delt (version 0.8.2)

densplit: Calculation of an overfitting histogram

Description

The function returns an overfitting histogram when a data matrix is given as an input. The output is an evaluation tree which is grown with greedy growing. The evaluation tree defines a partition of the sample space. The evaluation tree may be pruned to get a density estimate.

Usage

densplit(dendat, minobs=NULL, leaf=0, method="loglik", 
splitscan=0, seedf=1, suppo=NULL)

Arguments

dendat

n*d data matrix

minobs

non-negative integer; splitting of a bin will be continued if the bin containes "minobs" or more observations

leaf

internal (maximal number of leafs in the evaluation tree)

method

"loglik" or "projec"; the contrast function

splitscan

internal (random selection of splits)

seedf

internal

suppo

2*d vector of real numbers; the rectangle to be splitted; the rectangle has to contain the data

Value

Returns an evaluation tree as a list of vectors.

direc

integer in 1,...,d; variable which is splitted

split

real number; splitting point

mean

nonnegative number; value of the histogram on the rectangle corresponding to the node

nelem

nonnegative integer; number of observations in the rectangle corresponding to the node

ssr

real number; value of the likelihood criterion

volume

non-negative number; volume of the rectangle corresponding to the node

left

non-negative integer; link to the left child, 0 if terminal node

right

non-negative integer; link to the right child, 0 if terminal node

low

the lower vertice of the rectangles

upp

the upper vertice of the rectangles

N

the number of grid points at each direction

support

the support of the histogram

See Also

prune, eval.pick

Examples

library(denpro)

dendat<-sim.data(n=200,seed=5,type="mulmodII")
et<-densplit(dendat)

treeseq<-prune(et)
treeseq$leafs
len<-length(treeseq$leafs)

leaf<-treeseq$leafs[len-10]
leaf
etsub<-eval.pick(treeseq,leaf=leaf)

dp<-draw.pcf(etsub)
persp(dp$x,dp$y,dp$z,phi=25,theta=-120)