qle (version 0.18)

updateCovModels: Update covariance models

Description

The function updates the current covariance models stored in `qsd`.

Usage

updateCovModels(qsd, nextData, fit = TRUE, cl = NULL,
  controls = list(), verbose = FALSE)

Arguments

qsd

object of class QLmodel which is to be updated

nextData

object of class QLdata which includes new simulation results

fit

logical, if TRUE (default), re-estimate covariance parameters

cl

cluster object, NULL (default), of class "MPIcluster", "SOCKcluster", "cluster"

controls

list of control parameters passed to nloptr

verbose

logical, FALSE (default), whether to show intermediate output

Value

Object of class QLmodel as a list of updated covariance models

Details

The function updates both, the covariance models for kriging the statistics, and, if applicable, the ones for kriging the variance matrix of statistics based on the new data given by `nextData`. In practice, the user hardly needs to call this function except for empirical studies of how additional sample points might influence the overall predictive quality of the quasi-score and/or criterion function approximations.

If `fit=TRUE`, then the function re-estimates the covariance parameters for each statistic separately each time a total of `qsd$nfit` new sample points have been added. Thus, we can choose whether to fit the updated covariance models (by the REML estimation method) each time, e.g. during the estimation by qle if `qsd$nfit`=1, or after each 2nd, 3rd, and so on newly added point in order to limit the computational overhead. If bootstrapping was used to estimate the nugget variance of kriging models of the variance matrix, then these are taken from `nextData`.

See Also

setQLdata, simQLdata, QLmodel

Examples

# NOT RUN {
data(normal)

# old design
X <- as.matrix(qsd$qldata[c(1,2)])

# augment design with two additional points
Xnew <- multiDimLHS(N=2,qsd$lower,qsd$upper,X=X,
           method="augmentLHS",type="matrix")

# new simulations
Xsim <- simQLdata(sim=qsd$simfn,nsim=10,X=Xnew)

# prepare data
Xdata <- setQLdata(Xsim,Xnew)

# do not re-estimate covariance parameters
qsd2 <- updateCovModels(qsd,Xdata,fit=FALSE) 
 
# }