Get Initial Parameter Estimates for Nonlinear Least Squares Models.

getInitial

Description

Used by nls to construct initial parameter estimates for a nonlinear regression model.

Usage

# Generic function
getInitial(object, data, ...)
## Default S3 method:
getInitial(object, data, mCall, LHS = NULL, ...)
## S3 method for class 'formula':
getInitial(object, data,...) 
## S3 method for class 'selfStart':
getInitial(object, data, mCall, LHS = NULL, ...)

Arguments

object	a formula or a selfStart model that defines a nonlinear regression model
data	an environment, or something that can be used as an environment such as a list or data.frame, in which the expressions in the formula or arguments to the selfStart model can be evaluated.
mCall	a named list of name objects: the names of the list are the names of the arguments of the objective function. The name objects are the names used in the call to the objective funcion. Often this is the result of as.list(match.call(ObjectiveFunc, RHS)), where RHS is the call to ObjectiveFunc on the right hand side of the model formula. See the description of the argument initial from the function selfStart.
LHS	the expression from the left hand side of the model formula. See the description of argument initial from the function selfStart.
...	optional additional arguments.

Details

If object is a formula and if data has an attribute called "parameters" then that attribute is returned. Otherwise the formula is examined to see if its right hand side is a call to a selfStart object whose initial attribute can be evaluated with the given data.

If object is a selfStart function, then you must also supply the call to it as mCall and the response as LHS. The initial attribute of object is called with those arguments.

Any other class of object causes an error.

This function is intended to be used by nls but can be used to test functions created with selfStart.

Value

returns a named numeric vector, or a list of starting estimates for the parameters, suitable for use as the start argument to nls.

The construction of some selfStart models is such that these starting estimates are, in fact, the converged parameter estimates. For others, they are estimates based on a heuristic analysis of the objective function.