Dear all,
I am trying to fit some experimenta data which has a S-shaped form into a sigmoid function (which is also S-shaped). I am using the generalised logistic function as custom equation in the curve fitting toolbox, but I am unable to get a fit to the data.
The current default curves yield by the matlab toolbox are completly off the data. I have tried different (and more simple logistic functions), different fititing algorithms and different robustness methods. I am unsure how MATLAB computes the fit curve, but playing around with the starting points of the coefficients, sometimes I can get a good fit, eg. starting points: a = b = c = k = 0 and q = 1, yields a relative good fit:
A different set of initial values (eg. starting points: a = b = c = k = 0 and q = 2 or 0.5) yields a completly different fit.
I am seeking a robust way of fitting the data, (I need to do this for thousands sets of data), but I have been struggling to do such a simple fit. It may be the starting point of the fit, the method or even the equation.
Any suggestion for obtaining a robust fit (unsupervised) of S- and Z-shaped data would be extremely welcome.
Daniel.
