New features

• Newlink argument added to theCategorical class to support binary classification via the'probit' link function, in addition to'logit'.
• Platform-specific Conda environment YAMLs added for easier installation of the development version.
• UpdatedREADME.md to reflect the revised installation instructions for the development version from GitHub using the added Conda environment YAMLs.
• The package now compatible with Python 3.10-3.12 on Conda and Python 3.9-3.12 on PyPI.

Changes

• Enhanced prediction efficiency for small test datasets by reducing overhead from multi-threading.
• Up to ~30× faster and morenumerically stable DGP emulation with heteroskedastic likelihoods and replicates, with or without the Vecchia approximation.
• Improved initialization of DGP emulators with Poisson, heteroskedastic, and categorical likelihoods for more robust performance.
• The formerclassify method has been merged intopredict for DGP emulators with categorical likelihoods.
  • predict now supports both mean-variance and sample-based class probability predictions.
• UpdatedDGP classification demo to reflect the new API and functionality.
• Significant inference speed-up (~10×+) for GPs and DGPs with homogeneous noise and replicates.
• Various bug fixes and stability improvements.