threshold {fanovaGraph}

Threshold indices

Description

All indices below a treshold are set to be zero.

Usage

threshold(graphlist, delta, scaled = TRUE, robust = FALSE)

Arguments

graphlist

an object of class graphlist as obtained from estimateGraph

delta

numeric threshold, between 0 and 1 if scaled = TRUE

scaled

optional boolean, if TRUE, indices are normalized by the overall variance before for threshold cut, defaults to TRUE

robust

optional boolean, if TRUE, upper confidence intervals limits are used for the threshold cut instead of indices themselves, confidence intervals must be provided in graphlist, defaults to FALSE

Value

an object of class graphlist where the indices are thresholded the clique structure is updated respectively, see estimateGraph for a detailed description

Warning

The threshold cut is by default performed on scaled indices. For a cut on the original unscaled indices set scaled = FALSE.

Author(s)

J. Fruth, T. Muehlenstaedt, O. Roustant

References

Muehlenstaedt, T.; Roustant, O.; Carraro, L.; Kuhnt, S. (2011) Data-driven Kriging models based on FANOVA-decomposition, Statistics and Computing.

Examples

# Kriging model prediction
x <- matrix(runif(100*3,-pi,pi),100,3)
KM <- km(~1, design = data.frame(x), response = ishigami.fun(x))
krigingMean <- function(Xnew) predict(object = KM, newdata = Xnew, 
    type = "UK", se.compute = FALSE, checkNames = FALSE)$mean
    
# full graph estimation
g <- estimateGraph(krigingMean, d=3, n.tot=10000, q.arg=list(min=-pi, max=pi))
print(g[c(2,6)])
# threshold graph
g.cut <- threshold(g, delta = 0.1)
print(g.cut[c(2,6)])