Dear Reader,
You will find in the associated text files ("mpdp-commandlinesR.txt", "mpdp-routinesR.txt" and "flr-routinesR.txt")
the materials needed for implementing the statistical methodology detailed in the paper entitled
"Most predictive design points for functional data predictors" and co-authored by Frederic Ferraty, Peter Hall,
Philippe Vieu (to appear in Biometrika).
The file "mpdp-routinesR.txt" contains R routines:
1) "mpdp" ==> stepwise algorithm (forward selection + backward deletion) allowing the selection of the
most predicitve design points
2) "predict.mpdp" ==> allows to predict responses from an object of class "mpdp"
3) "quadratic2" ==> asymmetrical quadratic kernel
The file "flr-routinesR.txt" contains R routines ("Splinemlf", "Bspline.ini", "approx.spline.deriv",
"interp.spline.deriv") allowing the implementation of the alternative functional linear regression
method (in order to compare with)
The file "mpdp-commandlinesR.txt" contains all R commandlines allowing to implement the stepwise algorithm
(i.e. selection of most predictive design points) on simulations and real datasets.
How to use this material?
1) open an R session
2) follow the instructions given in the file "mpdp-commandlinesR.txt" (copy-paste the R commandlines into your R session)
IMPORTANT REMARK: this is an intensive computational method; the computational labour may become very huge
with large dataset (one run with chopped pork dataset - 159 curves with 100 design points - lasts about
5 minutes on a standard personal computer; one run with orange juice dataset - 149 curves with 700 design points -
lasts about 30 minutes).
All this material is freely downloadable from the NPFDA (NonParametric Functional Data Analysis) website
"http://www.math.univ-toulouse.fr/staph/npfda" which contains materials for implementing various nonparametric
statistical methods for functional data analysis. Use for any commercial purpose is forbidden.