Bayesian Network Repository

Several Bayesian networks are commonly used in literature as benchmarks. They are available in different formats from several sources, the most famous one being the Bayesian network repository hosted at the Hebrew University of Jerusalem. Others are shipped as examples of various Bayesian network-related software like Hugin or described in reference books such as Korb & Nicholson's “Bayesian Artificial Intelligence” or Koller & Friedman's “Probabilistic Graphical Models”.

Here I collected all the networks I used for simulations and for testing the implementations of read.bif(), read.dsc() and read.net() in bnlearn. All the networks are available in the BIF, DSC and NET formats and have been quality-checked and fixed as needed (i.e. all conditional probability distributions sum to one, no dummy nodes with a single level, no dangling dependencies on non-existent nodes, etc.). R objects with the corresponding bn.fit object are also provided.

 

Small Networks (<20 nodes)

Name

Nodes

Arcs

Parameters

ASIA8818
CANCER5410
EARTHQUAKE5410
SACHS1117178
SURVEY6621

Medium Networks (20–60 nodes)

Name

Nodes

Arcs

Parameters

ALARM3746509
BARLEY4884114005
CHILD2025230
HAILFINDER56662656
INSURANCE2752984
MILDEW3546540150
WATER326610083

Large Networks (60–100 nodes)

Name

Nodes

Arcs

Parameters

HEPAR II701231453
WIN95PTS76112574

Very Large Networks (100–1000 nodes)

Name

Nodes

Arcs

Parameters

ANDES2233381157
DIABETES413602429409
LINK724112514211
MUNIN (4 subnetworks)186–1041273–138815622–80352
PATHFINDER13520077155
PIGS4415925618

Massive Networks (>1000 nodes)

Name

Nodes

Arcs

Parameters

MUNIN (full network)1041139780592
MUNIN (4 subnetworks)186–1041273–138815622–80352