In order to fit a non-linear regression model, e.g. with the
function in R, you usually need to provide start values in order to
obtain reasonable fits. Usually, it does not make sense to use generic
start values, since they need to be as close as possible to the true
parameter values of the given model instance.
procedures for guessing start values from given data are of interest.
In this post I present on general idea for such procedures. It's an idea
rather than a procedure itself, because it is not guaranteed to be
always applicable and still requires some calculation and creativity
from its user to obtain a formula or algorithm for a specific model.
(This was originally posted on my old wordpress blog and not imported in full text.)