Janne H. Korhonen > AISTATS 2013
Exact Learning of Bounded Tree-width Bayesian Networks
Software

This page contains the best-w-tree software package discussed in Section 6 of the paper

  • Learning bounded tree-width Bayesian networks
    Janne H. Korhonen and Pekka Parviainen. In 16th International Conference on Artificial Intelligence and Statistics (AISTATS 2013), pages 370–378. JMLR, 2013.

The package is available for download under GPL version 3 license. The documentation below is also included in the package.

README
                    
best_w_tree is a simple software tool for finding optimal bounded tree-width Bayesian networks.                    
                    
INSTALLATION
                    
To use this software, you will need to have Python 2.7 and Cython (http://www.cython.org) installed. The you can compile the software by running
                    
$ python setup.py build_ext --inplace
                    
                    
                    
USAGE
                    
To find an optimal network with tree-width 2 for included test data, simply run
                    
$ python best_w_tree.py 2 data/test.dat
                    
The first parameter defines the tree-width bound and the second is the input file.
                    
To run the software with the scores obtained from Adult and Housing data sets of UCI Machine Learning Repository (http://archive.ics.uci.edu/ml), try the following commands
                    
$ python best_w_tree.py -n 10 -z 2 data/housing.dat
$ python best_w_tree.py -n 10 -z 2 data/adult.dat
                    
The option -n limits the number of included variables and -z tells the program to treat score value "0.0" in the input files as minus infinity.
                    
For a full list of options, run
                    
$ python best_w_tree.py --help
                     
                     
                     
INPUT FORMAT
                    
The score file should contain one line per variable. If there are n variables, each line should contain a whitespace-separated list of 2^n values. The j:th value on line i is interpreted as the local score f_i(J), where J is the set obtained by selecting all elements x in {0,1,2,...,n-1} such that the bit x is 1 in j. For example, 
                    
106 = 1101010 = { 1, 3, 5, 6}
                    
Note that there is a degree of redundancy here; entries with i:th bit set are required on line i, but they will never be used by the software, as i cannot be in its own parent set.
                    
Any "-" in the input is interpreted as minus infinity, as is "0.0", if the option -z is given to the software.



LICENSING

best_w_tree is distributed under GPL version 3, see the source code.