donlp2Control: Control variables for Rdonlp2
In Rdonlp2: Rmetrics - An R extension library to use Peter Spelluci's DONLP2 from R

Description Usage Arguments

View source: R/donlp2Control.R

Collection of Control Variables

donlp2Control( 
    iterma = 4000, nstep = 20,fnscale = 1, report = FALSE, 
    rep.freq = 1, tau0 = 1.0, tau = 0.1, del0 = 1.0, epsx = 1e-5, 
    delmin = 0.1, epsdif=1e-8, nreset.multiplier = 1,
    difftype=3, epsfcn = 1e-16,taubnd = 1.0, hessian = FALSE,
    te0 = TRUE, te1 = FALSE, te2 = FALSE, te3 = FALSE,
    silent = TRUE, intakt = TRUE)

`iterma`	maximum number of iterations.
`nstep`	maximum number of tries in the backtracking.
`fnscale`	set -1 to maximize the object function.
`report`	If `TRUE`, a list object which contains detailed information will be passed to `control.fun()` in `donlp2`.
`rep.freq`	the frequency of report(positive integer). the report will be passed to `control.fun` every `rep.freq` iterations.
`tau0`	the positive amount how much any constaint other than abound can be violated. A small `tau0` diminishes the efficiency of DONLP2, while a large `tau0` may degarde the reliability of the code.
`tau`	gives a weight between descent for `fn` and infeasibility and is also used as a safety parameter for chosing the penalty weigths. It can be chosen larger zero at will, but useful values are between 0.1 and 1. The smaller tau, the more may `fn` be scaled down. Tau is also used as an additive increase for the penalty-weights. Therefore it should not be chosen too large, since that degrades the performance.
`del0`	The positive amount by which constraints are considered binding. If too small, the indentification of correct sets of binding constraints may be delayed. If too large, DONLP2 will escape to the full regularlized SQP method(quite costly). Good values are in [0.01, 1.0].
`epsx`	successful temination is indicated if the Kuhn-Tucker criteria are satisfied within the value.
`delmin`	constraints are considered as sufficiently satisfied if absolute values of their violation are less than the value.
`epsdif`	relative precision in the gradients.
`nreset.multiplier`	maximum number of steps (`nreset.multiplier` times `n`) until a “restart” of the accumulated quasi-newton-update is tried. Value should be integer between 1 and 4.
`difftype`	1,2,3. numerical differentiation algorithm. The algorithm with `difftype=2` is nearly identical to one used in `optim()`. See PDF manual attached in this package.
`epsfcn`	relative precision of the function evaluation routine.
`taubnd`	The positive amount by which bounds may be violated if numerical differention is used.
`hessian`	if `TRUE`, numeric Hessian matrix is calculated by numerical differentiation algorithm specified in `difftype`.
`intakt`	if TRUE, informations about current iteration step in optimization and final results are output to R console. The amount of information depends on `te0`, `te1`, `te2`, `te3`.
`te0`	if TRUE one-line-output for every step is printed.
`te1`	if TRUE post-mortem-dump of accumlated information is printed.
`te2`	if TRUE, more detailed information is 'pretty-printed'.
`te3`	if TRUE, the gradients and approximated Newton-Raphson updates(in upper triangler matrix) are printed.
`silent`	If `TRUE`, `donlp2()` runs quietly, i.e., `intakt=FALSE`, .pro/.mes files are not created, and `te0,te1,te2,te3` are omitted.