label_switch {mixAR}

A posteriori relabelling of a Markov chain

Description

Takes the output from a MCMC simulation of parameters of a mixture, and detects whether labels switch has occured while sampling, using the method by Celeux (2000).

Usage

label_switch(x, m)

Arguments

x

output from an MCMC sampling of a mixture. A matrix, each column corresponds to one component of the mixture.

m

the number of observations in the sample that will be used to initialise the algorithm. m~100 is recommended.

Details

Function can be directly executed when x is one of mix_weights, scale, precision, shift or mu from bayes_mixAR output. ARcoeff cannot be input as it is, but element from the list may be used.

Value

A list of 2:

x

The input matrix, with adjusted labels

true_perm

The "true" permutation at each iteration.

Note

There is no absolute choice on what x should be to obtain the "true" permutation at any given point. User is subject to make the most suitable choice, given output of their MCMC.

Author(s)

Davide Ravagli

References

Celeux G (2000). Bayesian Inference of Mixture: The Label Switching Problem.. Payne R., Green P. (eds) COMPSTAT. Physica, Heidelberg.

See Also

bayes_mixAR

Examples

model <- new("MixARGaussian",
             prob   = exampleModels$WL_At@prob,      # c(0.5, 0.5)
             scale  = exampleModels$WL_At@scale,     # c(1, 2)        
             arcoef = exampleModels$WL_At@arcoef@a ) # list(-0.5, 1.1)

y <- mixAR_sim(model, n = 300, init = rep(0, which.max(model@order)))

## just examples, use larger numbers in practice
nsim <- 30   # 200
burnin <- 10  # 100
x <- bayes_mixAR(y, model, fix_shift = FALSE, tau = c(.15, .25),
              nsim = nsim, burnin = burnin)

label_switch(x$mix_weights, m = 5)