I read it, the original intent of the Deming/Stephans (1940)paper. This is the way I orginally learned of it in sampling and demography.

Is there any chance to get this automated in future versions of Hugin?

thanks

Bill



 

7

HUGIN GUI Discussion / Entering Marginal Distributions as Evidence

« on: May 22, 2012, 23:14:37  »

As I understand it, HUGIN allows us to enter single case data as evidence. A recent paper (Peng et.al., 2012, Internation Journal of Uncertainty Fuzziness and Knowledge Based Systems) make a passing reference to  the Madsen & Jensen(1999) as implementing IPFP for BN Junction Trees (to match marginals). Is that capability available in HUGIN? If so, how would I do that? I have tried entering likelihoods, but the second marginal entered as a likelihood will modify the first when it is propogated.

As an example, suppose I have a BN constructed on some historical data, and have observed the marginals for some of the variables in a new population. I would like to adapt the BN to match those marginals and see how that propogates through the network, test possible interventions in that new populations, etc.

Thanks

8

HUGIN GUI Discussion / Unable to complete graph layout using DOT

« on: October 25, 2011, 21:21:07  »

Hello,
I attempted to use DOT to layout a graph and recieved the mesage:
Unable to complete graph layout using DOT
String index out of range: -27
using internal algorithm

I am using HUGIN Developer 7.5 and Graphviz 2.28.
The path to the dot executable is C:\Program Files (x86)\Graphviz 2.28\bin\dot.exe

How do I fix this error?

Bill
added after initial post:
I tried this with a different network. Not only did it fail (same error, different index) it tried to send in a bug report. That failed, too with an i/o error.

9

General Discussion / Re: Degrees of freedom for CMI test

« on: January 24, 2011, 18:57:04  »

Thank you. I neglected the q term in my note.  In summary, then,  if I have an ordinal pair and wish to use an ordinal measure of association (such as Kendall's tau or Quade's partial form) I should do that offline and use the results to construct a constraints file. 

10

General Discussion / Degrees of freedom for CMI test

« on: January 20, 2011, 18:11:15  »

As I understand it, Hugin uses Conditional Mutual information tot test for lack of association between two nodes. If node_1 has r states and node_2 has s states, does the Chi-Square use (r-1)*(c-1) degrees of freedom?

When both nodes represent ordered states, the usual statistical test would be a Cochran Mantel Haenzel test with 1 degree of freedom, so that an (r-1)*(c-1) d.f. test is really conservative. If this is the case, is there any way to get the (approximately) correct test, other than by changing the significance and or changing the case weights?