optim_weights: Weight functions for optim_fit

In OptimModel: Perform Nonlinear Regression Using 'optim' as the Optimization Engine

Weight functions for optim_fit

Description

Weight functions for optim_fit. May be used when fit.method=="irwls" or fit.method=="mle". Generally, not called by the user.

Usage

 
       weights_varIdent(phi, mu)
       weights_varExp(phi, mu)
       weights_varPower(phi, mu)
       weights_varConstPower(phi, mu)
       weights_tukey_bw(phi = 4.685, resid)
       weights_huber(phi=1.345, resid)

Arguments

phi	Variance parameter(s)
mu	Vector of means
resid	Vector of model residuals

Details

• weights_varIdent returns a vector of ones.

• weights_varExp returns \exp(\phi*\mu)

• weights_varPower returns |\mu|^{\phi}

• weights_varConstPower returns \phi_1 + |\mu|^{\phi_2} where \phi_i =\phi[i]

• weights_tukey_bw is a Tukey bi-weight function. Let

  r=\tfrac{|\text{resid}|}{\text{mad}(\text{resid},\text{center}=\text{TRUE})}.

  Then this function returns

  \left(1-\left(\tfrac{r}{\phi}\right)^2\right)^2\text{ whenever } r <= \phi\text{ and }0{ o.w.}

  For this the user should use phi.fixed=TRUE in the optim_fit function.

• weights_huber is a Huber weighting function that returns \min(1, \phi/r), where r = |\text{resid}|/\text{sig} and \text{sig} = \text{mad}(\text{resid}, \text{center} = \text{TRUE}). For this the user should use phi.fixed = TRUE in the optim_fit function.

Value

A vector of numeric weights.

Author(s)

Steven Novick

See Also

optim_fit, rout_fitter

OptimModel documentation built on May 29, 2024, 9:47 a.m.