GPmp-contrib: computer experiments and sequential design

Website and documentation | Examples | API reference | PyPI | GPmp

GPmp-contrib provides computer-experiment objects, multi-output model containers, Matérn model classes, sequential-design procedures, set-estimation methods, plots, and relaxed Gaussian-process utilities. It builds these objects on the models, kernels, parameter-selection functions, diagnostics, and numerical backends provided by GPmp.

When to use GPmp-contrib

Use gpmp-contrib when you need a ComputerExperiment, a ModelContainer, provided Matérn model classes, sequential strategies, Bayesian optimization, excursion-set estimation, set inversion, or reGP.

Use gpmp directly when you need the numerical GP model, covariance functions, parameter-selection functions, diagnostics, posterior samplers, or plotting helpers without computer-experiment or sequential-design objects.

Core features

• Computer experiments: input boxes, normalized inputs, objectives, constraints, and multi-output evaluations.
• Model containers: one gpmp.core.Model per output, with parameter selection, prediction, diagnosis, and readable parameter objects.
• Matérn model classes: fixed half-integer regularity (Maternp) or jointly selected regularity nu > 0 (Matern). The available classes cover ML, REML, REMAP, and noisy observations.
• REMAP priors: inspect and modify resolved prior parameters with get_prior(...) and set_prior(...).
• Sequential strategies: fixed candidate grids, SMC particle sets, and BSS-style particle sets.
• Optimization and set estimation: expected improvement, excursion sets, set inversion, and Pareto utilities.
• Relaxed Gaussian processes: reGP procedures for threshold-oriented prediction problems.
• Posterior parameter sampling: access GPmp's MH, HMC, NUTS, and SMC samplers through ModelContainer.sample_parameters(...).

Numerical backends

GPmp-contrib uses the numerical backend selected by GPmp:

• NumPy: often fast for small-to-medium exact GP computations.
• PyTorch: provides automatic differentiation and is useful when gradient information is needed, especially in higher-dimensional parameter settings.

Set the backend before importing either package:

export GPMP_BACKEND=torch
export GPMP_DTYPE=float64

Model and kernel computations use gpmp.num backend objects. High-level ModelContainer methods convert inputs and outputs where documented in the API reference.

Package split

gpmp
  core GP model
  covariance functions
  parameter objects and selection
  diagnostics
  posterior samplers
  plotting helpers

gpmp-contrib
  computer experiments
  model containers and provided Matérn classes
  sequential strategies
  Bayesian optimization
  excursion-set estimation
  set inversion
  reGP utilities

Install

pip install gpmp-contrib
python -c "print(__import__('gpmpcontrib').__version__)"

The verification command prints the installed GPmp-contrib version. The installation also installs gpmp and the other runtime dependencies declared in pyproject.toml.

For local development:

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

Use pip install -e ".[docs]" for documentation tools. When testing against a local GPmp checkout, install that checkout first and then install gpmp-contrib in editable mode.

GPmp-contrib requires Python 3.9 or later, GPmp 0.9.38 or later, NumPy, SciPy, and Matplotlib. PyTorch is optional and is installed separately when automatic differentiation is needed.

Documentation

The documentation is available at https://gpmp-dev.github.io/gpmp-contrib/. It includes a complete Hartmann4 example, model construction and state, parameter selection, diagnostics, sequential design, excursion sets, set inversion, reGP, and the public API.

To build it locally:

pip install -e ".[docs]"
python docs/make_example_results.py
sphinx-build -M html docs/source docs/_build -E

Public API

The intended public API is organized around:

• gpmpcontrib.ComputerExperiment
• gpmpcontrib.modelcontainer
• the model classes exported by gpmpcontrib
• gpmpcontrib.SequentialPrediction
• gpmpcontrib.SequentialStrategyGridSearch
• gpmpcontrib.SequentialStrategySMC
• gpmpcontrib.SequentialStrategyBSS
• gpmpcontrib.samplingcriteria
• gpmpcontrib.optim
• gpmpcontrib.regp
• gpmpcontrib.test_problems

How to cite

If you use GPmp-contrib in research, please cite:

@software{gpmpcontrib2026,
  author       = {Emmanuel Vazquez},
  title        = {GPmp-contrib},
  year         = {2026},
  url          = {https://github.com/gpmp-dev/gpmp-contrib},
  note         = {Version 0.9.38},
}

Update the version number when citing another release.

Minimal example

The basic sequence is: choose a computer experiment, build a model container, select covariance parameters, predict, and inspect the result.

import gpmp as gp
import gpmp.num as gnp
import gpmpcontrib as gpc

gnp.set_seed(1234)

problem = gpc.test_problems.hartmann4
xi = gp.misc.designs.ldrandunif(problem.input_dim, 40, problem.input_box)
zi = problem(xi)
xt = gp.misc.designs.ldrandunif(problem.input_dim, 300, problem.input_box)

model = gpc.Model_ConstantMean_Maternp_REML(
    "hartmann4",
    output_dim=problem.output_dim,
    mean_specification={"type": "constant"},
    covariance_specification={"p": 3},
)

model.select_params(xi, zi)
zpm, zpv = model.predict(xi, zi, xt)
model.run_diagnosis(xi, zi)

The final call prints parameter-selection, parameter, and observation summaries. See the getting-started example for prediction checks, performance measures, stored model state, and the corresponding figure.

Authors

See AUTHORS.md for details.

License

GPmp-contrib is free software released under the GNU General Public License v3.0 or later. See LICENSE for details.