TGASPmetrics: Performance measurement of a T-GASP

In LinkedGASP: Linked Emulator of a Coupled System of Simulators

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.

Details

See examples which illustrate the use of the function.

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

Author(s)

Ksenia N. Kyzyurova, ksenia.ucoz.net

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

## 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)))

LinkedGASP documentation built on May 2, 2019, 2:08 a.m.