[R-pkgs] new package benchden 1.0.0 : benchmark densities for nonparametric density estimation

Thoralf Mildenberger mildenbe at statistik.uni-dortmund.de
Thu Jun 28 15:26:28 CEST 2007


The new package "benchden" 1.0.0 implements 28 benchmark densities for
nonparametric density estimation that were introduced by A. Berlinet and
L. Devroye ("A Comparison of Kernel Density Estimates", Pub. Inst. Stat.
Univ. Paris, XXXVIII, fasc. 3, 1994, 3-59,
http://cg.scs.carleton.ca/~luc/devs.html ). This collection includes a
variety of densities with different degrees of smoothness, different
tail behaviour, different number of modes, with and without infinite 
peaks and also some standard densities like the normal and the uniform. 
There is also a small intersection (e.g. the claw density) with the 
collection of normal mixtures introduced by Marron and Wand and 
implemented in R in the package "nor1mix".
Similar to the test bed functions by Donoho and Johnstone (Blocks, Bumps
etc.) commonly used in regression or the "Peppers" and "Lena" images
popular in image analysis, the densities in this collection should be 
useful for testing and comparing new and existing density estimators.
"benchden" 1.0.0 contains functions for the generation of random
variates as well as density-, distribution- and quantile-functions.
Everything is implemented in typical R-style and the package should
reduce the programming effort needed for simulation studies in
nonparametric density estimation. It also allows for better
reproducibility of the results.

Thoralf Mildenberger, Henrike Weinert and Sebastian Tiemeyer




More information about the R-packages mailing list