Kansas State University is pleased to announce the release of Bayesian Network Tools (BNT), an open-source software development toolkit for research in probabilistic reasoning. This toolkit consists of several utility classes whose collective purpose is to manipulate graphical models of probability.

BNT is governed by the GNU Public License.

The current version, 1.0 alpha (04 May 2002 build), contains the following modules for working with Bayesian networks with discrete chance nodes:

• Bayesian network core classes that model the structure of a Bayesian network
• XML Bayesian Network Interchange Format (XBN) converter that converts networks of other formats to and from XBN (1.0 beta)
• Graphical editor for interchange format
• K2 algorithm for learning the structure of a Bayesian network
• Lauritzen-Spiegelhalter (LS) algorithm for exact inference on a Bayesian network
• Importance sampling algorithms evaluated using exact root mean squared error
• Forward sampling with likelihood weighting
• Logic sampling
• Self-importance sampling
• Adaptive importance sampling
• GASLEAK, a research genetic algorithm (GA)-based wrapper that searches for inferential loss-minimizing orderings of variables in Bayesian network structure learning

Because this is the first public version of BNT, we anticipate many types of inquiries. All comments and feedback, incluidng usage questions and error reports, are welcome. We especially encourage suggestions on improvements and desired features and invite participation by developers of new modules.

If you would like to be part of any discussions relating to the future development of this toolkit, please feel free to join our Yahoo! Group, the BNT Development Team.