link: Linking two emulators

Description Usage Arguments Details Value Author(s) References Examples

Description

Function constructs a linked GASP emulator of a composite computer model f2(f1).

Usage

1
link(f1_MLEs, f2_MLEs, test_input)

Arguments

f1_MLEs

Parameters of the emulator of a simulator f1.

f2_MLEs

Parameters of the emulator of a simulator f2.

test_input

Testing inputs.

Details

See examples which illustrate inputs specification to the function.

Value

Four types of the linked GASP.

em1

Type 1 emulator, which uses MAP estimates of parameters.

em2

Type 2 emulator within partial objective Bayesian (POB) implementation.

emT

T-GASP emulator within objective Bayesian (OB) implementation.

em3

Approximated T-GASP emulator with the Gaussian distribution.

Author(s)

Ksenia N. Kyzyurova, kseniak.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

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
## Function f1 is a simulator
f1<-function(x){sin(pi*x)}
## Function f2 is a simulator
f2<-function(x){cos(5*x)}

## Function f2(f1) is a simulator of a composite model
f2f1 <- function(x){f2(f1(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)
GASP_type2_f1 <- eval_type2_GASP(as.matrix(xn),f1_MLEs)

par(mfrow = c(1,1))
par(mar = c(6.1, 6.1, 5.1, 2.1))
ylim = c(-1.5,1.5)
GASP_plot(GASP_type2_f1,f1,data.f1,"Type 2 GASP",ylab = " f",xlab = "x",
ylim = ylim, plot_training = TRUE)

s = GASP_type2_f1$mu
s.var = diag(GASP_type2_f1$var)

x2 = seq(-0.95,0.95,length = 6)#f1(x1)
data.f2 <- list(training = x2,fD = f2(x2), smooth = 2) # linking requires this emulator 
# to have smoothness parameter equal to 2

f2_MLEs = eval_GASP_RFP(data.f2,list(function(x){x^0},function(x){x^1}),1,FALSE)

GASP_type1_f2 <- eval_type1_GASP(as.matrix(seq(-3.5,3.5,.01)),f2_MLEs)
GASP_type2_f2 <- eval_type2_GASP(as.matrix(seq(-1,1,.01)),f2_MLEs)
TGASP_f2 <- eval_TGASP(as.matrix(seq(-1,1,.01)),f2_MLEs)

ylim = c(-1.5,1.5)
# labels = c(expression(phantom(x)*phantom(x)*phantom(x)*f(x[1])),
# expression(f(x[2])*phantom(x)*phantom(x)*phantom(x)),
# expression(f(x[3])),expression(f(x[4])),
# expression(f(x[5])),expression(f(x[6])))

par(mar = c(6.1, 6.1, 5.1, 2.1))
GASP_plot(GASP_type2_f2,f2,data.f2, "Type 2 GASP",labels = x2,xlab= "z",ylab = " g",
ylim = ylim,plot_training = TRUE)

le <- link(f1_MLEs, f2_MLEs, as.matrix(xn))

## Plot second type of the linked GASP
data.f2f1 <- list(training = x1,fD = f2f1(x1), smooth = 2)

par(mar = c(6.1, 6.1, 5.1, 2.1))
GASP_plot(le$em2,f2f1,data.f2f1,"Linked",labels = x1,
ylab = expression("g" ~ scriptscriptstyle(O) ~ "f"),xlab = "x",ylim = ylim)

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

Related to link in LinkedGASP...