calibrate: Generates 'calibrate.class' objects

In CaliCo: Code Calibration in a Bayesian Framework

Description

calibrate is a function that allows to generate a calibrate.class class in which the estimation is done for a defined model.class and prior.class objects.

Usage

calibrate(md, pr, opt.estim, opt.valid = NULL)

Arguments

md	a model.class object
pr	a prior.class object
opt.estim	estimation options Ngibbs Number of iteration of the algorithm Metropolis within Gibbs Nmh Number of iteration of the Metropolis Hastings algorithm thetaInit Initial point r regulation percentage in the modification of the k in the Metropolis Hastings sig Covariance matrix for the proposition distribution (k*sig) Nchains Number of MCMC chains to run (if Nchain>1 an output is created called mcmc which is a coda object codamenu) burnIn Number of iteration to withdraw
opt.valid	list of cross validation options (default value opt.valid=NULL) nCV Number of iterations for the cross validation type.valid Type of cross validation selected. "loo" (leave one out) is the only method emplemented so far.

Value

calibrate returns a calibrate.class object. Two main methods are available:

• plot(mdfit, x, graph) displays a series of graphs (ACF, MCMC, density a priori vs a posteriori , correlation between parameters, results on the quantify of interest, etc..) or return a list with all the graphs:

  • mdfit The calibrated model (a calibrate.class object)

  • x The x-axis

  • graph Allows to select the wanted display. By default all the layout pannel graphs are displayed and graph="all". If graph="chains", only the layout of the autocorrelation, chains points and densities a priori and a posteriori is produced. If graph="corr", only the layout of the correlation graph between each parameter is displayed. If graph="result", only the result on the quantity of interest is given. If graph=NULL, no graphs are produced automatically.

• print(mdfit) returns the main information concerning the calibrate.class object

Author(s)

M. Carmassi

See Also

prior, calibrate, forecast, sequentialDesign

Examples

## Not run: 
###################### The code to calibrate
X <- cbind(seq(0,1,length.out=10),seq(0,1,length.out=10))
code <- function(X,theta)
{
  return((6*X[,1]*theta[2]-2)^2*theta[1]*sin(theta[3]*X[,2]-4))
}
Yexp <- code(X,c(1,1,11))+rnorm(10,0,0.1)

############### For the first model
###### Definition of the model
md <- model(code,X,Yexp,"model1")
###### Definition of the prior densities
pr <- prior(type.prior=c("gaussian","gaussian","gaussian","gamma"),opt.prior=
list(c(1,0.01),c(1,0.01),c(11,3),c(2,0.1)))
###### Definition of the calibration options
opt.estim=list(Ngibbs=200,Nmh=400,thetaInit=c(1,1,11,0.1),r=c(0.3,0.3),
sig=diag(4),Nchains=1,burnIn=100)
###### Run the calibration
mdfit <- calibrate(md,pr,opt.estim)
####### The plot generated is a list of ggplot
p <- plot(mdfit,X[,1])
p$out
print(mdfit)

## End(Not run)

CaliCo documentation built on May 2, 2019, 4:05 p.m.