Probabilistic graphical models in python

This code is intended mainly as proof of concept of the algorithms presented in [1]. The implementations are not particularly clear, efficient, well tested or numerically stable. We advise against using this software for nondidactic purposes.

This software is licensed under the MIT License.

Features

• Models

  • Bayesian network (table conditional probability distributions)
  • Markov network (table potentials)
  • Influence diagram

• Inference

  • Variable elimination
  • Forward sampling
  • Gibbs sampling

• Learning

  • Parameter learning (maximum likelihood, uniform BDe, expectation maximization for missing data)
  • Structure learning (local search, likelihood score, BIC score, Bayesian score)

Examples

See the examples directory.

References

[1] Koller, D. and Friedman, N. Probabilistic Graphical Models: Principles and Techniques. The MIT Press. 2009.