On January 13th, 2003 - we have released new versions of the HUGIN Graphical User Interface (v6.2) and HUGIN Decision Engine (v6.0).
The main new features of this release are: support for object-oriented Bayesian networks and influence diagrams in the HUGIN Decision Engine, a new algorithm for learning the structure of a Bayesian network from a database of cases in the HUGIN Graphical User Interface, and various other improvements of the HUGIN Graphical User Interface.
The PC algorithm for learning the graphical structure of a Bayesian network from a database of cases has been extended with a necessary path condition. This condition is a necessary condition for the existence of a perfect map of the conditional dependence and independence statements derived by statistical tests. This has produced a new algorithm, which will be referred to as the NPC algorithm. The NPC algorithm has a number of advantages over the PC algorithm. For instance, the user now has the possibility to interact with the learning process in order to resolve structural uncertainties by making decisions on the presence or absence of uncertain edges. Both the PC and the NPC algorithm now supports user decisions on the directionality of edges, which cannot be directed using data alone. In the HUGIN Graphical User Interface it is now possible to learn arbitrarily complex structures, as tables for the conditional probability distributions are not created as part of the structural learning.
Various other aspects of the HUGIN Graphical User Interface have been improved: node tables, belief monitors, and operations on sets of nodes.
The node table has been greatly improved. For instance, expressions and manually specified tables can now coexist, cut-and-paste functionality has been improved (e.g., it is possible to cut and paste to/from Excel), tables can be exported to text files, it is possible to work with bars instead of numbers and to have numbers and bars simultaneously, the functionality for printing a table has been improved, and other new functionalities have been added to the node table.
It is possible to specify different display mode for the belief monitors in runmode. This facilitates viewing the changes in beliefs induced by a propagation of evidence.
It is now possible to perform an increased number of operations on sets of nodes (i.e., set type and set interface of sets of nodes simultaneously).
Finally, the HUGIN Decision Engine has been extended with support for object-oriented Bayesian networks and influence diagrams to make this functionality accessible through our Application Programming Interfaces (APIs). This implies that object-oriented Bayesian networks and influence diagrams are now supported both by the HUGIN Decision Engine and by the HUGIN Graphical User Interface.