BN2O type networks
At this web page we provide details about Bayesian networks of the BN2O
type used in the experiments with arithmetic circuits described in the
paper:
J. Vomlel and P. Savicky. Arithmetic circuits
of the noisy-or models. In the Proceedings of the Fourth European Workshop on
Probabilistic Graphical Models (PGM'08).
All networks and all results of the experiments from the paper may be downloaded below.
BN2O models
The name of the BN2O model in the form of bn2o-x-y-e-i
contains information about the BN2O structure:
- x is the number of nodes in the top level,
- y is the number of nodes in the bottom level,
- e is the total number of edges in the BN2O model, and
- i is the model index within the x-y-e type.
For each x-y-e type x, y=10,20,30,40,50 and e/y=2,5,10,20, excluding those with e/y > x we generated randomly ten models (indexed by i=0,...,9). For every node from the bottom level we randomly selected e/y nodes from the top level as its parents.
Model formats
There are four files for each BN2O model in the tar file:
- pa-x-y-e-i.txt
text file, where each line corresponds to one node from the bottom level
and contains parents of this node.
- bn2o-x-y-e-i-orig.xmlnet
an XML file, describing the BN2O model using the XML tags for noisy-or
- bn2o-x-y-e-i-orig-prep.xmlnet
an XML file, describing the model after the transformation using parent divorcing.
- bn2o-x-y-e-i-transf.xmlnet
an XML file, describing the model after the transformation using tensor rank-one decomposition.
All BN2O models are in the XMLNET format of Ace 3.0.
Download
All models can be downloaded as one models-svetice.tar.gz file.
Complete results of experiments
For the experiments we used a development release of Ace, version 3.0.
Ace is a package that compiles a BN into an AC using one of
two available methods - c2d or the tabular compilation - see the Ace manual for details.
Since randomness is used during the Ace compilation process, Ace often
provides different circuits for the same input model. Therefore for
every model we performed ten experiments
five with c2d and five with the tabular method for both the original and the transformed models.
All lines of the result-x-y-e-i.txt files have the following format:
x y e i j o meth_o max_cluster_size_o t meth_t max_cluster_size_t
- where x, y, e, i are described above,
- j is the index of the experiment within the x-y-e-i model type
- o is the size (number of edges) of AC for the original model (constructed by the parent divorcing method)
- meth_o is the method used to construct the AC (c2d or tabular) of the original model
- max_cluster_size_o is the size of the largest clique of the
triangulated graph created during the compilation process of the
original model
- t is the size of the transformed model (constructed by the tensor rank-one decomposition)
- meth_t is the method used to construct the AC (c2d or tabular) of the transformed model
- max_cluster_size_t is the size of the largest clique of the
triangulated graph created during the compilation process of the transformed model.
All results can be dowloaded as one results-svetice.tar.gz file.