Making Existing Data Accessible

One of the most important practical issues when working with statistical relational learning methods is how we can easily make transform any kind of data that we may have available into a format that is usable by the learning and inference methods - as either a training database or an evidence database.

The SRLDB Library provides a unified data model for relational data, such that any required format (be it for MLNs or BLNs) can easily be generated. The representation of data in an SRLDB database object has several advantages:

• Data can be saved to MLN, BLN/BLOG and Proximity databases.
• Continuous values can be automatically discretized (using EM or k-means clustering).
• Data types can be automatically created and their domains built up (avoiding the need to provide explicit type and domain declarations).
• Basic model declarations can be automatically generated from the database's data dictionary and need not be manually specified (for MLNs and BLNs)

The SRLDB library can either be used directly as a Java library or via Jython:

• The use of the Jython scripting library is usally the more convenient choice. The Jython wrapper for the SRLDB library is the Datagen Library. Please read the overview article on this library.
• For information on the underlying Java library, please refer to the Java package probcog.srldb (and the javadoc therein)