BNB
The Bayes Net Builder Tool
The Bayes Net Builder Tool (BNB) is used to model the Bayesian Advisor project and the Welfare to Work project. The Bayesian Advisor project is concerned with automating decision making and constraint checking aspects of advising in academia. The goal of the Welfare to Work project is to build software to support Welfare to Work case managers' advising of clients. Both the projects are composed of several members working together on various aspects of the problem. The BNB tool builds the initial Bayesian Network(s) (fragments) that will model the domain. The model includes random variables along with their domain, relationship between random varibales, and the context variables. There are two levels of context information: global context and local context. The global context is applicable to the entire project (project is collection of all the action fragments). The local context is applicable at the node level as well as at the fragment level. The Semistructured Probabilistic Database Management System (SPDBMS) stores various probabilistic information and meta-information (context) about the bayesian network(s). The Bayesnet Server serves as a central repository of the bayesian networks created by the BNB tool, so that other tools like the Probability Elicitation Tool (PET), Preference Elicitation Tool and other planning tools can access the bayesian networks. PET is an application that allows experts in various domains to associate probabilities in the bayesian network that will be used in planning. These probabilities associated with different bayesian networks are stored in the SPDBMS.
These are the functionalities implemented in the BNB tool:
The tool recognizes fragments belonging to a single project
The tool allows to define domain of the nodes (random variables) in the bayesian network
The tool allows to define domain of the context variables (local as well as global context)
The tool communicates with the Bayesnet server for storing and retreiving fragments, local and global context information
These are the functionalities that are not yet implemented in the BNB tool:
The tool should communicate with the SPDBMS to get the probabilities
The tool should create situation specific networks
The tool should implement Knowledge Based Model Construction (KBMC) algorithms to form probability distributions of situation specific networks from the underlying data available in the SPDBMS.
These are the functionalities that are future work for the BNB tool:
The tool should be able to combine fragments and probability distributions
The tool should be made more user-friendly.
The Java runtime environment is needed to run the project. For Windows operating system, kindly set the classpath variable to include the two jar files that come along with the project download (bnb.zip). For Unix operating system, the startBNB script file needs to be updated to include the correct path to the jar files and the project executable. The main class file that needs to be executed is bnbtool.bnbproject
Here is more information on the Bayes Net Server.
Project instructions file - readme.txt
Sample network (fragment) file
Sample local context domain file
Sample global context domain file
Sample file assigning weights to different context elements
The schemas for the above mentioned xml files is available at schema link
Page last modified 6/22/2005
problems? -- kkmath2@uky.edu