neighbors tests whether neighbors of point k,P can be used to re-initialize the EM algorithm and to improve the log-likelihood.

neighborsbis(
  kv.hull,
  x,
  L,
  k,
  param,
  P,
  lmin,
  eps,
  sameSigma = TRUE,
  pureR = FALSE
)

Arguments

kv.hull

convex hull of likelihood

x

the initial dataset

L

the likelihood

k

the points of interest

param

param outputs of segmentation

P

the number of class

lmin

minimal size of the segment to be implemented

eps

eps

sameSigma

should segments have same variance ?

pureR

should algorithm use only R functions or benefit from Rcpp faster algorithm

Value

smoothing likelihood