EBSICL:
Model Selection by Integrated Completed Likelihood criterion
Description
Computes the exact ICL criterion: -Loglikelihood (data,K) + H(m|K) where H is the entropy of the segmentation, and chooses the optimal number of segments as k= argmin(ICL)
Usage
EBSICL(x, prior=numeric())
Arguments
x
An object of class EBS returned by function EBSegmentation applied to data of interest.
prior
A vector of length Kmax giving prior probabilities on the value of K. Default value is uniform on 1:Kmax.
Value
NbICL
An integer containing the choice of the optimal number of segments.
ICL
Vector of length x$Kmax containing the ICL values.
Details
This function is used to compute the entropy of the segmentation in k segments (for k in 1 to Kmax) and choose the optimal K according to the ICL criteria.
References
Rigaill, Lebarbier & Robin (2012): Exact posterior distributions over the segmentation space and model selection for multiple change-point detection problems Statistics and Computing