An active set approach to minimizing objective in Unlinked Monotone Regression
UMRactiveSet_trust(
yy,
ww_y = NULL,
grad,
hess,
UMR_curv,
CDF,
init,
counts = rep(1, length(init)),
stepsize,
MM,
tol_end = 1e-04,
tol_collapse,
printevery,
filename
)Y (response) observation vector (numeric)
Weights (nonnegative, sum to 1) corresponding to yy. Samelength as yy. Or NULL in which yy are taken as being evenly weighted.
Is function(mm, ww_m). (Will be defined based on yy [and maybe ww_y] before being passed in.) Returns vector of length(mm). Gradient of objective function.
Is function(mm, ww_m). (Will be defined based on yy [and maybe ww_y] before being passed in.) Returns matrix of dimensions length(mm) by length(mm). Hessian of objective function.
A curvature function object (giving mathfrak(C) in the paper; and related to "C" in the paper). See UMR_curv_generic() and examples. This is generally a "curried" version of UMR_curv_generic with densfunc and BBp passed in.
This is the error (cumulative) distribution function, a function object. Function accepting vector or matrix arguments.
Initial value of estimate ('mm'). Vector, length may be different than length(yy). See 'counts' input.
Together 'init' and 'counts' serve as the initialization; the implied initial vector is rep.int(init, counts).
Stepsize for moving out of saddle points.
A number of iterations. May not use them all. MM is not exactly the total number of iterations used in the sense that within each of MM iterations, we will possibly run another algorithm which may take up to MM iterations (but usually takes many fewer).
Used as tolerance at various points . Generally algorithm (and some subalgorithms) end once sum(abs(mm-mmprev)) < tol, or you hit MM iterations.
Collapsing roughly equal mm values into each other.
integer value (generally << MM). Every 'printevery' iterations, a count will be printed and the output saved.
filename (path) to save output to.
Uses first order (gradient) for optimization, and uses certain second derivative computations to leave saddle points. See Balabdaoui, Doss, and Durot (2021). Note that yy and mm (i.e., number covariates) may have different length.