profkern: Calculates a level set tree of a kernel estimate
In denpro: Visualization of Multivariate Functions, Sets, and Data

Description Usage Arguments Value Note Author(s) See Also Examples

View source: R/profkern.R View source: R/denpro.R

Given a data matrix, returns a level set tree of a kernel estimate, when the dimension of the estimate is less or equal to 4.

1
2
3

profkern(dendat, h, N, Q, cvol=TRUE, ccen=TRUE, cfre=FALSE,
kernel="epane", compoinfo=FALSE, trunc=3, threshold=0.0000001,
sorsa="crc",hw=NULL)

`dendat`	n*d data matrix; d<=4
`h`	positive real number or vector of positive real numbers; smoothing parameters from the largest to the smallest
`N`	d-vector of dyadic integers >=4; number of knots for every direction: kernel estimate will be evaluated on these knots
`Q`	positive integer; number of quantization levels
`cvol`	TRUE or FALSE; TRUE if one wants that the volumes of the separated parts of the level sets are returned, note that one needs the volumes for drawing a volume plot
`ccen`	T or F; TRUE if one wants that the barycenters of the separated parts of the level sets are returned, note that one needs the barycenters for drawing a barycenter plot
`cfre`	T or F; TRUE if one wants that the frequencies for separated parts of the level sets are returned (not implemented)
`kernel`	either "epane" or "gauss"; the kernel of the kernel estimate
`compoinfo`	TRUE or FALSE; whether definitions of separated regions of level sets will be returned
`trunc`	positive real number; truncation of the Gaussian kernel to the interval [-trunc,trunc]
`threshold`	positive real number; kernel will be truncated to 0 if the value is smaller than the threshold (not implemented)
`sorsa`	if not "crc" uses slower R code
`hw`	lower level parameter

Returns a list of level set trees: for each value of h the level set tree of the corresponding kernel estimate. If h is not a vector but scalar, then only one level set tree will be returned. The level set tree is a list of vectors. The elements of the vectors supply information for each node of the tree. Below we denote with "nodenum" the number of nodes of the tree.

`parent`	"nodenum"-vector of integers in the range 0,..., nodenum-1; links to the parent of each node. Root nodes are marked with 0.
`level`	"nodenum"-vector of positive real numbers; level of the level set from which the set corresponding to the node is a part of.
`volume`	"nodenum"-vector of positive real numbers; volumes of sets corresponding to each node
`center`	d*nodenum-matrix; barycenters of sets corresponding to each node
`invalue`	"nodenum"-vector of positive integers; level of the level set in terms of original frequencies (these values are not normalized so that the estimate would integrate to one
`component`	Will be returned if "compoinfo"=TRUE; for each node of the level set a tree pointer to "AtomlistAtom"
`AtomlistAtom`	Will be returned if "compoinfo"=TRUE; pointers to "index"
`AtomlistNext`	Will be returned if "compoinfo"=TRUE; pointers which define the list of atoms for each separated component (for each node of the level set tree)
`index`	Will be returned if "compoinfo"=TRUE; matrix with d columns: determines a knot of the multivariate grid where estimate was evaluated
`nodefrek`	"nodenum"-vector of positive integers; number of observations in the set corresponding to the node. This is useful in cluster analysis applications. (Not implemented.)

Applies the DynaDecompose algorithm, described in the article "Algorithms for manipulation of level sets of nonparametric density estimates", by Jussi Klemela

Jussi Klemela

plotvolu, plotbary, plottree,

set.seed(1)
dendat<-matrix(rnorm(20),10)      # 10*2 data-matrix
pk<-profkern(dendat,h=1,N=c(8,8),Q=4)
plotvolu(pk)

dendat<-sim.data(n=200,type="mulmod")
pk<-profkern(dendat,h=1,N=c(64,64),Q=30)
plotvolu(pk)