Here are some contributions to software development (R packages) in computer experiments, sensitivity analysis and Poincaré-type functional inequalities.
Two R packages for the analysis and optimization of costly simulators, co-authored with D. Ginsbourger, Y. Deville, and V. Picheny (for DiceOptim), with contributions by M. Binois, C. Chevalier, S. Marmin, Y. Richet, T. Wagner…
- Download (on CRAN): DiceKriging, DiceOptim
- Related publications: JSS article, update
- Demonstrators with Shiny (by Y. Richet):
Customizing kernels with kergp
kergp is an experimental R package co-authored with Y. Deville and D. Ginsbourger for defining its own kernel. Useful functionalities are implemented including maximum likelihood parameter estimation, prediction, leave-one-out validation. Some predefined classes are implemented such as radial kernels or categorical kernels, and operations on kernels are possible via a formula mechanism.
DiceDesign is a R package for designing and evaluating computer experiments, co-authored with J. Franco, D. Dupuy, G. Damblin, B. Iooss. I implemented the radial scanning statistic, which angularly scans space-filling designs in order to detect undesirable patterns (such as alignments).
fanovaGraph is a R package co-authored with J. Fruth, T. Muehlenstaedt and M. Jastrow devoted to block-additive kernels. They are useful to discover and exploit additive structures in multivariate functions.
sensitivity, Poincaré constants
sensitivity is a R package for quantifying the influence of input variables on an output of interest, authored by B. Iooss, A. Janon and G. Pujol. I implemented functions for support analysis, helpful for choosing the boundaries of the input domain, and a function computing optimal Poincaré constants, useful for screening by using the information given by derivatives.