denpro (version 0.9.2)

pcf.kern: Computes a multivariate kernel estimate

Description

Computes a multivariate kernel estimate and gives the output as a piecewise constant function object.

Usage

pcf.kern(dendat, h, N, kernel = "gauss", weights = NULL, support = NULL,
lowest = 0, radi = 0)

Arguments

dendat

n*d matrix of real numbers; the data matrix

h

d vector of positive real numbers; vector of smoothing parameters

N

vector of d positive dyadic integers; the dimension of the grid where the kernel estimate will be evaluated; we evaluate the estimate on a regular grid which contains the support of the kernel estimate

kernel

"gauss", "epane", "bart", or "uniform"; the kernel is either the standard Gaussian, Epanechnikov product kernel, Bartlett kernel, or uniform kernel

weights

n vector of nonnegative weights, where n is the sample size; sum of the elements of "weights" should be one; these are the weights of a time localized kernel estimator

support

2*d vector of reals gives the d intervals of a rectangular support in the form c(low1,upp1,...,lowd,uppd)

lowest

a real value; the density estimate will be truncated to take value zero, if the value of the estimate is less or equal to "lowest"

radi

a nonnegative real number; the support is estimated as the smallest rectangle containing the observations with an additional band whose width is equal to "radi"

Value

a piecewise constant function object, see the web site

References

http://www.rni.helsinki.fi/~jsk/denpro/

See Also

draw.pcf lstseq.kern

Examples

n<-100
dendat<-sim.data(n=n,type="mulmod")

h<-1
pcf<-pcf.kern(dendat,h=h,N=c(32,32))
dp<-draw.pcf(pcf)
contour(dp$x,dp$y,dp$z,drawlabels=FALSE)

d<-2
h<-(4/(d+2))^(1/(d+4))*n^(-1/(d+4))*apply(dendat,2,sd)
pcf<-pcf.kern(dendat,h=h,N=c(32,32))
dp<-draw.pcf(pcf)
contour(dp$x,dp$y,dp$z,drawlabels=FALSE)

# we use now nonuniform weighting of kernels

weights<-matrix(0,n,1)
threshold<-4
for (i in 1:n){
    eta<-(n-i)
    if (eta/h>threshold) result<-0 else result<-exp(-eta^2/(2*h^2))
    weights[i]<-result
}
weights<-weights/sum(weights)

pcf<-pcf.kern(dendat,h=1,N=c(32,32),weights=weights)

dp<-draw.pcf(pcf)
contour(dp$x,dp$y,dp$z,drawlabels=FALSE)