Example Models#

pgmpy ships a registry of ready-made Bayesian network models from standard repositories. These range from structure-only causal graphs to fully parameterized networks that can be queried and simulated immediately — useful for benchmarking, demonstrations, and exploring the API.

Tip

When to use this vs. Example Datasets: Example Models provide pre-built graph structures (and optionally parameters) ready for inference and simulation. Example Datasets provide data tables for learning structures and parameters from scratch.

At a Glance#

Unified API: Discover models with list_models(...) and load them with load_model(...).
Rich Filtering: Filter models by type, parameterization, and size before loading.
Multiple Model Families: Discrete, continuous, and structure-only models from bnlearn, bnrep, and dagitty repositories.

API#

The example-model API mirrors the dataset API — discover, then load:

from pgmpy.example_models import list_models, load_model

names = list_models(is_parameterized=True, is_discrete=True)
model = load_model(names[0])

print(model)
print(len(model.nodes()), len(model.edges()))

The returned object type depends on the model family (e.g., DiscreteBayesianNetwork, LinearGaussianBayesianNetwork, or DAG).

Filtering#

list_models(**filters) narrows the registry before loading. Supported filters include is_parameterized, is_discrete, is_continuous, is_hybrid, n_nodes, and n_edges.

Model Families#

Discrete parameterized: Classic networks with full CPD tables, ready for inference and simulation.
Continuous: Networks with linear Gaussian parameters.
Structure-only: Causal graphs without parameters, useful for identification and causal reasoning tasks.

Loaded models integrate directly with the rest of the pgmpy stack — inference, simulation, parameter estimation, and plotting all work out of the box.

API Reference#

For the full list of available models: