# Olivier Roustant

#### 4 packages on CRAN

Estimation, validation and prediction of kriging models. Important functions : km, print.km, plot.km, predict.km.

Gaussian process regression with an emphasis on kernels. Quantitative and qualitative inputs are accepted. Some pre-defined kernels are available, such as radial or tensor-sum for quantitative inputs, and compound symmetry, low rank, group kernel for qualitative inputs. The user can define new kernels and composite kernels through a formula mechanism. Useful methods include parameter estimation by maximum likelihood, simulation, prediction and leave-one-out validation.

Estimation and plotting of a function's FANOVA graph to identify the interaction structure and fitting, prediction and simulation of a Kriging model modified by the identified structure. The interactive function plotManipulate() can only be run on the RStudio IDE with RStudio's package 'manipulate' loaded. RStudio is freely available (www.rstudio.org), and includes package 'manipulate'. The equivalent function plotTk() bases on CRAN Repository packages only.