nlrob.control function

Control Nonlinear Robust Regression Algorithms

Allow the user to specify details for the different nonlinear robust regression algorithms in nlrob.

nlrob.control(method,
              psi = c("bisquare", "lqq", "welsh", "optimal", "hampel", "ggw"),
              init = c("S", "lts"),
              optimizer = "JDEoptim", optArgs  = list(),
              ...)

Arguments

• method: character string specifying the method

• psi: string specifying the psi-function which defines the estimator.

• init: for some methods, currently, "MM" only, a string specifying the initial estimator.

• optimizer: currently only "JDEoptim" from package list("DEoptimR").

• optArgs: a list of optional arguments to the optimizer. Currently, that is JDEoptim from package list("DEoptimR").

• ...: optional arguments depending on method, such as fnscale, tuning.chi or both tuning.chi.tau and tuning.chi.scale; for method = "MM" also optim.control

  to be passed to the optim(.., hessian=TRUE) call.

  Internally, nlrob.control() will choose (or check) defaults for the psi/rho/chi related tuning parameters, also depending on the method

  chosen; see e.g., the Examples .

Returns

a list with several named components. The contents depend quite a bit on the method.

See Also

nlrob; for some details, nlrob.algorithms.

Examples

## Show how the different 'method's  have different smart defaults :
str(nlrob.control("MM"))
str(nlrob.control("MM", psi = "hampel"))# -> other tuning.psi.M and tuning.chi.scale defaults
str(nlrob.control("MM", psi = "lqq", tol = 1e-10))# other tuning.psi.M & tuning.chi.scale defaults
str(nlrob.control("tau"))
str(nlrob.control("tau",psi= "lqq"))
str(nlrob.control("CM")) # tuning.chi undefined, unneeded
str(nlrob.control("CM", psi= "optimal"))
str(nlrob.control("mtl"))