MSL: Maximum likelihood from an inferred likelihood surface

MSLR Documentation

Maximum likelihood from an inferred likelihood surface

Description

This computes the maximum of an object of class SLik representing an inferred (summary) likelihood surface

Usage

MSL(object, CIs = TRUE, level = 0.95, verbose = interactive(),
    eval_RMSEs = TRUE, cluster_args=list(),init=NULL, prior_logL=NULL,
    ...)

Arguments

object

an object of class SLik_j as produced by infer_SLik_joint (or, in the primitive workflow, of class SLik as produced by infer_surface.logLs).

CIs

If TRUE, construct one-dimensional confidence intervals for all parameters.

level

Intended coverage probability of the confidence intervals.

verbose

Whether to display some information about progress and results.

eval_RMSEs

Logical: whether to evaluate prediction uncertainty for likelihoods/ likelihood ratios/ parameters.

cluster_args

A list of arguments, passed to makeCluster, to control parallel computation of RMSEs. Beware that parallel computation of RMSEs tends to be memory-intensive. The list may contain a non-null spec element, in which case the nb_cores global Infusion option is ignored. Do *not* use a structured list with an RMSE element as is possible for refine (see Details of refine documentation).

init

Initial value for the optimiser. Better ignored.

prior_logL

(effective only for up-to-date workflow using gaussian mixture modelling of a joint distribution of parameters and statistics) a function that returns a vector of prior log-likelihood values, which is then added to the likelihood deduced from the summary likelihood analysis. The function's single argument must handle a matrix similar to the newdata argument of predict.SLik_j.

...

Further arguments passed from or to other methods.

Details

If Kriging has been used to construct the likelihood surface, RMSEs are computed using approximate formulas for prediction (co-)variances in linear mixed midels (see Details in predict). Otherwise, a more computer-intensive bootstrap method is used. par_RMSEs are computed from RMSEs and from the numerical gradient of profile log-likelihood at each CI bound. Only RMSEs, not par_RMSEs, are compared to precision.

Value

The object is returned invisibly, with the following added members, each of which being (as from version 1.5.0) an environment:

MSL

containing variables MSLE and maxlogL that match the par and value returned by an optim call. Also contain the hessian of summary likelihood at its maximum.

RMSEs

containing, as variable RMSEs, the root mean square errors of the log-likelihood at its inferred maximum and of the log-likelihood ratios at the CI bounds.

par_RMSEs

containing, as variable par_RMSEs, root mean square errors of the CI bounds.

To ensure backward-compatibility of code to possible future changes in the structure of the objects, the extractor function get_from should be used to extract the RMSEs and par_RMSEs variables from their respective environments, and more generally to extract any element from the objects.

Examples

## see main documentation page for the package

Infusion documentation built on Sept. 29, 2022, 1:05 a.m.

Related to MSL in Infusion...