This webpage provides access to source code, examples, and other resources for the BayesBase learning system. The system is mainly an implementation of the Learn-and-Join Algorithm. For background concepts and pseudocode, please see our paper. We make heavy use of the Tetrad System. The previous version is available here. The current version outputs Markov Logic Networks.

Basic Workflow

The input to the system is a relational schema hosted on a MySQL. The output is a Bayes net that shows probabilistic dependencies between the relationships and attributes represented in the database. The output comes in two formats.

  1. A representation of the Bayes net model in relational tables. We treat as first-class relational citizens all structured objects such as the Bayes net graph, conditional probability tables, statistical scores. SQL can be used to query the Bayes net model components.
  2. A Bayes Net Interchange Format file for viewing in a standard Bayes net reader. We have tested this with UBC's AIspace tool.