gatherReplicas collects all sample points, from all files, which are assumed to be exact replicas, with different seeds (and possibly sizes). This function uses mclapply to process the files, which may be quicker than gatherSample. The temperature is disregarded, assuming that no parallel tempering was used. To facilitate the loading of a very big sample, this function will analyse the auto-correltation within each file and returned a thinned subsample of size N/2*tauint (effective sample size). There is no need to further reduce the rfesult.

For small samples, it is better to load the entire sample and analyse it in full. This function is intended for samples that are so big that they challenge the memory of the machine.

Usage

gatherReplicas(files)

Arguments

files

a list of file names

Value

a matrix of sampled points, all with the same temperature

Details

This function is quicker if you have used trivial parallelism, without mpi communication between the ranks (or another method of obtaining several replicas, like forking or sequetial reprtition).

This function assumes that each supplied RDS file contains a matrix of model MCMC parameters. The returned value X will be the rbind of all the smaller x contained in the individual files. The value X will have several attributes attached to it:

• logLikelihood: log(likelihood(X[i,])), one value per row of X

• stepSize: the MCMC step size used in each given file#'