pgmpy is a Python package for causal inference and probabilistic inference using Directed Acyclic Graphs (DAGs) and Bayesian Networks with a focus on modularity and extensibility. Implementations of various algorithms for Causal Discovery (a.k.a, Structure Learning), Parameter Estimation, Approximate (Sampling Based) and Exact inference, and Causal Inference are available.

Possible Workflows in pgmpy for Directed Acyclic Graphs (DAGs) and Bayesian Networks (BNs).

Supported Data Types

	Causal Discovery	Parameter Estimation	Causal Inference	Probabilistic Inference	Simulations
Categorical	Yes	Yes	Yes	Yes	Yes
Continuous	Yes	Yes	Yes (partial)	Yes	Yes
Mixed	Yes	No	No	No	Yes
Time Series	No	Yes	Yes (ApproximateInference)	Yes	Yes

Algorithms

Causal Discovery / Structure Learning	Parameter Estimation	Probabilistic Inference	Causal Inference
PC with variants	Maximum Likelihood	Variable Elimination	do-operation
Greedy Equivalence Search(GES)	Bayesian Estimator	Belief Propagation	adjustment sets
Hill-Climb Search	Expectation Maximization (EM)	MPLP
Expert In The Loop		Sampling methods
Tree Search
Max-Min Hill-Climb
Exhaustive Search

Examples

Example notebooks: Examples

Tutorial notebooks: Tutorial Notebooks

Citation

If you use pgmpy in your scientific work, please consider citing us:

Ankur Ankan, & Johannes Textor (2024). pgmpy: A Python Toolkit for Bayesian Networks. Journal of Machine Learning Research, 25(265), 1–8.

Bibtex:

@article{Ankan2024,
  author  = {Ankur Ankan and Johannes Textor},
  title   = {pgmpy: A Python Toolkit for Bayesian Networks},
  journal = {Journal of Machine Learning Research},
  year    = {2024},
  volume  = {25},
  number  = {265},
  pages   = {1--8},
  url     = {http://jmlr.org/papers/v25/23-0487.html}
}