December 18, 2012: New release – HUGIN 7.7
Today we are releasing a new version of the HUGIN software (v7.7). The main new features of this release are the introduction of a new node type called the “discrete function” node type, and the new associated operation “aggregate”.
A discrete function node is similar to a function node. Whereas a function node represents a numerical value computed from the results of propagation in a network, a discrete function node represents a set of numerical values. That is, a discrete function node has a subtype, e.g., interval or labelled, and a set of states. A discrete function node allows the user to define mathematical expressions representing a distribution computed after a successful propagation in a network. The new operator “aggregate” can be used to aggregate two distributions (representing frequency and severity) into a single distribution. The expression is defined using the Table Generator.
It is possible to use function nodes and, for instance, the “probability” operator to connect (otherwise disconnected) sub-graphs of chance nodes. This means that a probability computed in one network can be transferred to another network.
The development of the functionality concerning the discrete function node type, as well as the aggregate and probability-transfer operators has been sponsored by the research project “Operational risk in banking and finance”. The project is dedicated to strengthening management of operational risk in the banking and finance sector, including the development of Basel II compliant operational risk measurement and management tools in accordance with the Advanced Measurement Approach (AMA). The project is financed by the University of Stavanger and a consortium of Norwegian banks consisting of Sparebank 1 SR-Bank, Sparebank 1 SNN, Sparebank 1 SMN, Sparebanken Hedmark and Sparebank 1 Oslo and Akershus.
HUGIN Graphical User Interface v7.7
The HUGIN Graphical User Interface has been improved with various newfeatures. These include:
- The discrete function node type as well as the “aggregate” and “probability” operators are supported by the HUGIN GUI.
- The data preprocessor for manipulating raw data in the Learning and EM Learning Wizards has been improved with new features. This includes the possibility to delete columns and entropy-based discretization for supervised discretization.
- The HUGIN GUI now supports simulation in a LIMID under the current strategy.
- It is now possible to define attributes on Domain objects in the HUGIN GUI.
- A Utility Sensitivity Analysis Dialog has been added to the HUGIN GUI. This allows the user to perform sensitivity analysis on the utility parameters of a LIMID. The sensitivity analysis is solving the LIMID for a set of user defined parameter values.
- The Code Wizard has been extended and improved with new functionality. This has improved the performance of the Code Wizard.
- The performance of the “propagate ALL rows” in the DataSet Dialog has been improved significantly.
- The d-separation code in the HUGIN GUI has been replaced to improve performance.
- The mean and variance calculations for numeric nodes have been improved.
- It is possible to inspect the Table Generator expression for a node in Run-Mode using CTRL+right-click.
- The Node States Generator has been extended to support copying states between nodes of different category and identical sub-types (e.g., to copy states between a labelled discrete chance node and a labelled discrete decision node).
- Other minor improvements.
Finally, efforts have been made into improving the performance of the HUGIN Graphical User Interface.
HUGIN Decision Engine v7.7
The HUGIN Decision Engine has been extended with the following features:
- A new node type has been introduced: The “discrete function” node type. This node type is a combination of the “discrete” and “function” node types. The function nodes implemented in previous versions of the HUGIN API are now referred to as “real-valued function” nodes.
- A new operator to express “aggregate distributions” has been introduced. An aggregate distribution is the distribution of the sum of a random number of independent identically distributed random variables. This operator can only be used in models for discrete function nodes.
- A new operator to express a probability computed from the beliefs of a discrete node has been introduced. This operator can only be used in models for function nodes.
- More general network topologies can now be specified. It is now possible to have a directed path containing “functional links” between two nodes that are not real-valued function nodes. This feature is essential in order to use the new expression operators.
- Two new parameters have been introduced for the total-weight triangulation method. These parameters can be used to produce better triangulations as well as speed up the triangulation process.
- An “initial triangulation” can be specified: If a good triangulation is already known, then this can be used to speed up the search for a better triangulation.
- The search for minimal separators can be sped up by discarding separators larger than a prespecified size. This also makes it possible to handle larger prime components (without having to split them into smaller graphs).
- The NET language has been extended in order to support discrete function nodes and the new expression operators.
- HUGIN API libraries for the Windows platforms are now provided for Visual Studio 2012 (in addition to Visual Studio 6.0, Visual Studio .NET 2003, Visual Studio 2005, Visual Studio 2008, and Visual Studio 2010).
- The HUGIN Web Service API has been extended with new widget types for creating belief bars, doing vector graphic plotting and table editing.