Classical Strategies and Concepts in Causal Discovery
summary
Explains how to run PC and score-based search end to end, with concrete tests (the Fisher Z partial-correlation test for Gaussians, chi-square/G-tests for discrete data), nonparametric options like the kernel-based conditional independence (KCI) test and generalized conditional covariance measures, and a full GES procedure using BIC with forward/backward moves. Proves consistency under strong faithfulness and gives high-dimensional rates, notes computational costs (KCI ~O(N^3)), error propagation in PC, and why GES often recovers mistakes in the backward pass.
AI-written summary · about
chapters (18)
- 0:00Start
- 4:21Recursive Definition of Partial Correlation
- 5:35Fischer Z Transformation
- 6:53Non-Parametric Tests
- 9:20Non-Parametric Conditional Dependence Test
- 9:31Test Based on the Generalized Conditional Covariance Measure
- 11:49Virtues of Pc
- 17:41Posterior Probability
- 32:22Multivariate Gaussian
- 36:35Pseudo Code for Ges
- 36:46Forward Search
- 37:56Backward Stage
- 44:25Score Properties
- 45:16Score Equivalence
- 45:48Decomposability
- 46:38Local Consistency
- 50:15Virtues and Drawbacks of Score Based Search
- 1:01:21Strong Faithfulness
related talks
31mIntroduction to Causal Inference using pgmpy
24mA Friendly Introduction To Causality
31mLearning And Testing Causal Models: A Property Testing Viewpoint
32mNew Approaches To Learning Nonparametric (Latent) Causal Graphical Models
49mCausal Inference in Complex Systems: Network Interference, Strategic Agents, and Beyond
43mA Practical Guide to Causality in Python (For The Perplexed)
31mPython and the Holy Grail of Causal Inference
01h 01mThe Mathematics of Causal Inference, with Reflections on Machine Learning and the Logic of Science