nlr (version 0.1-3)

nlr: Non-Linear Robust fitt.

Description

This is generic function fits a nonlinear mixed-effects model using robust methods described in Riazoshams et al (2015), allowed to be correlated and/or have unequal variances.

Usage

nlr(formula, data = parent.frame(), start = getInitial(formula, data), 
control = nlr.control(minlanda=1 / 2 ^ 10, 
maxiter=25 * length(start)), weights = NULL, robustobj = NULL, 
robustform = c("hampel", "huber", "bisquare",
"andrew", "halph huber", "hampel 2", "least square"), 
varianceform = NULL, tau = NULL, correlation = NULL, 
covariance = NULL, ...)

Arguments

formula

can be a nl.form object of the nonlinear function model, or a model formula with the response on the left of a ~ operator and an expression involving parameters and covariates on the right. See nl.form object.

data

an optional data frame or list of data with the response and predictor as name of variable. In heterogeneous case if it include response variable values of heterogenous variance function it asume variance function is function of predictor \(H(x_i,\tau)\), otherwise it assume is a function of predictor \(H(f(x_i,\theta),\tau)\).

start

list of starting value parameter, name of parameters must be represented as names of variable in the list.

control

nlr.control, include controles for nlr, see nlr.control.

weights

User optional matrix of variance covariance matrix of error, a general weight that can be used by user, but if correlation, covariace or varianceform argument given, then will be ignored.

robustobj

=NULL, optional nl.form object of robust loss function defined by user. User can define his/her own function, to be nl.form object, Or use defined robloss function in robustform.

robustform

nl.form object of robust function used for downgrading. functions("hampel","huber","bisquare","andrew","halph huber","hampel 2","least square").

varianceform

NULL, nl.form object if given heterogeneous variance will be fited.

tau

NULL, list or data.frame of initial values for heterogeneous variance function parameter. the stored value in vardnc object of nl.form will be stored.

correlation

autocorrelated error, form of corStruct but not in nlme, is extra arguments to any of "nl.forms".

covariance

optional covariance matrix of errors. If given Generalized estimates will be calculated.

…

Any extra arguments to any function sources, such as nonlinear regression model, heteroscedastic variance function, robust loss function or optimization object function.

Value

Output depends on the method called as:

• Nonlinear Least Square Estimate(NLLS): nl.fitt object.

• Generalized NLLS: nl.fitt.gn object.

• Robust Estimate: nl.fitt.rob object.

• Generalized Robust Estimate: nl.fitt.rgn object.

Details

nlr is non-linear robust inference. It is mostly gradient and hessian based. The classic estimates also is avilable.

References

Riazoshams H, Midi H, and Ghilagaber G, 2018,. Robust Nonlinear Regression, with Application using R, Joh Wiley and Sons.

See Also

nl.form, nl.fitt, nl.fitt.gn, nl.fitt.rob, nl.fitt.rgn

Examples

# NOT RUN {
## Chicken fitt without hetroscedasticity
 d<-list(xr=Weights$Date, yr=Weights$Weight)
 fitt.chicken1 <- nlr(nlrobj1[[2]],data=d,control=nlr.control(method = "OLS",trace=TRUE))
fitt.chicken1$parameters
plot(fitt.chicken1)
## Chicken fitt with hetroscedasticity
# RME by default
fitt.chicken2<-nlr(formula=nlrobj1
                   [[14]],data=d,start=list(p1=2300,p2=42,p3=.11), 
                   robustform ="hampel",
                   tau=list
                   (sg=.09,landa=1),varianceform=nlrobjvarmdls1[[1]],
                   control=nlr.control(tolerance=1e-3))
fitt.chicken2$parameters
fitt.chicken2$hetro$parameters
plot(fitt.chicken2)
# autocorrelated case
xr = trade.ir[, 1]
yr = trade.ir[, 2]
a1 <-  nlr( nlrobj5[[18]],data=list(xr = xr, yr = yr),
correlation=list(StructName="corAR1"))
a1$parameters
plot(a1)
# }