dgpsi v2.5.0

New features

• A Vecchia approximation has been implemented, enabling efficient large-scale emulation for all package functionalities.
• Flexibly switch between Vecchia and non-Vecchia modes of emulators built with the Vecchia approximation.
• Added support for both binary and multi-class DGP classifications through the Categorical class.
• The new log_loss() method of the emulator class allows you to evaluate the performance of constructed classifiers.

Changes

• Training times for DGP emulators are now 30%-40% faster.
• Computing (D)GP predictions and Leave-One-Out (LOO) evaluations is now 6-7 times faster.
• The nb_parallel argument has been removed, as multi-threading is now seamlessly integrated by default.
• Two new demos showcasing the new functionalities:
  • Large-Scale Emulation with Vecchia Approximation
  • DGP Classifications
• Various bug fixes and stability improvements.