Introduction to Causal Inference using pgmpy
summary
Demonstrates the DAG approach with a hands‑on walkthrough on real data using pgmpy, showing causal discovery algorithms (PC, hill‑climb) that produce very different, often wrong graphs and a concrete failure on the Adult census example; introduces expert-in-the-loop and LLM prompts that reduce manual work and improve directions; runs implied conditional‑independence tests, a FERC‑style fit, correlation/structure scores, shows back‑door/front‑door/instrumental identification, and simulates from parameterized DAGs so estimates can actually be computed or shown to fail.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 1:05What is causal inference
- 3:12Causal graphs applications
- 5:18Two frameworks for inference
- 7:03Causal Discovery workflow
- 8:09Limitations of automated Discovery
- 9:14Example on real-world data
- 11:03Expert knowledge in Discovery
- 12:47Expert-in-the-loop methods
- 14:10Evaluating causal graphs
- 15:57Metrics for model fit
- 17:24Identification of causal effects
- 18:06When effects are identifiable
- 19:10Backdoor and frontdoor criteria
- 20:35Estimation strategies
- 21:17Future plans for pgmpy
- 22:00Q&A: Reverse causality
- 23:25Q&A: Redundant variables
- 24:31Q&A: Interpreting colliders
- 25:18Q&A: Expert graph variability
- 26:19Q&A: When to use methods
- 27:45Q&A: Benchmark datasets
- 28:49Q&A: Sample size limits
- 29:53Q&A: Framework choice
- 30:35Closing remarks
related talks
01h 02mUnmeasured Confounding and More Recent Developments/Challenges in Causal Discovery
01h 23mProbabilistic Graphical Models, HMMs using PGMPY
43mA Practical Guide to Causality in Python (For The Perplexed)
41mOn the Path to Causal Inference
01h 08mUnmeasured Confounding and More Recent Developments/Challenges in Causal Discovery
01h 06mIntroduction to Causal Graphical Models: Graphs, d-separation, do-calculus
01h 02mClassical Strategies and Concepts in Causal Discovery
29m🧠🧹 Causality - Mental Hygiene for Data Science