objectiveFunctionEvaluation: objectiveFunctionEvaluation Objective Function Evaluation

In SPOT: Sequential Parameter Optimization Toolbox

objectiveFunctionEvaluation Objective Function Evaluation

Description

This function handles the evaluation of the objective function in spot. This includes handling of the random number generator stream, variable transformations (transformX) as well as the actual evaluation.

Usage

objectiveFunctionEvaluation(x = NULL, xnew, fun, control = list(), ...)

Arguments

x	matrix of already known solutions, to determine whether RNG seeds for new solutions need to be incremented.
xnew	matrix of new solutions.
fun	objective function to evaluate the solutions in xnew.
control	control list with the following entries: seedFun initial seed to be used for the random number generator seed. Set to NA to avoid using a fixed seed. noise: logical parameter specifying whether the target function is noisy. verbosity: verbosity. Default: 0. transformFun: transformation functions applied to xnew. See transformX
...	parameters passed to fun.

Value

the matrix ynew, which are the observations for fun(xnew)

See Also

spot for more details on the parameters, e.g., fun

transformX

spotControl

Examples

## 1) without noise
x <- NULL
xnew <- matrix(1:10, ncol=2)
fun <- funSphere
control <- spotControl(dim(xnew)[2])
control$verbosity <- 0
objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
##
fun <- funMoo
objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
## 2) with noise
fun = function(x){funSphere(x) + rnorm(nrow(x))}
control$noise <- TRUE
objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
## 3) known solutions
x <- matrix(11:20, ncol=2)
xnew <- matrix(1:10, ncol=2)
fun <- funSphere
objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
## 4) known solutions with noise and repeats
x <- matrix(1:20, ncol=2, byrow=TRUE)
xnew <- matrix(1:10, ncol=2, byrow=TRUE)
fun = function(x){funSphere(x) + rnorm(nrow(x))}
objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
## 5) identical solutions with noise and repeats
x <- matrix(1:10, ncol=2, byrow=TRUE)
xnew <- x
fun = function(x){funSphere(x) + rnorm(nrow(x))}
y <- objectiveFunctionEvaluation(x=NULL, xnew=x, fun=fun, control=control)
y1 <- objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
y2 <- objectiveFunctionEvaluation(x=NULL, xnew=xnew, fun=fun, control=control)
print(cbind(x, y))
print(cbind(xnew, y1))
print(cbind(xnew, y2))
identical(y, y1) # FALSE
identical(y, y2) # TRUE
## 6) known solutions with noise and repeats. function sets seed
x <- matrix(1:20, ncol=2, byrow=TRUE)
xnew <- matrix(1:10, ncol=2, byrow=TRUE)
fun <- function(x,seed){
  set.seed(seed)
  funSphere(x)+rnorm(nrow(x))}
control$seedFun <- 1
y1 <- objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
y2 <- objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
identical(y1, y2) # TRUE
control$seedFun <- 2
y3 <- objectiveFunctionEvaluation(x=x, xnew=xnew, fun=fun, control=control)
identical(y1,y3) # FALSE
## 7) spot examples:
res1a <- spot(,function(x,seed){set.seed(seed);funSphere(x)+rnorm(nrow(x))},
c(-2,-3),c(1,2),control=list(funEvals=25,noise=TRUE,seedFun=1))
res1b <- spot(,function(x,seed){set.seed(seed);funSphere(x)+rnorm(nrow(x))},
c(-2,-3),c(1,2),control=list(funEvals=25,noise=TRUE,seedFun=1))
res2 <- spot(,function(x,seed){set.seed(seed);funSphere(x)+rnorm(nrow(x))},
c(-2,-3),c(1,2),control=list(funEvals=25,noise=TRUE,seedFun=2))
sprintf("Should be equal: %f = %f. Should be different:  %f", res1a$ybest,
res1b$ybest, res2$ybest)

SPOT documentation built on June 26, 2022, 1:06 a.m.