Release 6.6

On September 12th, 2005 - we have released new versions of the HUGIN Graphical User Interface (v6.6) and HUGIN Decision Engine (v6.4).

The main new features of this release are:

  • Improved data mining functionality of the HUGIN Graphical User Interface.

  • Value of Information analysis on discrete random variables in Bayesian networks and influence diagrams in both the HUGIN Decision Engine and the HUGIN Graphical User Interface.

  • Improved performance of the HUGIN Decision Engine with respect to inserting and propagating multiple pieces of hard evidence in the junction forest.

  • HUGIN Decision Engine libraries for Windows platforms are now provided for Visual Studio .NET 2003 (in addition to Visual Studio 6.0).

Hugin Graphical User Interface v6.6:

The Learning Wizard of the HUGIN Graphical User Interface has been improved with a number of new features:

  • The Data Analysis tool of the Learning Wizard shows a number of statistics on the data analyzed including an indication of the amount of missing data and the number of distinct values for each variable.

  • The Data Dependency window of the Learning Wizard reports the p-value of an edge when the edge is selected and model size when the edge is deselected.

  • The Structure Uncertainties window of the Learning Wizard reports the p-value of an edge when the edge is selected.

  • The log information generated as part of the learning process is available to the user for inspection after completing the Learning Wizard.

The EM Learning Wizard of the HUGIN Graphical User Interface has been improved with a number of new features:

  • The EM algorithm reports the log-likelihood of the model given data to the network log after each iteration.

  • The log information generated as part of the learning process is available to the user for inspection after completing the EM Learning Wizard.

A few features related to the learning functionality of the HUGIN Graphical User Interface have been included or improved:

  • Better display of the network log information and a separate usage log window.

  • Automatic documentation of models. This feature enables the user to produce HTML documentation of a network, including the descriptions included in the network.

  • A print preview has been included in the printing functionality.

  • Online help has been improved to support drag’N’drop.

The HUGIN Graphical User Interface has been improved with various new features. This includes:

  • A line specifying the data file used for structure learning has been included in the Structure Learning window.

  • A line specifying the data file used for parameter learning has been included in the EM Learning window.

  • It is now possible to add experience tables to all selected discrete chance nodes.

  • It is now possible to add fading tables to all selected discrete chance nodes.

  • An invocation of “generate table” is enforced on all nodes with a model prior to parameter learning using the EM Learning Wizard. This ensures that the conditional probability table of each node has been generated form the model and it will be shown in the Prior Distribution window of the EM Learning Wizard.

  • The explicit creation of experience tables for all discrete chance nodes has been disabled prior to parameter learning using the EM Learning Wizard. This implies that the user must create experience tables prior to using the EM Learning Wizard or in the Prior Distributions step of the EM Learning Wizard.

The HUGIN Graphical User Interface have been extended with support for value of information analysis on discrete random variables in Bayesian networks and influence diagrams. Value of information analysis is based on computing entropy and mutual information on nodes and pairs of nodes.

The functionality of the HUGIN Graphical User Interface supporting object-oriented models has been extended to include a few new features:

  • It is no longer a requirement that object-oriented models are saved with “.oobn” suffix (both .net and .oobn are supported).

  • Improved functionality for saving and loading object-oriented models with expanded / collapsed instance nodes. If an instance node is expanded when the model is saved, then the instance node will be expanded and placed at the same location when the model is loaded subsequently.

The HUGIN Graphical User Interface has been improved with various new features. This includes:

  • Functionality for loading and saving node elimination orders in the HUGIN Graphical User Interface. This is very useful when a good or even optimal triangulation of a network is available or has been identified.

  • Help facilities made available in the “Update HUGIN” window. Due to strict security policies some users have experienced problems using the “live update” functionality of the HUGIN Graphical User Interface. Help facilities have been included in order to clearly explain how to update the installation in case of problems performing the “live-update”.

  • Improved usage log information.

  • Probabilities and utilities are shown as inconsistent when switching from edit mode to run mode on a previously compiled domain when tables have been updated.

  • Using right mouse pop-up menu to change the subtype of a node to numbered creates a node with states starting at 0 … (not 1 as was previously the case).

  • The functionality for graph Layout now supports continuous chance nodes and instance nodes.

Finally, efforts have been put into improving the stability of the HUGIN Graphical User Interface.

HUGIN Decision Engine v6.4:

The HUGIN Decision Engine has been extended with the following features:

  • Functions for computing entropy and mutual information have been added. These functions are useful for value of information analysis.

  • Direct access to the pseudo-random number generator implemented in HUGIN Decision Engine is now provided through API functions.

  • The EM algorithm reports the log-likelihood of the model given data to the network log after each iteration.

  • Improved performance of the HUGIN Decision Engine with respect to inserting and propagating multiple pieces of hard evidence in the junction forest.

Finally, HUGIN Decision Engine libraries for Windows platforms are now provided for Visual Studio .NET 2003 (in addition to Visual Studio 6.0).