fitlmm: Learn RLL or Log-posterior for two variance MM.
In andrewpbray/lmmoptim: Optimization routine for linear mixed models

Description Usage Arguments Details Value

Learns the shape of the objective function within epsilon.

1 2	fitlmm(lines, startbox, eps = 0, delE = 0, delS = 0, M = Inf, maxit = 10, ratio = FALSE, lognote = "summary")

`lines`	a dataframe that contains the constants that define the line represented by each term in the sum. Created as output from `findlines`.
`startbox`	a list of boxes and their bounds, possibly the output from a previous call to `fitlmm`. If left empty, will create a new startbox from a call to `makebox`.
`eps`	a non-negative numeric indicating the tolerance within which you would like to learn the function. Default is 0.
`delE`	a non-negative numeric indicating the width of σ^2_e beyond which the algorithm will stop branching boxes. Default is 0.
`delS`	a non-negative numeric indicating the width of σ^2_s beyond which the algorithm will stop branching boxes. Default is 0.
`M`	a non-negative numeric indicating the size of the buffer between the highest global lower bound and the upper bound of a given box past that box will now be branched further. Default is `Inf`.
`maxit`	a positive integer indicating the maximum number of iterations of the algorithm. Default is 10.
`ratio`	a logical indicating if `delE` and `delS` are specified as ZZQ. Default is `FALSE`.
`lognote`	a string to append to the "log.out" log file to annotate the purpose of each run of the algorithm. For use in benchmarking computation time. Default is `"summary"`.

This is the primary function that implements the branch-bound-kill algorithm described in "Approximately Exact Calculations for Linear Mixed Models" by Lavine & Hodges (2015). This an interative algorithm that can require substantial computation time. It is recommended that the user start with a conservative maxit and do additional computations as necessary.

The arguments of this function include five settings to control when to stop branching (subdividing) a box:

when the upper and lower bounds are within eps of one another.
when the width of σ^2_e is less than delE.
when the width of σ^2_s is less than delS.
when the upper bound is more than M below the highest global lower bound.
when the number of iterations reaches maxit.

A list of boxes including their limits in σ^2_e and σ^2_e as well as their bounds. Running this function also results in the creation of a log file called "log.out" containing box counts at every iteration.

andrewpbray/lmmoptim documentation built on May 10, 2019, 11:10 a.m.