Search through a range of initial parameter values to obtain convergence

searchdrc provides a facility for searching through a range of initial values for a single parameter in order to obtain convergence of the non-linear estimation procedure used in dose-response curve fitting.

Usage

searchdrc(object, which, range, len = 50, verbose = FALSE)

Arguments

object: an object of class 'drc', which must have valid $start and $parNames fields populated. This is typically an object from a model that failed to converge but was still constructed with initial parameter values.
which: a character string containing the parameter name without the curve suffix (e.g., "b" not "b:1"). Must exactly match one of the parameter names in the model object.
range: a numeric vector of exactly length 2 specifying the interval endpoints c(lower, upper) for the search range. The two endpoints must be different.
len: a positive integer (minimum 2). The maximum number of evenly spaced starting values to try within range. The search stops early as soon as convergence is achieved, so the actual number of attempts may be less than len. Defaults to 50.
verbose: logical. If TRUE, prints progress messages indicating which starting value is currently being tried. Defaults to FALSE.

Value

If convergence is achieved, returns the fitted model object of class 'drc', corresponding to the first starting value in the search grid that led to a successful fit. If no starting value leads to convergence, the function throws an error.

Details

The function iterates through at most len evenly spaced values within the specified range, using each as a starting value for the chosen parameter. The search stops as soon as the first successful model fit is found. You would need to identify the parameter which is most likely to cause problems for the estimation procedure.

Parameter names should be provided without the curve suffix. For example, use "b" rather than "b:1". The function internally matches the parameter using the pattern "^<which>:" against the full parameter names stored in the model object.

Author

Christian Ritz, Hannes Reinwald.

Examples

if (FALSE) { # \dontrun{
library(drc)

# Fit an initial model (which may fail to converge)
myModel <- drm(response ~ dose, data = myData, fct = LL.4())

# Search over a range of starting values for the slope parameter "b"
myModelFixed <- searchdrc(myModel, which = "b", range = c(-5, 5), len = 100)

# With progress messages enabled
myModelFixed <- searchdrc(myModel, which = "b", range = c(-5, 5),
                          len = 100, verbose = TRUE)
} # }