StochTree Vignettes
Extended worked examples for the stochtree package, covering core models, practical topics, and advanced causal inference methods. Each vignette presents R and Python implementations side-by-side.
Core Models
| Vignette | Description |
|---|---|
| BART | Bayesian Additive Regression Trees for Supervised Learning |
| BCF | Bayesian Causal Forests for Treatment Effect Estimation |
| Probit BART | Probit BART for Binary Outcomes |
| Heteroskedastic BART | BART with a Forest-based Variance Model |
| Ordinal Outcome Modeling | BART with the Complementary Log-Log Link for Ordinal Outcomes |
| Multivariate Treatment BCF | BCF with Vector-valued Treatments |
Practical Topics
| Vignette | Description |
|---|---|
| Model Serialization | Saving and Loading Fitted Models |
| Multi-Chain Inference | Running and Combining Multiple MCMC Chains |
| Tree Inspection | Examining Individual Trees in a Fitted Ensemble |
| Summary and Plotting | Posterior Summary and Visualization Utilities |
| Prior Calibration | Calibrating Leaf Node Scale Parameter Priors |
| Scikit-Learn Interface | Using Stochtree via Sklearn-Compatible Estimators in Python |
Low-Level Interface
| Vignette | Description |
|---|---|
| Custom Sampling Routine | Building a Custom Gibbs Sampler with Stochtree Primitives |
| Ensemble Kernel | Using Shared Leaf Membership as a Kernel |
Advanced Methods
| Vignette | Description |
|---|---|
| BART with Targeted Smoothing | tsBART: Leaf-Regression BART as GP Approximation |
| Regression Discontinuity Design | BARDDT: Leaf-Regression BART for RDD |
| Instrumental Variables | IV Analysis via a Custom Monotone Probit Gibbs Sampler |