Movatterモバイル変換

dgpsi

The R packagedgpsi provides R interface to Pythonpackagedgpsifor deep and linked Gaussian process emulations using stochasticimputation (SI).

Hassle-free Python Setup
You don’t need prior knowledge of Python to start using the package, allyou need is a single click in R (seeInstallation section below) that automaticallyinstalls and activates the required Python environment for you!

Features

dgpsi currently has following features:

• Gaussian process emulations with separable or non-separable squaredexponential and Matérn-2.5 kernels.
• Deep Gaussian process emulations with flexible structures including:
  • multiple layers;
  • multiple GP nodes;
  • separable or non-separable squared exponential and Matérn-2.5kernels;
  • global input connections;
  • non-Gaussian likelihoods (Poisson, Negative-Binomial,heteroskedastic Gaussian, and Categorical).
• Linked emulations of feed-forward systems of computer models bylinking (D)GP emulators of deterministic individual computermodels.
• Fast Leave-One-Out (LOO) and Out-Of-Sample (OOS) validations for GP,DGP, and linked (D)GP emulators.
• Multi-core predictions and validations for GP, DGP, and Linked (D)GPemulators.
• Sequential designs for (D)GP emulators and bundles of (D)GPemulators.
• Automatic pruning of DGP emulators, both statically anddynamically.
• Large-scaleGP, DGP, and Linked (D)GP emulations.
• ScalableDGP classification using Stochastic Imputation.
• Bayesianoptimization.

Getting started

• CheckA QuickGuide to dgpsi to get started with the package.
• For experimental features, check out ourwebsite for thedevelopment version.

Installation

You can install the package from CRAN:

install.packages('dgpsi')

or its development version from GitHub:

devtools::install_github('mingdeyu/dgpsi-R')

After the installation, run

library(dgpsi)

to load the package. To install or activate the required Pythonenvironment automatically, you can either rundgpsi::init_py() explicitly or simply call any functionfrom the package. That’s it - the package is ready to use!

Note
After loadingdgpsi, the package may take some time tocompile and initiate the underlying Python environment the first time afunction fromdgpsi is executed. Any subsequent functioncalls won’t require re-compiling or re-activation of the Pythonenvironment, and will be faster.

If you experience Python related issues while using the package,please try to reinstall the Python environment:

dgpsi::init_py(reinstall = T)

Or uninstall completely the Python environment:

dgpsi::init_py(uninstall = T)

and then reinstall:

dgpsi::init_py()

Research Notice

This package is part of an ongoing research initiative. For detailedinformation about the research aspects and guidelines for use, pleaserefer to ourResearchNotice.

References

Ming, D. and Williamson,D. (2023) Linked deep Gaussian process emulation for model networks.arXiv:2306.01212

Ming, D.,Williamson, D., and Guillas, S. (2023) Deep Gaussian process emulationusing stochastic imputation.Technometrics. 65(2),150-161.

Ming, D. and Guillas, S.(2021) Linked Gaussian process emulation for systems of computer modelsusing Matérn kernels and adaptive design,SIAM/ASA Journal onUncertainty Quantification. 9(4), 1615-1642.