Python and the Holy Grail of Causal Inference
summary
Claims that ruined randomized tests can be fixed with observed covariates, demonstrated by a heat‑pump deployment where propensity score matching corrects timing and selection bias and yields an average treatment effect for the test sample (not the full population) after extrapolating seasonal gas usage; a cooler placement study builds virtual twins with a tree ensemble, achieves R² ≈ 50%, ranks outlets by predicted uplift and actually prioritizes installs, all under the strong assumption of conditional mean independence.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 0:36What is Causal Inference
- 1:18"Slogan" for the Talk
- 1:35Typical Randomized Testing
- 2:15Use Case 1: Heat Pump Savings
- 2:45Measurement Data: Daily Gas Usage ~ Outside Temp
- 3:32Experiment Setup: Heat Pump Savings
- 5:40Fixing Group Imbalance
- 6:50Propensity Score Matching
- 7:47Recap of Heat Pump Use Case (Correcting Assumptions and Limitations)
- 9:07Use Case 2: Effect of Cooler Placement
- 9:40Experiment Setup: Cooler Placement
- 11:47Problem 1: Test and Control Groups are Statistically Different
- 13:17Code Example: Single Covariate Simulated Example
- 14:01Treatment Effect Correction
- 16:14Code Example: Treatment Effect Correction
- 16:28Assumptions: Conditional Mean Independence & Conditional Independence
- 18:44Tree Ensemble: Virtual Twins
- 20:21Results: Cooler Placement
- 22:18Talk Recap
- 24:50Q&A 1
- 26:38Q&A 2
- 27:32Q&A 3
- 29:00Q&A 4
- 30:29Q&A 5
related talks
41mOn the Path to Causal Inference
31mA Gentle Introduction To Causal Inference
31mLearning And Testing Causal Models: A Property Testing Viewpoint
31mIntroduction to Causal Inference using pgmpy
24mAn introduction to Causal Inference with Python – making accurate estimates of cause and effect from
43mA Practical Guide to Causality in Python (For The Perplexed)
26mCausal inference with Causallib
43mBeyond Just Prediction: Causal Thinking in Machine Learning