Heteroskedastic Gaussian Process Modeling and Design under Replication

Performs Gaussian process regression with heteroskedastic noise following Binois, M., Gramacy, R., Ludkovski, M. (2016) . 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.


News

hetGP 1.1.1

change / hetGP 1.1.0

  • add vignette describing the package
  • precisions on noise.var with predict.hetGP in the documentation
  • change print method to return summary
  • add plot methods to GP and TP models
  • catch optim errors when at least one likelihood evaluation worked
  • lower and upper are now optional parameters
  • IMSPE include the extra term with estimated beta0 (but not crit_IMSPE at this stage)
  • IMSE.search and IMSE_nsteps_ahead are now a single function: IMSPE_optim
  • update defaults of IMSE.search
  • new functions crit_optim mimicking IMSPE_optim for contour/optim criteria
  • catch numerical errors in updated variance in contour criteria
  • improve partial derivatives wrt theta for Matern kernels
  • add time slot in model fitting outputs
  • faster trace computations in the symmetric case
  • initial theta values when providing lower and upper is now 10% of the range
  • fix strip and crit_EI methods (thanks to Rob Smith)

hetGP 1.1.0

change / hetGP 1.0.2

  • better handling of the penalty term
  • update the documentation, rename nu2_hat by nu_hat for consistency with JCGS paper
  • several performance improvements (e.g., predict with xprime argument)
  • contour finding infill criteria related to arXiv:1807.06712
  • add Expected Improvement criterion (beta version)
  • correct bug with known mean not used for initialization with mleHetGP/TP
  • add leave one out predictions

hetGP 1.0.2

change / hetGP 1.0.1

  • homTP/heTP update methods
  • option settings$return.hom to return or not modHom models with hetGP and hetTP
  • new functions to export/import hetGP objects, including a robust re-computation of inverse covariance matrices
  • accordingly, predict now only threshold negative predictive variances occuring due to numerical instability with Cholesky inverse
  • better initialization with mleHetTP
  • improved warnings and error handling

hetGP 1.0.1

change / hetGP 1.0.0

  • add Student-t modeling (equivalent of mleHetGP and mleHomGP, with predict functions)
  • various typos and bugs corrections
  • now works on Solaris

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("hetGP")

1.1.1 by Mickael Binois, 3 months ago


Browse source code at https://github.com/cran/hetGP


Authors: Mickael Binois , Robert B. Gramacy


Documentation:   PDF Manual  


LGPL license


Imports Rcpp, MASS, methods, DiceDesign

Suggests knitr, monomvn, lhs

Linking to Rcpp


See at CRAN