ShrinkageTrees

This package provides functions for fitting Horseshoe Trees, Causal Horseshoe Forests, and their more general counterparts: Shrinkage Trees and Causal Shrinkage Forests.

These models allow for global-local shrinkage priors on tree step heights, enabling adaptive modeling in high-dimensional settings.

The functions can be used for:

1) High-dimensional prediction
2) High-dimensional causal inference 3) Estimaton of heterogeneous (conditional average) treatment effects

Supported outcome types: continuous, binary, and right-censored survival times.

The mathematical background and theoretical foundation for these models is described in the preprint Horseshoe Forests for High-Dimensional Causal Survival Analysis by T. Jacobs, W.N. van Wieringen, and S.L. van der Pas (arXiv:2507.22004).

✨ Features

• Horseshoe, forest-wide horseshoe, empirical Bayes Horseshoe, and half-Cauchy priors
• Flexible tree-based non-linear modeling of the ATE and CATE
• Supports survival data with right-censoring (accelerated failure time model)
• Efficient C++ backend via Rcpp

📦 Installation

The released version of ShrinkageTrees can be installed from CRAN:

r install.packages("ShrinkageTrees")

You can install the development version from GitHub:

```r

Install devtools if not already installed

install.packages("devtools") devtools::install_github("tijn-jacobs/ShrinkageTrees") ```

🚀 Example

```r library(ShrinkageTrees)

set.seed(42) n <- 100 p <- 1000

Generate covariates

X <- matrix(runif(n * p), ncol = p) Xtreat <- Xcontrol <- X treatment <- rbinom(n, 1, X[, 1]) tau <- 1 + X[, 2]/2 - X[, 3]/3 + X[, 4]/4

Generate survival times (on log-scale)

truetime <- X[, 1] + treatment * tau + rnorm(n) censortime <- log(rexp(n, rate = 0.05)) followup <- pmin(truetime, censortime) status <- as.integer(truetime <= censor_time)

Fit a standard Causal Horseshoe Forest (without propensity score adjustment)

fithorseshoe <- CausalHorseForest( y = followup, status = status, Xtraincontrol = Xcontrol, Xtraintreat = Xtreat, treatmentindicatortrain = treatment, outcometype = "right-censored", timescale = "log", numberoftrees = 200, k = 0.1, Npost = 5000, Nburn = 5000, storeposterior_sample = TRUE )

Posterior mean CATEs

CATEhorseshoe <- colMeans(fithorseshoe$trainpredictionssample_treat)

Posteriors of the ATE

postATEhorseshoe <- rowMeans(fithorseshoe$trainpredictionssampletreat)

Posterior mean ATE

ATEhorseshoe <- mean(postATE_horseshoe)

Plot the posterior of the ATE

```

📄 Documentation

• In R: ?ShrinkageTrees, ?HorseTrees, ?CausalHorseForest, and ?CausalShrinkageForest for detailed help.
• Examples and parameter descriptions can be found in each function’s documentation.

🤝 Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request. The software is designed to be flexible and modular, allowing for a wide variety of global-local shrinkage priors to be easily implemented and extended in future versions.

📜 License

This package is licensed under the MIT License.

🇪🇺 Acknowledgments

This project has received funding from the European Research Council (ERC) under the European Union’s Horizon Europe program under Grant agreement No. 101074082. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them