update_km_noisyEGO {DiceOptim}

Update of one or two Kriging models when adding new observation

Description

Update of a noisy Kriging model when adding new observation, with or without covariance parameter re-estimation. When the noise level is unkown, a twin model "estim.model" is also updated.

Usage

update_km_noisyEGO(
  model,
  x.new,
  y.new,
  noise.var = 0,
  type = "UK",
  add.obs = TRUE,
  index.in.DOE = NULL,
  CovReEstimate = TRUE,
  NoiseReEstimate = FALSE,
  estim.model = NULL,
  nugget.LB = 1e-05
)

Arguments

model

a Kriging model of "km" class

x.new

a matrix containing the new points of experiments

y.new

a matrix containing the function values on the points NewX

noise.var

scalar: noise variance

type

kriging type: "SK" or "UK"

add.obs

boolean: if TRUE, the new point does not exist already in the design of experiment model@X

index.in.DOE

optional integer: if add.obs=TRUE, it specifies the index of the observation in model@X corresponding to x.new

CovReEstimate

optional boolean specfiying if the covariance parameters should be re-estimated (default value = TRUE)

NoiseReEstimate

optional boolean specfiying if the noise variance should be re-estimated (default value = TRUE)

estim.model

optional input of "km" class. Required if NoiseReEstimate=TRUE, in order to deal with repetitions.

nugget.LB

optional scalar: is used to define a lower bound on the noise variance.

Value

A list containing:

model

The updated Kriging model

estim.model

If NoiseReEstimate=TRUE, the updated estim.model

noise.var

If NoiseReEstimate=TRUE, the re-estimated noise variance

Author(s)

Victor Picheny

References

V. Picheny and D. Ginsbourger (2013), Noisy kriging-based optimization methods: A unified implementation within the DiceOptim package, Computational Statistics & Data Analysis