lc.control: Set the Control Parameters for 'logcon'

In logconcens: Maximum Likelihood Estimation of a Log-Concave Density Based on Censored Data

Set the Control Parameters for logcon

Description

Allows to set the control parameters for the more technical aspects of the function logcon and provides default values for any parameters that are not set.

Usage

lc.control(maxiter=49, move.prec=1e-5, domind1l=1, domind2r=1, force.inf=FALSE,
                red.thresh=NULL, check.red=TRUE, addpoints=FALSE, addeps=NULL,
                preweights=NULL, minw=0, show=FALSE, verbose=FALSE)

Arguments

maxiter	the maximal number of iterations in the main EM algorithm. Default is 49 rather than 50, because this goes well with plotting in case of show = TRUE.
move.prec	a real number giving the threshold for the L_1-distance between densities in subsequent steps below which the algorithm is stopped.
domind1l, domind2r	index numbers in the vector of sorted interval endpoints that specify the left and right boundary of the maximal domain to be considered by the algorithm; see the details section of the help for logcon. The indices are counted from the left and from the right, respectively. So the default values of domind1l = 1 and domind2r = 1 mean that the largest possible domain is used.
force.inf	logical. For experimental use only. Should the domain interval be forced to be right-infinite (if there is a right-infinite data interval)?
red.thresh	a real number indicating the threshold below which the boundary integrals are considered too small; see the details section of the help for logcon. There is a sensible default, which depends on check.red.
check.red	logical. If a boundary integral is deemed too small, should the derivative of the augmented log-likelihood be checked to confirm the domain reduction.
addpoints	logical. Should extra exact observations be added to the data at the left- and rightmost finite interval endpoints to prevent domain reduction? These observations obtain a small weight <1 as compared to the weight of 1 for all the other observation intervals. The weight is specified by addeps.
addeps	a positive real number. If NULL, a default value of 1/n^2 is computed where n is the number of observation intervals. See addpoints.
preweights	a vector of weights for the observation intervals. Defaults to rep(1,n).
minw	a positive real number. This gives another way for preventing domain reduction. Instead of adding observations the weights for the internal active set algorithm are kept at or above minw at the boundary of the domain.
show	logical. Should progress of the algorithm be plotted? Warning: if TRUE, this may open many new graphics devices in case of complicated data sets.
verbose	logical. Should additional information about the progress of the algorithm be printed? This mainly prints quantities important for the decision to reduce the domain of the function and about the progress of the EM algorithm.

Details

For further explanations about the algorithm see the help for logcon. In summary:

maxiter and move.prec provide stopping criteria for the EM algorithm.

domind1l, domind2r, force.inf, red.thresh, and check.red control aspects related to domain reduction.

addpoints, addeps, preweights, winw allow for reweighing of data interval, mainly for increasing numerical stability by preventing domain reduction.

show and verbose give illustrations and background information of the run of the algorithm.

Value

A list with all of the above components set to their (specified or default) value.

Author(s)

Dominic Schuhmacher [e-mail]
Kaspar Rufibach [e-mail]
Lutz Duembgen [e-mail]

See Also

logcon

Examples

## See the examples for logcon

logconcens documentation built on Aug. 19, 2023, 5:12 p.m.