create nlr.control options for using in diferent part of the package.
nlr.control(maxiter = 50, tolerance = 0.0001, minscale = 0.001, trace = F,
minlanda=1e-16,derivfree=F,robscale=T,
algorithm = c("Levenberg-Marquardt", "Nelder-Mead","Gauss Newton"),
method=c("default","RME","CME","CLSME","RGME","WME","MLE","OLS","TS","RTS","lms"),
initials=c("manuall","lms","OLS","quantile"),history=F,length.out=NULL,singlePlot=F,
singularCase=1,
JacobianLeverage = c("default", "classic", "robust"))maximum number of iteration. To be used optimization or fitt procedures.
Tolerance of convergence.
Minimum of scale value.
Default is False, if True draw the graph of convergence in iterations, depends on availability for that cases.
Minimum of landa coeficient values in stepest descend or levenberg marquardt. When the step length does not reduce the objective function this value decrease.
Default value is False. If True the implicitly program use derivative free optimization methods based on Nelder-Mead method. This is very slow convergence in that case the number of maximum iteration should be increased, at least maxiter=500 is sugested. nlr is derivative based but in the cases where problem happenes in derivative computation Nelder-Mead will be used.
Default is True, when False the clasic standard error of residuals will be used in computations. It is not recomneded to use this option because reduce the robustness of estimators.
of optimization method, default is "Levenberg-Marquardt", is robust in computation when outlier happense. "Nelder-Mead" is derivative free method, and "Gauss Newton" method is used for fast computation but might face with some singularity in hessian when outlier happense.
Select how to solve the singular gradient matrix case in MM-estimate procedure of levenberg marquard method. The efault value equal 1 add up a value to diagonal elements, and 2 add up proportion to size of diagonal values. They are same but Generally case 1 works in most of cases, but in case of divergence or singularity problem can use 2.
character name of jacobian leverage to be used incomputation.
"default":default value assigned by any function
"classic":cllasic nonrobust value
"robust":robust jacobian leverage value
of computation using several type of estimators.
"RME": Heteroscedastic error Robust Multi Stage Estimate.
"CME": Heteroscedastic error Classic Multi Stage Estimate.
"CLSME": Heteroscedastic error Classic Least Square based Estimate.
"RGME": Heteroscedastic error Robust Generalized Estimate.
"WME": Heteroscedastic error Weighted M-Estimate, See Lim 2010.
"MLE": Maximum Likelihood Estimate.
"OLS": Ordinary Least square, constant and uncorrelated error.
"TS": Autocorelated error, classic Two Stage Estimate.
"RTS": Autocorelated error, Robust Two Stage Estimate.
"lms": Least median of squared residuals estimate. (Non efficient)
is used to define initializing parameters using a specific estimatro. Might be used when initial values are dificult to find manually, or getInitial function.
"manuall" default value is manually provided initial values by user.
"lms" robust high breakdown point least median of squares of errors estimator.
"ols" ordinary least square estimator.
"quantile" robust least quantile regression.
Length for incrementing independent variable to be used in ploting commands to have a more smoother curve.
Default=F, Used in plot command, to draw the history of convergence, default is F, the TRUE value force the function to draw the history.
Default=F, By default the plot function draw the fited model (in one dimentional case), and residuals in second collumn, singlePlot=T causes the plot to draw the two graph in two windows.
list of controls.
nlr.control function create nlr.control variable and mostly used as control argument to the functions. It define convergence and many other parameters depends on the function operation.
Riazoshams H, Midi H, and Ghilagaber G, 2018,. Robust Nonlinear Regression, with Application using R, Joh Wiley and Sons.
# NOT RUN {
## The function is currently defined as
"nlr.control"
nlr.control()
# }
Run the code above in your browser using DataLab