LinkedGASP (version 1.0)

TGASPmetrics: Performance measurement of a T-GASP

Description

Evaluates frequentist performance of a T-GASP.

Usage

TGASPmetrics(TGASP, true_output, ref_output)

Arguments

TGASP

TGASP emulator (in the paper this is done within an objective Bayesian implementation - OB emulator.)

true_output

Output from the simulator.

ref_output

Heuristic emulator output.

Value

List of performance measures.

RMSPE_base

Root mean square predictive error with respect to the heuristic emulator output.

RMSPE

Root mean square predictive error for the emulator output

ratio

ratio of RMSPE_base to RMSPE. Ratio = RMSPE_base/RMSPE

CIs

95% central credible intervals

emp_cov

95% empirical coverage within the CIs

length_CIs

Average lenght of 95% central credible intervals

Details

See examples which illustrate the use of the function.

References

Ksenia N. Kyzyurova, James O. Berger, and Robert L. Wolpert. Coupling computer models through linking their statistical emulators. SIAM/ASA Journal on Uncertainty Quantification, 6(3): 1151-1171, 2018

Examples

# NOT RUN {
## Function f1 is a simulator
f1<-function(x){sin(pi*x)}

## One-dimensional inputs are x1
x1 <- seq(-1,1,.37)

## The following contains the list of data inputs (training) and outputs (fD) together with 
## the assumed fixed smoothness of a computer model output.
data.f1 <- list(training = x1,fD = f1(x1), smooth = 1.99)

## Evaluation of GASP parameters
f1_MLEs = eval_GASP_RFP(data.f1,list(function(x){x^0},function(x){x^1}),1,FALSE)

## Evaluate the emulator
xn = seq(-1,1,.01)
TGASP_f1 <- eval_TGASP(as.matrix(xn),f1_MLEs)

## Plot the emulator
par(mfrow = c(1,1))
par(mar = c(6.1, 6.1, 5.1, 2.1))
ylim = c(-1.5,1.5)
TGASP_plot(TGASP_f1,f1,data.f1,ylim = ylim)

## Measure the performance of the emulator
TGASPmetrics(TGASP_f1,f1(xn),mean(f1(xn)))
# }