initZtwDEMCNormal: initZtwDEMCNormal

In twDEMC: parallel DEMC

Description

Generate an initial population of states for twDEMCBlockInt.

Usage

initZtwDEMCNormal(thetaPrior, covarTheta, nChainPop = 4, nPop = 2, 
    m0 = ceiling(calcM0twDEMC(length(thetaPrior), nChainPop)/(m0FiniteFac)), 
    m0FiniteFac = 1, doIncludePrior = TRUE)

Arguments

thetaPrior	numeric vector (nParm) of point estimate of the mean
covarTheta	numeric matrix (nParm x nParm) the covariance of parameters. << Alternatively, can be given as vector (nParm) of the diagonal, i.e. variances, if all dimensions are independent
nChainPop	number of chains to run
nPop	number of independent populations among the chains
m0	number of initial states for each chain
m0FiniteFac	use a factor smaller than 1 to increase default m0 to account for only a portion of proposal results in finite densities
doIncludePrior	If TRUE, then set last sample of chain 1 to the prior estimate, which might be already a kind of best estimates by an optimization.

Details

There are several methods to establish the initial population for a twDEMC run.

• drawing from a multivariate normal distribution: this method

• subsetting the result of a former twDEMC run: initZtwDEMCSub.twDEMC or sample matrix initZtwDEMCSub.matrix

• extending the result of a former twDEMC run to include more parameters: initZtwDEMCExt.twDEMC

• selecting the N closes points from a sequence of points in parameter space constrainNStack

• selecting the points inside a confindenc ellipsis in parameter constrainCfStack

• replacing cases with in initial proposals that yield non-finite density replaceZinitNonFiniteLogDens

• general method for replacing cases in initial proposals replaceZinitCases

Value

a matrix of number of parameters by number of individuals (m0 x d x Npop), with d dimension of theta

Author(s)

Thomas Wutzler

See Also

twDEMCBlockInt calcM0twDEMC

Examples

data(twLinreg1)
attach( twLinreg1 )

.nChainPop=4
.nPop=2
.nPar=length(theta0)
Zinit <- initZtwDEMCNormal( theta0, diag(sdTheta^2), nChainPop=.nChainPop, nPop=.nPop)
head(Zinit[,,1])
all.equal( c(calcM0twDEMC(.nPar,.nChainPop), .nPar, .nChainPop*.nPop), dim(Zinit) )

detach()

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