pgmpy/pgmpy

Python Toolkit for Causal and Probabilistic Reasoning

What it solves

pgmpy provides a framework for causal and probabilistic reasoning using graphical models. It allows users to build, learn, and perform inference on complex models that represent dependencies between variables, enabling the discovery of causal relationships and the prediction of outcomes based on observed evidence.

How it works

The library implements various data structures for graphical models, including Directed Acyclic Graphs (DAGs), Bayesian Networks, and Structural Equation Models. It provides a modular, composable API for several key tasks:

  • Structure Learning: Learning the model structure from data, with the ability to integrate expert knowledge.
  • Parameter Estimation: Estimating conditional probability distributions (CPDs) from observed data.
  • Inference: Computing posterior distributions (probabilistic inference) or interventional distributions (causal inference).
  • Simulation: Generating synthetic data based on specified evidence or interventions.

The tool is designed to be scikit-learn compatible where possible, allowing it to be integrated into standard ML pipelines.

Who it’s for

It is designed for researchers and developers working with probabilistic graphical models, causal discovery, and causal inference.

Highlights

  • Broad Model Support: Supports DAGs, PDAGs, MAGs, PAGs, Bayesian Networks, Dynamic Bayesian Networks, and Structural Equation Models.
  • Causal Toolset: Includes algorithms for causal discovery, identification, and validation.
  • Flexible Data Handling: Works with discrete data, linear Gaussian data, and mixture data with arbitrary relationships (via a Torch backend).
  • Modular API: Uses a unified, composable API that makes the library extensible and compatible with scikit-learn pipelines.

Related

  • Project
    erdogant/bnlearn

    bnlearn is a Python library for learning, parameterising, and querying Bayesian networks. It supports structure discovery (various scores and search methods), CPT estimation, causal inference (do‑calculus), synthetic data generation, and visualisation, all wrapped in a simple API.

  • Project
    SheffieldML/GPy

    A Gaussian processes framework in Python that provides tools for implementing and optimizing probabilistic models for regression and classification.

  • Project
    cdt15/lingam

    A Python library for causal discovery that estimates linear non-Gaussian acyclic models to determine the causal relationships between variables.

  • Project
    py-why/dowhy

    DoWhy is a Python library for end‑to‑end causal inference. It lets you define a causal graph, automatically identifies the causal effect, estimates it with a variety of statistical methods, and runs refutation tests to check robustness. The library also supports graphical causal models for root‑cause analysis, counterfactuals, and interventional sampling. Install via `pip install dowhy` and follow the four‑step workflow (model → identify → estimate → refute).

  • Project
    pymc-labs/CausalPy

    A Python library for research-grade causal inference in quasi-experimental designs, offering Bayesian and OLS estimation with full uncertainty quantification.