Today we are releasing a new version of the HUGIN software (v7.5). The main new features of this release are new algorithms for learning the structure of a Bayesian network from a database of cases in the HUGIN Graphical User Interface (GUI) and a new HUGIN Web Service Application Programming Interface (API) to the HUGIN Decision Engine.

HUGIN Graphical User Interface 7.5

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

– The Learning Wizard now includes two new algorithms for learning the structure of a Bayesian network from a database of cases: a greedy search-and-score method and the Rebane & Pearl algorithm. The greedy search-and-score method constructs a Bayesian network by a greedy search through the space of possible network structures. It is possible to use either the BIC or AIC score to score candidate structures and to specify an upper limit on the number of parents of each node in the graph. The Pearl & Rebane algorithm constructs a poly-tree structure from the database of cases.

– The Learning Wizard has a new option to add no-links constraints to a selected set of nodes.

– The Analysis Wizard now supports the use of function nodes. This allows the user to process a large number of cases and evaluate the value of function nodes in each case.

– In the Sensitivity Set Graph of the Parameter Sensitivity Wizard it is now possible to open a table to see the sensitivity values associated with each node. In the Parameter Sensitivity Wizard there is also a new option for saving the sensitivity network panel as PDF.

– The Evidence Sensitivity Dialog has improved support for performing “what-if” analysis.

– The user now has the option to define her own language file.

– The Import Information function now enables the user to import state labels and values from a network specification file.

– When a node of a Class is selected in Run-mode, the status bar shows the entropy for discrete chance nodes, the entropy, mean and variance of discrete numeric chance nodes and the mean and variance of continuous chance nodes.

– The simulation of function nodes has beed added to the Generate Simulated Cases dialog.

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

HUGIN Decision Engine v7.5

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

– The performance of table operations involving discrete nodes only has been improved. This has led to improved performance of inference on many networks.

– Function and utility nodes can now be used as inputs and outputs of classes in OOBNs.

– A new HUGIN Web Service API is introduced including a set of pre-defined widgets to support easy deployment of HUGIN models using Javascript. Examples can be found at:
https://www.hugin.com/solutions/fraud-detection-management/online-demonstration
and
http://camvac.hugin.dk