py-why/causal-learn

What it solves

Causal-learn provides a platform for causal discovery, allowing users to recover the causal structure of a system from observational data. It eliminates the need for expensive or impossible interventional experiments to determine cause-and-effect relationships.

How it works

The library implements a wide range of causal discovery algorithms, including constraint-based, score-based, and permutation-based methods, as well as Granger causality and methods based on constrained functional causal models. It also includes hidden causal representation learning and a suite of utilities for independence tests, score functions, and graph operations to help users build their own custom methods.

Who it’s for

It is designed for researchers and developers working in machine learning and causality who need to recover causal graphs from data.

Highlights

• Implements both classical and state-of-the-art causal discovery algorithms.
• Provides provable correctness guarantees for structure recovery.
• Includes a comprehensive set of utilities for independence tests and graph evaluations.
• Offers an online agentic version via causallearn.com.