what to do if a nls model fails

[ybukhman]

1. Currently, if nls fails to converge, GCAT immediately switches to an alternative model. See if changing the initial guess would help first. Try randomly changing all initial guess values by 10% and fitting again. If still fails, what about a larger change, e.g. 30% or 50%?

2. If Richards model fails, GCAT switches to logistic. This is not necessarily the best choice. Try fitting both logistic and Gompertz, then choose the one with the lower RSS.

3. Record the fact that a fit has failed in a slot in the well object.