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 - Frank Jensen

Pages: [1] 2 3 ... 6

I have checked the file and discovered that it is a tar-file containing all the other files in the "models-ch3.tar.gz" file.  That must be a mistake.  I will notify the authors of the book.

Best regards,

C# / Re: Expression for Noisy Or
« on: September 07, 2016, 21:28:03  »
The NoisyOR operator is described in section 6.7.2 of the Hugin API Reference Manual (the "api-manual.pdf" file).

I hope this helps.


C / Re: structure learning
« on: August 29, 2016, 16:26:02  »
The only thing missing from your code is the specification of the states of the nodes.  You need to specify the number of states and the labels (names) of the states, assuming labeled states.  If the states represent numbers or (numeric) intervals, then state values must be specified.

For example, in the "asia" example, the nodes are binary (2 states) with labels "yes" and "no".  The labels must match exactly the labels used in the data file (that is, "yes", "Yes", and "YES" are all diffent).

I hope this helps.


There are no limits on the numbers of parents and children.

What is the reason for this question?  Is it related to structure learning?

General Discussion / Re: Chain graphs
« on: October 06, 2015, 22:06:40  »

1. Hugin only supports chain graphs that are specified using NET files.  These files can be loaded, and evidence can be entered for the nodes.  The evidence can be propagated, and beliefs can be retrieved.  But not much more than that.  Many API functions will return a "usage"-error if the network is a chain graph.

2. The data is a list of numbers in row-major order, as mentioned by Example 13.12 in the API manual.  Note that you can omit the (nested) parentheses.

3. There is no support for non-trivial (i.e., with more than one node) chain graph components containing CG nodes.

I hope this helps.

Java / Re: Generate multiple simulations in the API
« on: September 23, 2015, 17:45:24  »

First: The Java API documentation basically only explains the Java interface.  For the full explanation, you also need to read the Hugin API Reference Manual.  This is the "api-manual.pdf" file in the Hugin software package.

Second: There are two different file formats for cases -- the "case file" format and the "data file" format.  The first type only contains a single case, but the second type can contain multiple cases (this is basically a restricted CSV file format).  "saveCase" saves the currently entered evidence as a "case file", and "saveCases" saves the currently stored cases within the Domain object as a "data file".  In the latter case, you must create (sample) the cases and store them in the Domain object, before they can be saved.  In the "test" subdirectory of the Hugin distribution, there is a program called "generate_cases.c", which does exactly what you want, but it is written in C.

I hope this helps.  If not, then please ask for more help.


I still don't understand.  All probabilities in a Bayesian network are point probabilities.

It is possible to express some uncertainty on a probability by adding a parent node representing the probability.  That parent node would typically be an interval node ranging from 0 to 1 and be given a distribution over that interval (such as a Beta distribution).

How would a Normal or a Log-Normal distribution be used for your purpose?

It isn't clear to me what you're trying to achieve.  Could you give some examples?

General Discussion / Re: Expressions
« on: December 18, 2014, 15:25:30  »
I don't think there is an elegant way to do this.

However, if we assume that "f(y|z)" can be expressed using Hugin expression syntax, and we know that Z assumes one of a finite number of states, say z1, z2, and z3, then you can write max_z{f(y|z)} as follows:

max (f(y|z1), f(y|z2), f(y|z3))

Clearly not pretty, but it should work.


General Discussion / Re: Expressions
« on: December 16, 2014, 23:01:06  »
Hi Therese,

You can use Normal, LogNormal, and Gamma in expressions, see section 6.7.1 in the Hugin API Reference Manual.

But I suspect that it is not as simple as that.  Can you elaborate on the "slightly complicated to compute" part of your post?


ActiveX / Re: Using invoke function in Matlab
« on: July 09, 2014, 19:18:27  »
You cannot set the subtype to "boolean" unless the node has exactly two states.  In other words, you must first set the number of states for the node.

General Discussion / Re: Trouble with using Hugin in Matlab
« on: April 01, 2014, 14:36:22  »
The library files (DLL files) are not included in the Hugin Explorer package.  You need to upgrade to the Hugin Developer package in order to get the library files.

Java / Re: Direct entry of data into tables
« on: March 12, 2014, 17:03:12  »
In the Java API, you must use a function to change the data in a CPT.  That function handles this notification.  (In C it is possible to use a pointer to change the data without the rest of the API knowing about it.)

I hope this helps.

Java / Re: NULL pointer in learnClassTables
« on: March 02, 2014, 01:03:47  »
Your code is correct.  However, the learnClassTables method incorrectly diagnosed that the domain was not a proper runtime domain for the class.  The problem will be fixed in the next release of the Hugin software.

Sorry for the inconvenience.

Java / Re: NULL pointer in learnClassTables
« on: February 15, 2014, 23:20:55  »
It's hard to guess what is wrong (all the usage conditions should be mentioned in the Hugin API Reference Manual).  The most obvious things to check are that the domain is an unmodified runtime domain constructed from some class, and that the equilibrium is sum-normal with all evidence incorporated (by calling "propagate" if necessary).

I hope this helps.


Pages: [1] 2 3 ... 6