skdeshpande91.r-universe.dev

[skdeshpande91] flexBART 2.0.5

sameer.deshpande@wisc.edu (Sameer K. Deshpande) — Tue, 12 May 2026 12:41:51 GMT

Implements a faster and more expressive version of Bayesian Additive Regression Trees that, at a high level, approximates unknown functions as a weighted sum of binary regression tree ensembles. Supports fitting (generalized) linear varying coefficient models that posits a linear relationship between the inverse link and some covariates but allows that relationship to change as a function of other covariates. Additionally supports fitting heteroscedastic BART models, in which both the mean and log-variance are approximated with separate regression tree ensembles. A formula interface allows for different splitting variables to be used in each ensemble. For more details see Deshpande (2025) and Deshpande et al. (2026) .

[skdeshpande91] VCBART 1.2.5

sameer.deshpande@wisc.edu (Sameer K. Deshpande) — Tue, 12 May 2026 01:39:21 GMT

Fits linear varying coefficient (VC) models, which assert a linear relationship between an outcome and several covariates but allow that relationship (i.e., the coefficients or slopes in the linear regression) to change as functions of additional variables known as effect modifiers, by approximating the coefficient functions with Bayesian Additive Regression Trees. Implements a Metropolis-within-Gibbs sampler to simulate draws from the posterior over coefficient function evaluations. VC models with independent observations or repeated observations can be fit. For more details see Deshpande et al. (2026) .

[skdeshpande91] flexBCF 1.0.3

sameer.deshpande@wisc.edu (Sameer K. Deshpande) — Mon, 01 Dec 2025 14:36:30 GMT

A faster implementation of Bayesian Causal Forests (BCF; Hahn et al. (2020) ), which uses regression tree ensembles to estimate the conditional average treatment effect of a binary treatment on a scalar output as a function of many covariates. This implementation avoids many redundant computations and memory allocations present in the original BCF implementation, allowing the model to be fit to larger datasets. The implementation was originally developed for the 2022 American Causal Inference Conference's Data Challenge. See Kokandakar et al. (2023) for more details.