BBN — probabilistic reasoning

A very light-weight library for probabilistic reasoning on the basis of a directed graphs encoding causal relationships between variables.

We offer a Java library for probabilistic reasoning with Bayesian belief networks for unlimited variables with multinomial discrete values. It is appropriate for systems where the joint probability distribution can be decomposed into a sparsely connected directed graphs without cycles. The supplied inference algorithms are either bucket tree elimination of the clique tree algorithm.

The library is currently available in release uk.co.weirholme.wbbn-1.0-beta as a Java binary. If you are interested in testing this library with your applications, please contact us.

  1. Preliminary JavaDoc. This is strictly work in progress for the next release version 1.2 of the library.
  2. Sampling from the alarm network.