Model-based propositionalization for outlier detection
Transforms features learned with a relational BN model into a single table to be used in a standard single-table outlier detection method.
- A relational database
- A target entity set (e.g. Movies in IMDB)
- A learned Bayesian network (using FactorBase).
Output: A single data table where each row represents a target instance and each column represents a conjunctive relational feature.
How to Use - Overview
Run Bayesian Network to get Contingency Tables
We provide FactorBase to make a Bayesian network that shows probabilistic dependencies between the relationships and attributes represented in the database.
Note that the default setting of the original contingency tables in source code is "unielwin_CT" database. So before compile you should modify the
scripts/setup.sqlto replace "unielwin_CT" with correct original table path.
Point to the required database in your MySQL server
Download the source code to your local computer. Modify
src/config.cfgwith your own configuration according to the sample format explained in the image.
The "dbname" represents the name of schema which stores the converted tables.
The "column" represents the name of the column where the primary key is located.
The "tablename" represents the name of the contingency table in original schema.
Compile & Run
- Go into
javac -cp ".:./lib/*" Convert_table.java
java -cp ".:./lib/*" Convert_table > result.txt
Then you can check the converted table in your "dbname".
- Go into