libKriging

libKriging is a C++ library for Kriging / Gaussian process regression.

Main features of libKriging are:

• Standard implementations of the most common Kriging models:

  • ordinary / universal kriging

  • noise-free interpolation with Kriging(noise = NULL)

  • homogeneous nugget estimation with Kriging(noise = "nugget")

  • known heteroskedastic noise with Kriging(noise = <variance vector>)

  • hyper-parameter optimisation based on log-likelihood, leave-one-out, or log-marginal-posterior

  • optional normalisation of conditional data

• Kriging with per-variable input warping (WarpKriging):

  • continuous warps: affine, Box-Cox, Kumaraswamy, monotone neural net, free MLP

  • categorical embedding and ordinal level warps for discrete inputs

  • joint optimisation of warp and GP hyper-parameters

• Deep kernel learning with MLPKriging:

  • shared MLP feature extractor over all inputs

  • nonlinear latent feature space before GP evaluation

• Ports and comparisons with established Kriging libraries:

  • https://CRAN.R-project.org/package=DiceKriging

  • https://CRAN.R-project.org/package=RobustGaSP

  • https://github.com/stk-kriging

• Compatibility with common OS / architecture targets:

  • Windows

  • Linux

  • macOS (Intel and Apple Silicon)

• Wrappers for:

  • Python: https://pypi.org/project/pylibkriging/

  • R: https://github.com/libKriging/rlibkriging

  • Octave

  • Matlab

  • Julia

Check out the Usage section for further information, and how to Installation the project.

Note

This project is under active development.

Contents

• Installation
• Usage
• API
• Mathematical Background
• References