Today - March 25th, 2009 - we have released a new version of the HUGIN software (v7.1).
The main new features of this release are:
Parameter estimation from data for continuous nodes using the Expectation Maximization (EM) algorithm. The HUGIN tool has been extended with functionality to perform parameter estimation for both discrete and continuous nodes in a Bayesian network. The parameter estimation is performed using the EM algorithm supporting missing values in data.
Two-way parameter sensitivity analysis has been included in the Parameter Sensitivity Wizard. The Parameter Sensitivity Wizard has been extended with support to perform two-way parameter sensitivity analysis on an output probability given a set of evidence. This allows the user to efficiently investigate the impact of co-varying two parameters on a posterior probability.
HUGIN Graphical User Interface v7.1:¶
The HUGIN Graphical User Interface has been improved with various new features. This includes:
Learning Wizard: Functionality for saving and loading data preprocessing definitions, e.g., the specification of a discretization operation or value replacements.
Learning Wizard: Discretization using Information Entropy Minimization (U. M. Fayyad and K. B. Irani. ‘Multi-Interval discretization of continuous valued attributes’).
Learning Wizard: Discretization of multiple numeric variables have been improved to become more user friendly.
Analysis Wizard: Functionality for exporting the results of a series of belief update operations across a set of cases to a CSV file.
Analysis Wizard: Functionality for generating a PDF report over the results of a series of belief update operations across a set of cases.
Parameter Sensitivity Wizard: Computing Sensitivity on multiple cases from a case file.
In addition to parallelizing table operations when compiling and propagating in run-mode. Table operations when learning and in various wizards may also be parallelized.
Added option for specifying a background image for the network panel.
Allowing the creation and parsing of invalid Expressions in the Expression Builder Wizard.
Other minor improvements.
Finally, efforts have been put into improving the performance of the HUGIN Graphical User Interface.
HUGIN Decision Engine v7.1:¶
The HUGIN Decision Engine has been extended with the following features:
The EM algorithm has been extended to learn CG distributions for continuous nodes.
The EM algorithm now permits learning to be enabled/disabled for specific parent configurations: A nonnegative experience count enables learning, while a negative experience count disables learning.
The EM algorithm now checks that the equilibrium is ‘sum’ with no evidence incorporated. As a “sanity” check, it also verifies that learning is enabled (that is, there must exist at least one node with a nonnegative experience count), and that case data has been specified.
Because the EM algorithm is controlled by experience tables, continuous nodes may now also be given experience tables.
The HKB format has been updated (in order to support parameter learning for continuous nodes).
The model scoring functions now check that the junction tree potentials are up-to-date with respect to the node tables and their models (if any).