Mapping-Based Additive Gaussian Process Models
Overview
Fits mapping-based additive Gaussian process models for experiments in which each component has both a quantitative level and a position in an ordered sequence. Two model structures are available: a compact two-dimensional mapping and a full mapping with one fewer dimension than the number of components. Both models support parameter estimation, point prediction, and plug-in predictive uncertainty. Input checks validate the sequence data and apply consistent scaling to the quantitative inputs. Computationally intensive covariance and gradient calculations are implemented in C++ with 'Rcpp'. Initial-design functions combine a space-filling Latin hypercube with sequence permutations. The sequence portion can be generated randomly or optimized with simulated annealing or space-filling threshold accepting. Expected improvement can be optimized over both parts of the input, and a sequential interface supports Bayesian optimization of an expensive user-supplied objective. An integrated workflow can generate the initial design, evaluate the objective, and continue the sequential search in one call. The model was introduced by Xiao et al. (2024) doi:10.1080/01621459.2022.2123335.
- Limited parallel fitting and acquisition searches to two worker processes.
- Added random and space-filling threshold-accepting options for the sequence portion of an initial design. The existing simulated-annealing method remains the default for backward compatibility.
- Added C++ incremental scoring for the threshold-accepting search, including bounded initialization, threshold calibration, best-result retention, and detailed diagnostics.
- Connected the new sequence choices to the full initial-design and from-scratch Bayesian-optimization workflows.
- Added examples and regression tests for method selection, input validation, reproducibility, finite permutation spaces, and quantitative-design preservation.
Show all changes in 0.12.0Hide the rest of the changes in 0.12.0
- Rewrote the Bayesian optimization help pages and examples to explain function choice, arguments, workflow steps, and returned objects more directly.
- Added a pkgdown website with a getting-started guide, an optimization workflow, a reproducible covariance-engine benchmark, and a curated function reference.
- Added package-level citation metadata for both the software and the associated methodology paper.
- Added
magp_bayes_optimize_from_scratch()to connect simulated-annealing initial design, initial objective evaluation, and sequential MaGP Bayesian optimization in one call. - Added reproducible handling of stochastic objective functions while preserving the caller's random-number state.
- Included the complete initial design and its responses in the optimization result.
- Added early validation for design, fitting, acquisition, and objective controls before the initial-design search begins.
- Added expected improvement for both two-dimensional and full-mapping models.
- Added mixed acquisition search across quantitative bounds and sequence permutations, with multi-start and optional parallel execution.
- Added a sequential Bayesian optimization interface for expensive objective functions.
- Added deterministic sequence sampling for cases where complete permutation enumeration would be too large.
- Added duplicate-input protection, early stopping, complete search history, and final-model refitting.
- Added multi-start optimization to
magp2d_fit()andmagpfull_fit(). - Added optional local parallel execution through portable socket clusters.
- Made seeded sequential and parallel fits use the same parameter starts.
- Added start-level objective values, convergence results, warnings, and errors to each fitted object.
- Completed the quantitative-sequence initial-design workflow.
- Added
magp_quantitative_criterion()andmagp_quantitative_design()for maximin-style Latin hypercubes. - Added
magp_joint_criterion()to evaluate the combined separation of quantitative and sequence inputs. - Added
magp_initial_design()to construct and align both design portions while preserving their individual structures. - Added reproducible simulated-annealing searches and validation across several run sizes and component counts.
- Added tools for constructing the sequence portion of a quantitative-sequence initial design.
- Added
magp_sequence_criterion()to assess pair balance and Hamming-distance space filling in a sequence design. - Added
magp_sequence_design()to optimize the sequence portion with reproducible simulated annealing. - Added reproducible simulated-annealing search with configurable criterion weights and search settings.
From NEWS.md in the release
Install
Health
- OK13 OKNow
- OK7 OK2026-09-04
Documentation
- Help pages
- 22 (1 internal)includes a package overview page
- Examples that run
- 46%6 run in a plain check · 7 only when donttest examples are enabled (R CMD check --as-cran)
- Exported functions with examples
- 92%
- Documented parameters
- 92%
- Return-value docs
- 100%
- Help pages citing references
- 32%
Downloads
Repository
- PRs merged (all time)
- 100%
- Median days to close, 50 most recently updated closed issues
- not measured
- Median days to close, 50 most recently updated closed PRs
- 0
Commits over time
7 development-tooling and community-health practices detected across 4 families in the upstream repository
Checks run against github.com/twskr/magp on 2026-10-11.
The R CMD check workflow runs on Linux, macOS and Windows, and on R-devel
GitHub Pages site last deployed 2026-10-06 · The repository has a site deploy workflow
Dependencies
Code & Tests
Datasets
People & History
Pick two releases to compare their code metrics.
- 0.12.0Latest2026-09-24 · current release · diff ↗ ·
added 10 exports
magp_bayes_optimize, magp_bayes_optimize_from_scratch, magp_expected_improvement, magp_initial_design, magp_joint_criterion, magp_next_point, magp_quantitative_criterion, magp_quantitative_design, magp_sequence_criterion, magp_sequence_design - 0.8.02026-09-03
- RR 4.6.0 released · 2026-04-24
Package metadata
- First published
- 2026-09-03
- Total releases
- 2 / 1 yrs
- License
- MIT + file LICENSE OSI
- Download size
- 283 KB
Cite
Cite this package
Run in R for the authors' preferred citation:
citation("magp")The package provides a citation file that builds its entry when R runs it. Run citation() to see it.
The repository also keeps a CITATION.cff.
Cite the R Observatory
For a number measured here: a download total, a coverage figure, an archival date.
From data release v2026-10-18, which the citation names so these numbers can be found later. More on citing and the projects behind them.