Sample the parameter space

Description

These functions take an SLik object (as produced by MSL) and samples its parameter space in (hopefully) clever ways, not yet well documented. rparam calls sample_volume to define points targeting the likelihood maximum and the bounds of confidence intervals, with n for these different targets dependent on the mean square error of prediction of likelihood at the maximum and at CI bounds.

Usage

1
2
3
4
5
6
7
rparam(object, n= 1, useEI = list(max=TRUE,profileCI=TRUE,rawCI=FALSE), 
       useCI = TRUE, verbose = interactive(), tryn=30*n,  
       level = 0.95, CIweight=Infusion.getOption("CIweight"))

sample_volume(object, n = 6, useEI, vertices=NULL,
              dlr = NULL, verbose = interactive(), 
              fixed = NULL, tryn= 30*n)

Arguments

object

an SLik object

n

The number of parameter points to be produced

useEI

List of booleans, each determining whether to use an “expected improvement” (EI) criterion (e.g. Bingham et al., 2014) to select candidate parameter points to better ascertain a particular focal point. The elements max, profileCI and rawCI determine this for three types of focal points, respectively the MSL estimate, profile CI bounds, and full-dimensional bounds. When EI is used, n points with best EI are selected among tryn points randomlygenerated in some neighborhood of the focal point.

vertices

Points are sampled within a convex hull defined by vertices. By default, these vertices are taken from object$fit$data.

useCI

Whether to define points targeting the bounds of confidence intervals for the parameters. An expected improvement criterion is also used here.

level

If useCI is TRUE but confidence intervals are not available from the object, such intervals are computed with coverage level.

dlr

A (log)likelihood ratio threshold used to select points in the upper region of the likelihood surface. Default value is given by Infusion.getOption("LRthreshold")

verbose

Whether to display some information about selection of points, or not

fixed

A list or named vector, of which each element is of the form <parameter name>=<value>, defining a one-dimensional constraint in parameter space. Points will be sampled in the intersection of the volume defined by the object and of such constraint(s).

tryn

See useEI argument.

CIweight

For development purposes, not documented.

Value

a data frame of parameter points. Only parameters variable in the SLik object are considered.

References

D. Bingham, P. Ranjan, and W.J. Welch (2014) Design of Computer Experiments for Optimization, Estimation of Function Contours, and Related Objectives, pp. 109-124 in Statistics in Action: A Canadian Outlook (J.F. Lawless, ed.). Chapman and Hall/CRC.

Examples

1
2
3
4
5
if (Infusion.getOption("example_maxtime")>10) {
 data(densv)
 summliksurf <- infer_surface(densv) ## infer a log-likelihood surface
 sample_volume(summliksurf)
}