kergp (version 0.5.7)

predict.gp: Prediction Method for the "gp" S3 Class

Description

Prediction method for the "gp" S3 class.

Usage

# S3 method for gp
predict(object, newdata,
        type = ifelse(object$trendKnown, "SK", "UK"), 
        seCompute = TRUE, covCompute = FALSE,
        lightReturn = FALSE, biasCorrect = FALSE,
        forceInterp,        
        ...)

Value

A list with the following elements.

mean

GP mean ("kriging") predictor (including the trend) computed at newdata.

sd

GP prediction ("kriging") standard deviation computed at newdata. Not computed if seCompute is FALSE.

sdSK

Part of the above standard deviation corresponding to simple kriging (coincides with sd when type = "SK"). Not computed if seCompute is FALSE.

trend

The computed trend function, evaluated at newdata.

cov

GP prediction ("kriging") conditional covariance matrix. Not computed if covCompute is FALSE (default).

lower95,

upper95

Bounds of the 95 % GP prediction interval computed at newdata (to be interpreted with special care when parameters are estimated, see description above). Not computed if seCompute is FALSE.

c

An auxiliary matrix \(\mathbf{c}\), containing all the covariances between the points in newdata and those in the initial design. Not returned if lightReturn is TRUE.

cStar

An auxiliary vector, equal to \(\mathbf{L}^{-1}\mathbf{c}\) where \(\mathbf{L}\) is the Cholesky root of the covariance matrix \(\mathbf{C}\) used in the estimation. Not returned if lightReturn is TRUE.

Arguments

object

An object with S3 class "gp".

newdata

A data frame containing all the variables required for prediction: inputs and trend variables, if applicable.

type

A character string corresponding to the GP "kriging" family, to be chosen between simple kriging ("SK"), or universal kriging ("UK").

seCompute

Optional logical. If FALSE, only the kriging mean is computed. If TRUE, the kriging variance (actually, the corresponding standard deviation) and prediction intervals are computed too.

covCompute

Logical. If TRUE the covariance matrix is computed.

lightReturn

Optional logical. If TRUE, c and cStar are not returned. This should be reserved to expert users who want to save memory and know that they will not miss these values.

biasCorrect

Optional logical to correct bias in the UK variance and covariances. Default is FALSE. See Details below.

forceInterp

Logical used to force a nugget-type prediction. If TRUE, the noise will be interpreted as a nugget effect. This argument is likely to be removed in the future.

...

Not used yet.

Author

O. Roustant, D. Ginsbourger, Y. Deville

Details

The estimated (UK) variance and covariances are NOT multiplied by \(n/(n-p)\) by default (\(n\) and \(p\) denoting the number of rows and columns of the trend matrix \(\mathbf{F}\)). Recall that this correction would contribute to limit bias: it would totally remove it if the correlation parameters were known (which is not the case here). However, this correction is often ignored in the context of computer experiments, especially in adaptive strategies. It can be activated by turning biasCorrect to TRUE, when type = "UK"

See Also

gp for the creation/estimation of a model. See gls-methods for the signification of the auxiliary variables.