EGRI: Efficient Global Robust Inversion

In IRSN/RobustInv: "Robust inversion of expensive black-box functions"

Description

Sequential design of experiments for robust inversion.

Usage

EGRI(T, model, method = NULL, fun, iter = 1, batchsize = 1,
  opt.index, inv.index, lower = NULL, upper = NULL,
  new.noise.var = NULL, integcontrol = NULL, optimcontrol = NULL,
  kmcontrol = NULL, ...)

Arguments

T	Target threshold.
model	The current kriging model. km object.
method	Criterion used for choosing observations. Currently, only the "surnew" criterion is available.
fun	Objective function.
iter	Number of iterations. At each iteration, a batch of batchsize point is evaluated in parallel.
batchsize	The size of the batch of points evaluated at each iteration.
opt.index	Array with integers corresponding to the indices of the nuisance parameters.
inv.index	Array with integers corresponding to the indices of the controlled parameters.
lower	Array of size d. Lower bound of the input domain.
upper	Array of size d. Upper bound of the input domain.
new.noise.var	Optional scalar value of the noise variance of the new observations.
integcontrol	The integcontrol argument used in the integration_design_robinv function. This applies for the following criterion: "surnew". See the help of the integration_design_robinv function.
optimcontrol	A list used to set up the optimization procedure of the chosen sampling criterion. For the "surnew" criterion, see the help of the max_surnew_robinv function.
kmcontrol	Optional list representing the control variables for the re-estimation of the kriging model once new points are sampled. The items are the same as in the km function of the DiceKriging package.
...	Other arguments of the objective function fun.

Value

A list with the following fields. (i) par: The added observations (iter*batchsize) x d matrix, (ii) value: The value of the function fun at the added observations, (iii) lastmodel: The current (last) kriging model of km class, (iv) lastvalue: The value of the criterion at the last added point, (v) allvalues: If an optimization on a discrete set of points is chosen, the value of the criterion at all these points, for the last iteration, (vi) lastintegration.param: For debug. The last value returned by the integration_design_robinv function, if applicable.

Author(s)

Clement Chevalier [e-mail]

Examples

library(KrigInv)
myfun <- function(x) return(-1 * branin_robinv(x))
d <- 4

set.seed(8)

# an example with scaling
n0 <- 40
T <- -10
opt.index <- c(3,4)
inv.index <- c(1,2)
lower <- rep(0,times=d)
upper <- rep(1,times=d)

design <- matrix(runif(d*n0),nrow=n0)
response <- myfun(design)

knots.number <- c(2,2,2,2)
knots <- generate_knots(knots.number = knots.number , d = d)

model <- km(formula = ~1,design = design,response = response,covtype = "matern3_2",scaling = TRUE,knots = knots)

integcontrol <- list(distrib = "surnew",n.points = 100,finaldistrib="surnew",n.candidates=300,
                     nsimu=100,n.optpoints = 50,choose_optpoints=TRUE,n.optpoints.candidates=200)

optimcontrol <- list(method = "genoud", pop.size = 400, max.generations = 4, wait.generation = 1)
## Not run: 
obj <- EGRI(T = T,method="surnew",model = model,fun = myfun,iter = 2,
            batchsize = 2,opt.index = opt.index,inv.index = inv.index,
            integcontrol=integcontrol,optimcontrol=optimcontrol)

## End(Not run)

IRSN/RobustInv documentation built on Nov. 20, 2019, 10:46 p.m.