NEWS

---

shrinkGPR 2.0.1

• Updated README.md to be more informative, as well as include multivariate case
  • Fixed an issue with the distribution of the latent variables in shrinkMVGPR causing lack of convergence in some cases
  • Fixed a bug that didn't allow for covariates in the mean formula to be used in gen_marginal_samples()

shrinkGPR 2.0.0 (2026-03-30)

• Potentially breaking change: Changed name of global shrinkage parameter from lambda to tau, as to be more consistent with references
  • Added support for multivariate GPR and TPR under shrinkage
  • Added method to save and load model objects
  • Overhauled documentation to be more informative, including model specification
  • Fixed various bugs
  • Fixed issues on CRAN with examples being run when torch is not installed

shrinkGPR 1.1.3

• Fixed issues on CRAN with examples being run when torch is not installed

shrinkGPR 1.1.2

• Moved to torch for R version >= 0.16.0
  • Made training loop more numerically robust by falling back on robust Cholesky decomposition
  • Fixed error in documentation for sylvester flows
  • Fixed issues on CRAN with examples being run when torch is not installed

shrinkGPR 1.1.1 (2025-10-01)

• Fixed typo in prior on global shrinkage parameters
  • Fixed error when generating posterior samples with (optional) linear mean equation
  • Wrapped JIT compilation in torch::torch_is_installed() check to avoid errors when torch is not installed

shrinkGPR 1.1 (2025-08-19)

• Added student-t process as in Shah et al. (2014) as alternative to Gaussian process
  • Added option to estimate simulated data with multicollinearity
  • Added ability to estimate marginal effects of one or two parameters and associated plotting functions
  • Increased speed of estimation through use of JIT compiled torchscript
  • Changed default number of householder transformations in Sylvester flows to be min(5, d-1)
  • Corrected documentation for mean equation in shrinkGPR
  • Fixed bug in TG/NGG density in GPR_class