Implements the vine copula based kernel density estimator of
Nagler and Czado (2016)
This package implements a vine copula based kernel density estimator. The estimator does not suffer from the curse of dimensionality and is therefore well suited for high-dimensional applications (see, Nagler and Czado, 2016). The package is built on top of the copula density estimators in kdecopula and let's you choose from all its implemented methods. The package can handle discrete and categorical data via continuous convolution.
You can install:
the stable release on CRAN:
the latest development version:
A detailed description of of all functions and options can be found in the API documentaion. In short, the package provides the following functionality:
kdevine and its methods:
kdevine(): Multivariate kernel density estimation based on vine copulas. Implements the estimator of (see, Nagler and Czado, 2016).
rkdevine(): Density and simulation functions.
kdevinecop and its methods:
kdevinecop(): Kernel estimator for the vine copula density (see, Nagler and Czado, 2016).
rkdevinecop(): Density and simulation functions.
contour.kdevinecop(): Matrix of contour plots of all pair-copulas.
kde1d and its methods:
kde1d(): Univariate kernel density estimation for bounded and unbounded support.
rkde1d(): Density, cdf, and simulation functions.
lines.kde1d(): Plots the estimated density.
Nagler, T., Czado, C. (2016) Evading the curse of dimensionality in nonparametric density estimation with simplified vine copulas Journal of Multivariate Analysis 151, 69-89 (doi:10.1016/j.jmva.2016.07.003, preprint)
Nagler, T., Schellhase, C. and Czado, C. (2017) Nonparametric estimation of simplified vine copula models: comparison of methods arXiv:1701.00845
Nagler, T. (2017) A generic approach to nonparametric function estimation with mixed data arXiv:1704.07457
kspackage (and thereby its dependencies like rgl) by calling
KernSmooth::dpik()directly instead of
plot.kde1dtype for continuous data (was shown as histogram).
support for discrete variables based on continuous convolution (using R package cctools).
new bandwidth default
mult_1d = log(1 + d) in
First release on CRAN.