This collection consists of 5 structure learning datasets from the Bayesian Network Repository (Scutari, 2010).

Task: The dataset collection can be used to study causal discovery algorithms.

Summary:

• Size of collection: 5 datasets with 3 - 56 columns of various sizes
• Task: Causal Discovery
• Data Type: Discrete
• Dataset Scope: Collection
• Ground Truth: Known / Estimated
• Temporal Structure: No
• License: TBD
• Missing Values: No

Missingness Statement: There are no missing values.

Collection:

The alarm dataset contains the following 37 variables:

• CVP (central venous pressure): a three-level factor with levels LOW, NORMAL and HIGH.
• PCWP (pulmonary capillary wedge pressure): a three-level factor with levels LOW, NORMAL and HIGH.
• HIST (history): a two-level factor with levels TRUE and FALSE.
• TPR (total peripheral resistance): a three-level factor with levels LOW, NORMAL and HIGH.
• ... (33 more variables, see the corresponding .html file)

The binary synthetic asia dataset:

• D (dyspnoea), a two-level factor with levels yes and no.
• T (tuberculosis), a two-level factor with levels yes and no.
• L (lung cancer), a two-level factor with levels yes and no.
• B (bronchitis), a two-level factor with levels yes and no.
• A(visit to Asia), a two-level factor with levels yes and no.
• S (smoking), a two-level factor with levels yes and no.
• X (chest X-ray), a two-level factor with levels yes and no.
• E (tuberculosis versus lung cancer/bronchitis), a two-level factor with levels yes and no.

The binary coronary dataset:

• Smoking (smoking): a two-level factor with levels no and yes.
• M. Work (strenuous mental work): a two-level factor with levels no and yes.
• P. Work (strenuous physical work): a two-level factor with levels no and yes.
• Pressure (systolic blood pressure): a two-level factor with levels <140 and >140.
• Proteins (ratio of beta and alpha lipoproteins): a two-level factor with levels <3 and >3.
• Family (family anamnesis of coronary heart disease): a two-level factor with levels neg and pos.

The hailfinder dataset contains the following 56 variables:

• N07muVerMo (10.7mu vertical motion): a four-level factor with levels StrongUp, WeakUp, Neutral and Down.
• SubjVertMo (subjective judgment of vertical motion): a four-level factor with levels StrongUp, WeakUp, Neutral and Down.
• QGVertMotion (quasigeostrophic vertical motion): a four-level factor with levels StrongUp, WeakUp, Neutral and Down.
• CombVerMo (combined vertical motion): a four-level factor with levels StrongUp, WeakUp, Neutral and Down.
• AreaMesoALS (area of meso-alpha): a four-level factor with levels StrongUp, WeakUp, Neutral and Down.
• SatContMoist (satellite contribution to moisture): a four-level factor with levels VeryWet, Wet, Neutral and Dry.
• ... (49 more variables are in the correspondent .html file)

The lizards dataset contains the following 3 variables:

• Species (the species of the lizard): a two-level factor with levels Sagrei and Distichus.
• Height (perch height): a two-level factor with levels high (greater than 4.75 feet) and low (lesser or equal to 4.75 feet).
• Diameter (perch diameter): a two-level factor with levels narrow (greater than 4 inches) and wide (lesser or equal to 4 inches).

We provide 100 samples, each containing 5000 instances, of discrete Bayesian Networks from bnlearn's Bayesian Network Repository. Specifically, the BNs, along with their characteristics, are:

NETWORK	#NODES	#EDGES	#PARAMETERS	MAX. PARENTS	MEAN DEGREE
Cancer	5	4	10	2	2.00
Earthquake	5	4	10	2	2.00
Survey	6	6	21	2	2.00
Asia	8	8	18	2	2.00
Sachs	11	17	178	3	3.09
Child	20	25	230	4	2.50
Insurance	27	52	230	4	2.50
Water	32	66	10083	5	4.12
Mildew	35	46	540150	3	2.63
Alarm	37	46	509	5	2.49
Barley	48	84	114005	4	3.50
Hailfinder	56	66	2656	4	2.36
Hepar2	70	123	1453	6	3.51
Win95pts	76	112	574	7	2.95
Pathfinder	109	195	72079	5	3.58
Munin1	186	273	15622	3	2.94
Andes	223	338	1157	6	3.03
Diabetes	413	602	429409	2	2.92
Pigs	441	592	5618	2	2.68
Link	724	1125	14211	3	3.11
Munin2	1003	1244	69431	3	2.48
Munin4	1038	1388	80352	3	2.67
Munin3	1041	1306	71059	3	2.51
Munin	1041	1397	80592	3	2.68

Each dataset is sampled using Python and the bnlearn package. The BN structure is loaded from the .bif file using bif = bnlearn.import_DAG(path), and samples are generated with bnlearn.sampling(bif, n=5000, methodtype='bayes'). Post-processing is then applied to replace the generated numerical values with those categorical specified in the .bif structure file.

Additionally, ten extra old database samples of most BNs can be found on OpenML.