Skip to contents

Changelog

Source: NEWS.md

refundBayes 0.6.0

CRAN release: 2026-05-08

New features

  • Added fpca_bayes() for Bayesian Functional Principal Component Analysis, modelling a functional outcome as μ(t) plus a low-rank FPC expansion with posterior inference on the mean function, FPC scores, eigenvalue standard deviations, and the residual SD. Initial eigenfunctions are obtained from refund::fpca.sc() and held fixed during sampling.
  • Added a joint_FPCA argument to sofr_bayes(), fcox_bayes(), and fofr_bayes() for jointly modelling each functional predictor via FPCA alongside the regression coefficients. When enabled, the predictor is replaced by an FPCA representation and FPC scores are sampled jointly with β(·), propagating measurement-error uncertainty into the posterior of the regression coefficient (errors-in-variables-aware fit).

Documentation

  • Added a dedicated vignette for each new feature: Bayesian Functional Principal Component Analysis and Joint FPCA Modeling in refundBayes (covering joint FPCA usage in SoFR, FCox, and FoFR).
  • Expanded the pkgdown site to include reference entries and articles for fpca_bayes() and the Joint-FPCA option.
  • Annotated the Quick Start example in README.md with inline comments explaining the formula syntax and sampler arguments.
  • Aligned vignette YAML titles with their \VignetteIndexEntry to silence rmarkdown::html_vignette title-mismatch warnings.

Dependencies

  • Removed brms and dplyr from Imports. The two brms::brmsformula() call-sites were replaced with stats::as.formula(), and the .data pronoun used in ggplot calls is already re-exported by ggplot2. This trims the install dependency tree noticeably (brms transitively pulled in posterior, bridgesampling, loo, bayesplot, etc.).

refundBayes 0.5.1

CRAN release: 2026-04-07

New features

  • Added fofr_bayes() for Bayesian Function-on-Function Regression (FoFR), supporting functional responses with functional and scalar predictors. The bivariate coefficient surface β(s, t) is represented via a tensor-product basis with dual-direction smoothness (random-effect reparameterisation in the predictor direction and a penalty-matrix prior in the response direction).
  • Added a dedicated vignette Bayesian Function-on-Function Regression with the full model specification, prior table, and a worked example.
  • Expanded the pkgdown site to include the new FoFR reference entry and vignette.

Documentation

  • Rewrote README.md: added a supported-models table, links to per-function vignettes, a citation to Jiang et al. (2025, Statistics in Medicine), and CRAN status / downloads badges.
  • Minor fixes to fcox_bayes() examples so that pkgdown can parse and render the reference page.

Internal

  • Added a standalone Stan program (Simulation/StanFoFR_Gaussian.stan) and a formal simulation script (Simulation/FoFR_Simulation.R) for reproducible FoFR benchmarking without recompiling Stan code via refundBayes at every run.

refundBayes 0.5.0

Initial public release of refundBayes, a package providing a convenient interface for Bayesian functional regression using Stan. The package is designed to mirror the mgcv::gam formula syntax familiar to users of refund, while delivering full Bayesian posterior inference via rstan.

Supported models

  • sofr_bayes() — Bayesian Scalar-on-Function Regression, supporting Gaussian, binomial, and Poisson families, with one or more functional predictors alongside scalar covariates.
  • fosr_bayes() — Bayesian Function-on-Scalar Regression with FPCA-based residual structure for modelling subject-level functional deviations.
  • fcox_bayes() — Bayesian Functional Cox Regression for time-to-event outcomes with functional and scalar predictors, including posterior inference on the log-hazard ratio surface.

Modelling framework

  • Functional coefficients represented via spline bases constructed through mgcv::smoothCon(), with spectral reparameterisation (mgcv::smooth2random()) into fixed and random effect components.
  • Non-centered parameterisation of penalised spline coefficients for efficient HMC sampling.
  • Inverse-gamma priors on smoothing variance components and weakly informative priors on fixed-effect coefficients.
  • Support for multiple spline bases (cubic regression, thin-plate, P-spline, etc.) via the bs argument in the formula interface.
  • Generic summary() and plot() methods for posterior summaries and visualisation of functional coefficients with credible bands.

Documentation and infrastructure

  • Per-function vignettes illustrating model specification, prior choices, fitting, and post-processing for SoFR, FoSR, and FCox.
  • Example datasets (example_data_sofr, example_data_FoSR, example_data_Cox) shipped with the package.
  • pkgdown documentation site published at https://zirenjiang.github.io/refundBayes/.
  • Companion tutorial paper: Jiang, Crainiceanu, and Cui (2025), Tutorial on Bayesian Functional Regression Using Stan, Statistics in Medicine, 44(20–22), e70265, doi:10.1002/sim.70265.