mcmcpermind-class: Finmix 'mcmcpermind' class

In simonsays1980/finmix: An R package for Bayesian estimation of finite mixture distributions

Description

This class defines objects to store the outputs from permuting the MCMC samples. Due to label switching the sampled component parameters are usually not assigned to the same component in each iteration. To overcome this issue the samples are permuted by using a relabeling algorithm (usually K-means) to reassign parameters. Note that due to assignment of parameters from the same iteration to the same component, the sample size could shrink.

This class stores the permuted parameters together with the new MCMC sample size and the mixture log-likelihood, the prior log-likelihood, and the complete data posterior log-likelihood. All this slots are inherited from the parent class mcmcpermfix. In addition to these slots this class adds permuted weights, permuted indicators as well as the entropies and number of assigned observations per component.

Slots

relabel

A character defining the used algorithm for relabeling.

weightperm

An array of dimension Mperm x K containing the relabeled weight parameters.

entropyperm

An array of dimension Mperm x 1 containing the entropy for each MCMC permuted draw.

STperm

An array of dimension Mperm x 1 containing all permuted MCMC states, for the last observation in slot @y of the fdata object passed in to mixturemcmc() where a state is defined for non-Markov models as the last indicator of this observation.

Sperm

An array of dimension N x storeS containing the last storeS permuted indicators. storeS is defined in the slot @storeS of the mcmc object passed into mixturemcmc().

NKperm

An array of dimension Mperm x K containing the numbers of observations assigned to each component.

See Also

• mcmcpermute() for the calling function

• mcmcpermfix for the corresponding class for models with fixed indicators

simonsays1980/finmix documentation built on Dec. 23, 2021, 2:25 a.m.