Building on the understanding of Structural Causal Models (SCMs) and identification from Chapter 1, this chapter concentrates on methods for learning the causal structure directly from data. Inferring the correct causal graph, often represented as a Directed Acyclic Graph (DAG), from observational data presents significant challenges.

Here, you will investigate established and modern algorithms for causal discovery. We will cover:

• Constraint-based approaches: Techniques like the PC and FCI algorithms (and their extensions) that use statistical tests of conditional independence, such as $X \perp Y | Z$, to infer structural properties.
• Score-based approaches: Methods like GES and LiNGAM that search for the graph structure maximizing a defined scoring criterion.
• Utilizing diverse data: Techniques for incorporating interventional data and handling heterogeneous datasets to improve discovery accuracy.
• Practical considerations: Evaluating discovery algorithm performance, addressing the presence of latent variables, and implementing these methods using Python libraries.

Upon completing this chapter, you will have the necessary knowledge to apply and evaluate different causal discovery algorithms in practical settings.