Quickstart

Task-oriented examples to help you get started with pgmpy.

The sections below provide a minimal example for various supported tasks and a link to the corresponding User Guide page for the full workflow.

Causal Discovery / Structure Learning

User Guide: Causal Discovery and Structure Learning

Learn a graph structure directly from data.

from pgmpy.datasets import load_dataset
from pgmpy.causal_discovery import PC

dataset = load_dataset("sachs_discrete")
est = PC(ci_test="chi_square", return_type="dag")
est.fit(dataset.data)
print(est.causal_graph_)

Parameter Estimation

User Guide: Parameter Estimation

Fit conditional distributions for a known graph structure.

from pgmpy.estimators import MaximumLikelihoodEstimator
from pgmpy.models import DiscreteBayesianNetwork
from pgmpy.example_models import load_model

# Generate some data to use for fitting.
model = load_model("bnlearn/alarm")
df = model.simulate(n_samples=1000, seed=42)

# Create a network structure and fit data to it.
bn_struct = DiscreteBayesianNetwork(model.edges())
bn_struct.fit(df, estimator=MaximumLikelihoodEstimator)
bn_struct.cpds

Probabilistic Inference

User Guide: Probabilistic Inference

Query posterior distributions from a Bayesian Network.

from pgmpy.example_models import load_model
from pgmpy.inference import VariableElimination

model = load_model("bnlearn/alarm")
infer = VariableElimination(model)

result = infer.query(
    variables=["HISTORY"],
    evidence={"CVP": "LOW", "PCWP": "LOW"},
)
print(result)

Causal Identification

User Guide: Causal Identification

Check whether a causal effect is identifiable from the graph alone.

from pgmpy.base import DAG
from pgmpy.identification import Adjustment

dag = DAG(
    [("X", "Y"), ("Z", "X"), ("Z", "Y")],
    roles={"exposures": "X", "outcomes": "Y"},
)
identified_graph, is_identified = Adjustment(variant="minimal").identify(dag)
identified.get_role("adjustment")

Causal Inference

User Guide: Causal Estimation

Estimate a causal effect from data once you have a causal graph.

from pgmpy.example_models import load_model
from pgmpy.inference import CausalInference

model = load_model("bnlearn/sachs")
infer = CausalInference(model)
infer.query(variables=["Akt"], do={"PKC": "LOW"})

Example Datasets and Models

User Guides: Example Datasets | Example Models

Discover built-in datasets and example models.

from pgmpy.datasets import list_datasets, load_dataset
from pgmpy.example_models import list_models, load_model

print(list_datasets(is_discrete=True, has_ground_truth=True)[:3])
dataset = load_dataset("sachs_discrete")
print(dataset.name, dataset.data.shape)

print(list_models()[:3])
model = load_model("bnlearn/alarm")
print(len(model.nodes()), len(model.edges()))

Simulations

User Guide: Simulations

Generate synthetic data from a model for testing and experimentation.

from pgmpy.example_models import load_model

model = load_model("bnlearn/ecoli70")
data = model.simulate(int(1e3))
data.head()

Extend pgmpy

User Guide: Extensibility

Implement new datasets, models, metrics, or algorithms from the repository templates. pgmpy provides extension templates for each class of method. To add a new method, you can just follow the TODOs listed in the extension template for that method. All of our extension templates can be found at: https://github.com/pgmpy/pgmpy/tree/dev/devtools/extension_templates

Next Steps

• User Guide – Workflow-oriented documentation for each task area

• Examples – Jupyter notebooks with longer walkthroughs

• API Reference – Full public API documentation