GPmp Documentation

GPmp provides building blocks for Gaussian-process (GP) interpolation and regression, also called kriging: mean functions, covariance functions, covariance-parameter selection, diagnostics, conditional simulation, and posterior sampling of covariance parameters.

The package is meant for GP-based algorithms and research software. Its API is small and explicit: users provide the mean and covariance functions, choose or define selection criteria, inspect diagnostics, and keep numerical backend objects visible through gpmp.num. The backend can be either NumPy or PyTorch.

Features

  • GP interpolation and regression with known or unknown mean functions.

  • Maximum likelihood, restricted maximum likelihood, REMAP, and custom parameter-selection criteria.

  • Posterior parameter sampling with MH, NUTS, and SMC samplers.

  • Leave-one-out diagnostics, model reports, and selection-criterion plots.

  • Conditional sample paths.

  • Classical covariance functions and a callable interface that makes new covariance functions easy to add.

  • Matérn covariance functions with fixed half-integer regularity or a continuous regularity parameter \(\nu > 0\), which can be selected jointly with variance and lengthscales.

Positioning

GPmp is intended for research and for integration into GP-based algorithms. It favors a small, explicit API over a broad catalog of model classes. Modeling assumptions and numerical operations remain visible in the code, and the components can be inspected and modified. GPmp currently focuses on exact GP computations. Large-scale approximate inference is not implemented.

The companion package gpmp-contrib builds on GPmp to provide model containers and complete procedures for computer experiments, including Bayesian optimization, excursion-set estimation, set inversion, and sequential design.

Installation from source

Editable installation requires a local clone of the repository. Clone GPmp, enter the repository root, then run pip install -e .:

git clone https://github.com/gpmp-dev/gpmp.git
cd gpmp
pip install -e .

Documentation contents

  • GPmp Tutorial builds and diagnoses a Hartmann4 interpolation model.

  • Examples contains rendered scripts for interpolation, regression, Matérn covariance functions, Matérn regularity selection, posterior sampling, sample paths, and dataloader-based selection.

  • API reference documents the public API, including backend objects, core models, kernels, parameter objects, diagnostics, samplers, optional dataloaders, plotting functions, designs, and test functions.

  • References lists the literature cited by the tutorial and examples.

How to Cite

@software{gpmp2026,
  author       = {Emmanuel Vazquez},
  title        = {GPmp: the Gaussian Process micro package},
  year         = {2026},
  url          = {https://github.com/gpmp-dev/gpmp},
  note         = {Version 0.9.37},
}

Authors

See AUTHORS.md.

License

GPmp is free software released under the GNU General Public License v3.0. See LICENSE.txt for details.