Chain Graph Sampling Resources

On this page we make available the programs, source code, and samplesets from chain graph (CG) sampling methods presented in literature. It is our intent that the samples can be used to get a better insight into what systems the different CG interpretations can represent, but also that they can be used to get better evaluations when evaluating for example structure learning algorithms. Use the attached ReadMe-files to interpret the given graph files.

LWF CGs

Enumerated LWF CG models

The set of LWF CG models (largest chain graphs) that exist for 2 to 5 nodes. This set of models can be used to evaluate the correctness of other sampling methods.
Link to dataset.

MCMC-sampler of LWF CG models

Does: Samples LWF CG models uniformly using a Markov chain Monte Carlo approach. The sampled models are in the form of largest CGs.
Presented by: Peña, J. M.
Article: Approximate Counting of Graphical Models Via MCMC, In Proceedings of the 11th International Conference on Artificial Intelligence and Statistics (AISTATS 2007), 352-359. 2007.
Sourcecode and Program.
10^5 samples sampled with 10^5 iterations between each sample for 2 to 25 nodes

AMP CGs

Enumerated AMP CG models

The set of AMP CG models (Largest deflagged graphs) that exist for 2 to 5 nodes. This set of models can be used to evaluate the correctness of other sampling methods.
Link to dataset.

MCMC-sampler of AMP CG models

Does: Samples AMP CG models uniformly using a Markov chain Monte Carlo approach. The sampled models are in the form of largest deflagged graphs.
Presented by: Sonntag, D.
Article: On Expressiveness of the AMP CG Interpretation. (Under Review).
Sourcecode and Program.
10^5 samples sampled with 10^5 iterations between each sample for 2 to 25 nodes

MVR CGs

Enumerated MVR CG models

The set of MVR CG models (essential MVR CGs) that exist for 2 to 5 nodes. This set of models can be used to evaluate the correctness of other sampling methods.
Link to dataset.

MCMC-sampler of MVR CG models

Does: Samples MVR CG models uniformly using a Markov chain Monte Carlo approach. The sampled models are in the form of essential MVR CGs.
Presented by: Sonntag, D., Peña, J. M. and Gómez-Olmedo, M.
Article: Approximate Counting of Graphical Models Via MCMC Revisited. International Journal of Intelligent Systems, accepted. 2014.
Sourcecode and Program.
10^5 samples sampled with 10^5 iterations between each sample for 2 to 25 nodes