tomisilander / bene

An exact Bayesian network structure learning software based on dynamic programming.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

bene

This package contains software for constructing the globally optimal Bayesian network structure using decomposable scores AIC, BIC, BDe, fNML, qNML and LOO.

Requirements

bash, gcc. If you also have graphviz package (thus dot) you can get images of network structures.

Compile

Go to build directory and say ./build.sh

This is how the package is used

Run and study example/run.sh and then probably bin/data2net.sh, You may also take a look at util/data2net.py

Input data format

(see example directory for an example)

vdfile value description file in which each line containis the variable name and the names of the values all separated by tabulators.

datfile datafile for discretized data, data vectors on rows, variables on columns values numerated from zero on (0, 1, 2, ..)

DISCLAIMER

This work is anything but ready, tested etc. Use the source.

Contact

tomi.silander@iki.fi

About

An exact Bayesian network structure learning software based on dynamic programming.


Languages

Language:C 90.0%Language:Python 5.6%Language:Shell 4.4%