Examples

Working example notebooks are available in the example folder.

The notebooks are grouped by topic. Within each group they run from introductory to advanced, so a group’s first notebook is the place to start.

Getting started

The two canonical tours of the package: CATE estimation with the meta-learners, and uplift modeling with the tree-based learners.

• Meta-Learners Examples - Training, Estimation, Validation, Visualization
• Uplift Trees Example with Synthetic Data

Meta-learners

• Meta-Learners Examples - Single/Multiple Treatment Cases
• DR Learner vs. DR-IV Learner vs. X-Learner Benchmark with Synthetic Data

Uplift and causal trees

• Causal Trees/Forests Treatment Effects Estimation and Tree Visualization
• Causal Trees/Forests Treatment Effects Estimation and Tree Visualization: Multiple treatment groups
• Causal Trees/Forests Interpretation with Feature Importance and SHAP Values
• Honest causal trees: parity with grf and EconML

Evaluation and validation

Validating a CATE model without ground truth: uplift and Qini curves, TMLE-based evaluation, calibration, and sensitivity analysis.

• Uplift Curves with TMLE Example
• Propensity Score Calibration
• Modify above example to use IsotonicRegression
• Qini curves with multiple costly treatment arms
• Sensitivity Analysis Examples

Interpretation and feature selection

• Model Interpretation with Feature Importance and SHAP Values
• Uplift Trees/Forests Visualization
• Feature Selection for Uplift Trees by Zhao et al. (2020)

Neural models

DragonNet and CEVAE, each with a comparison of its available backends. These need the optional tf, torch or jax extras.

• DragonNet vs Meta-Learners Benchmark with IHDP + Synthetic Datasets
• DragonNet: JAX vs TF Benchmark
• CEVAE vs. Meta-Learners Benchmark with IHDP + Synthetic Datasets
• CEVAE: JAX vs PyTorch Backend Benchmark

Instrumental variables

• 2SLS Benchmarks with NLSYM + Synthetic Datasets

Policy and value optimization

Turning effect estimates into treatment decisions: policy learning, counterfactual unit selection and value optimization, and probabilities of causation.

• Policy Learner by Athey and Wager (2018) with Binary Treatment
• Unit Selection Based on Counterfactual Logic by Li and Pearl (2019)
• Counterfactual Value Estimation Using Outcome Imputation by Li and Pearl (2019)
• Calculating the Probabilities of Necessary and Sufficient Causation - Tian and Pearl (2000)

Datasets and benchmarks

The benchmark leaderboard regenerates every number published on the Benchmark Datasets page; the simulation studies compare estimators on synthetic and semi-synthetic data.

• Benchmark Leaderboard
• Meta-Learner Benchmarks with Synthetic Data in Nie and Wager (2020)
• Meta-Learner Benchmarks with Semi-synthetic Data in Schuler, A., Jung, K., Tibshirani, R., Hastie, T., and Shah, N. Synth-validation: Selecting the best causal inference method for a given dataset (2017)
• Logistic Regression Based Data Generation Function for Uplift Classification Problem