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 - Anders L Madsen

Pages: [1] 2 3 ... 20
1
General Discussion / Re: Parameter Selection and Cross-Validation
« on: August 02, 2023, 16:49:55  »
Hello Holly,

We do not have a tutorial showing the process for performing feature selection for a Bayesian network.

If we assume that you are considering a classification problem, then one approach is a wrapper search. Start with a model containing only the target variable and iteratively adding the feature which improves the score the most. Stop when there is no further improvement.

You can split the data into train, validate and test sets to measure the performance.

Best regards
Anders

2
General Discussion / Re: Parameter Learning
« on: June 17, 2023, 05:19:30  »
Hi Marco,
You can run the EM learning Wizard that has the option to reset all tables and all experience counts.
Best regards
Anders

3
HUGIN GUI Discussion / Re: Variance or Standard Deviation?
« on: March 10, 2023, 21:59:17  »
Hi Marco,

It should be the variance. Why do you have doubts?

best regards
Anders

4
HUGIN GUI Discussion / Re: Discrete vs Hybrid BN
« on: January 19, 2023, 18:40:36  »
Hello,

Here are some quick replies to your questions

  • It is easy to find VOI in the GUI however where and how I can see the posterior probability distribution for the target node?

    Each node has a monitor that displays the posterior probability distribution of the node. You need to compile the network, select the target node, right click on the target node and select toggle monitor window.
  • The VOI is found in discrete but not in hybrid networks. So my question what are the common metrics that can be evaluated for both types of networks?

    Value of information in the GUI does not support hybrid variables.
  • Also how are the sensitivity to findings and sensitivity to parameters evaluated in hybrid networks?

    In hybrid networks there is in many situations not a difference between a discrete observation and a continuous observation. Hence, the documentation applies to both situations. Parameter sensitivity analysis is not supported for the continuous parameters in hybrid networks

I hope this is useful
Anders

5
HUGIN GUI Discussion / Re: A More General Zoom Feature
« on: September 04, 2021, 16:23:14  »
Marco,

We are working on a new type of monitors that will support better zoom functionality.

best regards
Anders

6
HUGIN GUI Discussion / Re: Information Entropy Minmization Discretization
« on: September 04, 2021, 16:21:57  »
Hi Marco,

The IEM discretization method is based on the algorithm of the following paper:

U. M. Fayyad and K. B. Irani. 'Multi-Interval discretization of continuous valued attributes'. In Proceedings of the Thireenth International Joint Conference on Artificial Intelligence, pages 1022-1027, 1993.

If you have ideas for other (or better) discretization algorithms to implement, then please let us know.

Best regards
Anders

7
HUGIN GUI Discussion / Re: Plate Models
« on: March 30, 2021, 17:45:02  »
Hi Marco,

You can try to use the combined OOBN and DBN facility in HUGIN. Maybe it is sufficient for your needs? In the DBN you can define the number of time steps in the time window.

We have a YouTube tutorial on DBNs in HUGIN: https://www.youtube.com/watch?v=Go7O6-UfFWk

Hope this is useful
Best regards
Anders
 

8
RHUGIN / Problem running RHugin using latest version of HUGIN
« on: April 20, 2020, 10:40:31  »
The RHugin plugin (https://github.com/huginexpert/RHugin) searches for the HUGIN installation is a specific location. This location is no longer used to install HUGIN software.

Here is how to solve the issue on the Windows platform:

RHugin tries to locate the Hugin installation, but fails because the naming of the Hugin directory has changed.  Try the following workaround:
  • Start a “Command Prompt” with “Run as Administartor” privileges.
  • CD to the directory that contains the Hugin installation ("C:\Program Files\Hugin Expert").
  • Run this command: mklink /D "Hugin Researcher 8.8 (x64)" "HUGIN 8.8 (x64)"
     

Notice the Hugin/HUGIN distinction in the mklink-command.

Then follow the instructions for installing the RHugin package.

9
P(e) is the probability of the evidence. E.g., if you have entered X=x and Y=y (assuming discrete nodes), then P(e)=P(X=x,Y=y).

It is calculated as the normalization constant after propagation of evidence in the junction tree.

Anders

10
FAQ / Re: How can I use HUGIN in Matlab?
« on: February 06, 2020, 20:08:03  »
Due to an update in Matlab (version 2018a), the access to enums has changed. The above code should be updated to something like this:

Code: [Select]

ghapi = NET.addAssembly ('C:\Program Files\Hugin Expert\HUGIN 8.8 (x64)\HDE8.8CS\Lib\hugincs-8.8-2.0-x64.dll');
myEquilibirumEnumHandle = ghapi.AssemblyHandle.GetType('HAPI.Domain+Equilibrium');
myEvidenceModeEnumHandle2 = ghapi.AssemblyHandle.GetType('HAPI.Domain+EvidenceMode');

d = HAPI.Domain ('C:\Program Files\Hugin Expert\HUGIN 8.8 (x64)\Samples\ChestClinic.net', HAPI.DefaultClassParseListener);
L = d.GetNodeByName ('L');
d.Compile ();
L.GetBelief (0)
%ans =
%    0.0550
L.GetBelief (1)
%ans =
%    0.9450

X = d.GetNodeByName ('X');
X.SelectState (0);
d.Propagate (myEquilibirumEnumHandle.GetEnumValues().Get(1), myEvidenceModeEnumHandle2.GetEnumValues().Get(0));
L.GetBelief (0)
% %ans =
% %    0.4887
 L.GetBelief (1)
% %ans =
% %    0.5113

11
FAQ / CLI install documentation HUGIN 8.7
« on: October 18, 2019, 09:59:10  »
The following command-line is known to work (April 2019):

msiexec /i HuginDist87(x64).msi /L*v SilentInstallation.log /quiet PATH_LICENSE="C:\TMP\LICENSE-8.7"

A couple of words on the installation:
  • The PATH_LICENSE is of course the path to the location of the HUGIN License file.
  • The .MSI file does not perform a cryptographic check of the License Key - only a shallow sanity check. The deep check is performed every time HUGIN is launched. Therefore you must launch HUGIN to be sure that everything is working properly.
  • The .MSI file will write the license to the location: C:\ProgramData\Hugin Expert. If anything goes wrong, the file can be manually copied to this location.
   
   

12
HUGIN GUI Discussion / Re: Temporal Nodes
« on: October 08, 2019, 21:38:31  »
Dear Marco,

The number of time slices is set under "Network->Network Properties->DBN".

Hope this helps
Anders

13
Hello Raymond,

I assume that you by now have realised that "our" HUGIN is a tool for probabilistic graphical models and not images.

Sorry for the belated reply.

Anders

14
General Discussion / Re: Expressing Fault Tree Analysis in BBN
« on: June 06, 2018, 07:09:52  »
Hi Mel,

The "generalization" of OR / AND is MAX / MIN. If you encode false/correct as 0, true/failure as 1, and defective as 2 using Numbered nodes and use MIN/MAX instead of AND/OR, then you should be able to encode the desired CPTs.

The idea is that there must be an order on the states and this is enforced by the use of the Numbered node where 0 would reflect the absence of a cause and higher values reflect a more significant presence of the cause.

Let me know if this works for you.
Anders

15
FAQ / Re: How can I use HUGIN in Matlab?
« on: May 16, 2018, 10:00:09  »
The method
Code: [Select]
enterFindingis used for entering evidence on a node. There is also a selectState method to set the node to a specific state.

To change the CPT of the a node you need access the table for the node and manipulate its content. The documentation has a complete example building a network (here referring to the .NET API documentation):
http://download.hugin.com/webdocs/manuals/CS/html/d4648875-d41a-783b-d5f4-638df39ee413.htm#Example%202
Hope this helps

Pages: [1] 2 3 ... 20