Learn R Programming

⚠️There's a newer version (1.1.7) of this package.Take me there.

hetGP (version 1.1.4)

Heteroskedastic Gaussian Process Modeling and Design under Replication

Description

Performs Gaussian process regression with heteroskedastic noise following the model by Binois, M., Gramacy, R., Ludkovski, M. (2016) , with implementation details in Binois, M. & Gramacy, R. B. (2021) . The input dependent noise is modeled as another Gaussian process. Replicated observations are encouraged as they yield computational savings. Sequential design procedures based on the integrated mean square prediction error and lookahead heuristics are provided, and notably fast update functions when adding new observations.

Copy Link

Version

Install

install.packages('hetGP')

Monthly Downloads

659

Version

1.1.4

License

LGPL

Maintainer

Mickael Binois

Last Published

July 8th, 2021

Functions in hetGP (1.1.4)

Import and export of hetGP objects

Generic Log-likelihood function This function can be used to compute loglikelihood for homGP/hetGP models

Compute double integral of the covariance kernel over a [0,1]^d domain

Noisy 1d test function (1) Add Gaussian noise with variance r(x) = scale * (1.1 + sin(2 pi x))^2 to f1d

Gaussian process predictions using a homoskedastic noise GP object (of class homGP)

SIR test problem

Noisy 1d test function (2) Add Gaussian noise with variance r(x) = scale * (exp(sin(2 pi x)))^2 to f1d2

Student-t process predictions using a heterogeneous noise TP object (of class hetTP)

Allocation of replicates on existing designs

IMSPE optimization

Leave one out predictions

Likelihood-based comparison of models

Integrated Contour Uncertainty criterion

Sequential IMSPE criterion

Data preprocessing

Bayes Factor Data

Maximum Contour Uncertainty criterion

Maximum Empirical Error criterion

Assemble To Order (ATO) Data and Fits

pred_noisy_input

Gaussian process prediction prediction at a noisy input x, with centered Gaussian noise of variance sigma_x. Several options are available, with different efficiency/accuracy tradeoffs.

Gaussian process modeling with homoskedastic noise

Parallel Expected improvement

Student-t process predictions using a homoskedastic noise GP object (of class homGP)

t-MSE criterion

Correlation function of selected type, supporting both isotropic and product forms

Student-T process modeling with homoskedastic noise

Update "hetGP"-class model fit with new observations

Score and RMSE function To asses the performance of the prediction, this function computes the root mean squared error and proper score function (also known as negative log-probability density).

Gaussian process predictions using a heterogeneous noise GP object (of class hetGP)

deriv_crit_IMSPE

Derivative of crit_IMSPE

Derivative of EI criterion for GP models

Expected Improvement criterion

Update "hetTP"-class model fit with new observations

1d test function (1)

Gaussian process modeling with heteroskedastic noise

1d test function (2)

Student-t process modeling with heteroskedastic noise

Fast homGP-update

Fast homTP-update

Integrated Mean Square Prediction Error

Criterion optimization

Contour Stepwise Uncertainty Reduction criterion