twDEMC-package: parallel DEMC

In twDEMC: parallel DEMC

Description

Parallel version of a differential evolution markov chain (DEMC) combined with simulated annealing (SA). DEMC is a variant of Monte Carlo Markov Chain (MCMC).

Details

Method twDEMCSA automatically infers proper temperature decrease in simulated annealing (SA), based on mixing of different independent populations of DEMC chains.

Method twDEMCBlock.twDEMCPops can be used to continue sampling with explicitly specifying starting and end temperature.

Further functionality of the twDEMC package deals with

• Generating an initial population for twDEMC: initZtwDEMCNormal

• Transforming the objects of type twDEMCPops: subset.twDEMCPops

• Transforming the objects of type twDEMC obtained by concatPops.twDEMCPops: subChains.twDEMC

• Transforming the parameter space: transOrigPopt.default

• Invoking fLogDen with proposal in a parallel load balanced way: twCalcLogDenPar and twCalcLogDensPar

• Calculating marginal aggregates: marginals1d

• Laplace approximation of the normalizing constant: normConstLaplace and normConstLaplaceBridge

• Plotting: plotThinned.mcmc.list

• Scaling logDensities by temperature: calcBaseTemp, calcTemperatedLogDen.default

• Ordering multiple log-Densities orderLogDen

The workhorse for parallelized sampling of different populations and chains is function twDEMCBlockInt, which is usually called by twDEMCBlock.array or twDEMCBlock.twDEMCPops

Author(s)

Thomas Wutzler

twDEMC documentation built on May 2, 2019, 5:38 p.m.