Prototype implementation of direct and factored deletion algorithms to learn Bayesian network parameters from incomplete data under the MCAR and MAR assumptions. These algorithms are consistent, yet they only require a single pass over the data, and no inference in the Bayesian network.
Guy Van den Broeck, Karthika Mohan, Arthur Choi, Adnan Darwiche, Judea Pearl. Efficient Algorithms for Bayesian Network Parameter Learning from Incomplete Data, In Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence (UAI), 2015.
http://reasoning.cs.ucla.edu/deletion/
Guy Van den Broeck
Department of Computer Science
UCLA
http://web.cs.ucla.edu/~guyvdb/
guyvdb@cs.ucla.edu
The code takes as input Bayesian networks in the UAI file format. A simple example called fire_alarm.uai is shown below.
BAYES
6
2 2 2 2 2 2
6
1 0
1 1
2 0 2
3 0 1 3
2 3 4
2 4 5
2
0.01 0.99
2
0.02 0.98
4
0.9 0.1 0.01 0.99
8
0.5 0.5 0.99 0.01 0.85 0.15 0.0001 0.9999
4
0.88 0.12 0.001 0.999
4
0.75 0.25 0.01 0.99
This source code is licensed under the Apache License, Version 2.0: http://www.apache.org/licenses/LICENSE-2.0
This software uses the inference library inflib.jar, which is provided by the Automated Reasoning Group at UCLA. inflib.jar is licensed only for non-commercial, research and educational use.
