The function searches for new starting values. Either in a random fashion or based on a smarter search algorithm. The smart search algorithm is a rewritten version of the search algorithm implemented in the Apollo package. The function is rewritten to work with the current modeling framework.
search_start_values(
ll,
estim_env,
model_options,
control = NULL,
type = "simple",
n_candidates = 100,
n_return = 10,
multiplier = 1
)This is the 'raw' log-likelihood function passed to the estimation routine. It is important that the user takes into account whether the optimization routine is a minimizer (e.g. 'ucminf') or a maximizer (e.g. 'maxlik').
An estimation environment or list of estimation environments
returned from prepare
A list of model options. Note, this list is validated
for a second time here to set some necessary defaults. See
validate for details.
A list of control options that are passed to
set_controls.
A character string indicating whether to use a simple or a smart search algorithm. Default is 'simple'
An integer giving the number of candidates to evaluate. Default is 100.
An integer giving the number of parameter vectors to return. The default is 10
A double indicating a multiplier for the 'simple' search algorithm. The default is 1.
A matrix of starting values
Hess, S. & Palma, D., 2019, Apollo: A flexible, powerful and customisable f reeware package for choice model estimation and application, Journal of Choice Modelling, 32 Bierlaire, M., Thémans, M. & Zufferey, N., 2010, A heuristic for nonlinear global optimization, INFORMS Journal on Computing, 22(1):