IJGPIS
The IJGPIS package performs approximate inference in Bayesian networks. The current software allows approximate probability of evidence computation and belief updating.
Author
Description
IJGPIS is an importance sampling technique that uses the output of Iterative Join Graph Propagation (IJGP) to compute a importance distribution. It also uses relational consistency to solve the rejection problem. A more detailed description can be found in [1].
Input format:
The algorithm uses the Ergo_file_format.
Usage
Call the algorithms as follows:
ijgpis [parameters] -f <ergo-file>
The parameters are detailed below.
Parameters
--task: (int) 0 for computing probability of evidence and 1 for computing updated beliefs (default value is 0)
--ordering: <minfill,topological,mindegree> What ordering to use.
--i-bound: (int) An integer value for the i-bound (default value is 3)
--num-iterations: (int) The maximum number of iterations for which IJGP is run
--outfile (string): The path of the output file (default is out)
--num-samples: (int) The maximum number of samples to be drawn
--time-limt: (int) The maximum time in seconds for which the algorithm is run
--help : Print this help.
Download
A 32bit Linux executable is available here.
References
[1] Approximate Inference Algorithms for Hybrid Bayesian Networks with Discrete Constraints. Vibhav Gogate and Rina Dechter. In 11th Conference on Uncertainty in Artificial Intelligence (UAI), 2005. Link