Graphical Models and Causality

M2 Data Science course (Université d’Évry): directed and undirected graphical models, d-separation, Gaussian graphical models, interventions and structural causal models, structure learning, LiNGAM and OT-LiNGAM.

This course, taught in the M2 Data Science at Université d’Évry, goes from probabilistic graphical models to causal reasoning. It starts from directed acyclic graphs and the factorisation they encode, d-separation and Markov equivalence, then covers Markov random fields and Gaussian graphical models. The second half turns to causality: interventions, structural causal models and identification, structure learning with scores and conditional independence tests, and causal discovery with non-Gaussian noise (LiNGAM and OT-LiNGAM).

The course has six sessions, each with a lecture and a tutorial. The slides are available in two forms: interactive HTML slides and a printable PDF version, with four slides per page. Practical work is in Python.

Lectures

Lecture	Topic	HTML	PDF
1	Causality and directed graphical models	Slides	Printable version
2	d-separation and Markov equivalence	Slides	Printable version
3	Markov random fields and Gaussian graphical models	Slides	Printable version
4	Interventions, structural causal models and identification	Slides	Printable version
5	Structure learning: scores and conditional independence tests	Slides	Printable version
6	Causal discovery with non-Gaussian noise: LiNGAM and OT-LiNGAM	Slides	Printable version

Tutorials

Tutorial booklet — PDF version without solutions: one sheet per lecture, with theory exercises (pen and paper) and Python practice.