On 06.12.2013 21:00, smms wrote:
> So I would appreciate
> if someone could refer me to a good reference of how to make good guessing
> of initial parameters.
That's evidently impossible. If it there were such a thing as a
reference manual on how to do it, it would not be called "guessing".
Initial values have to come from your own understanding of the problem.
There is no magic bullet.
It says in "help fit" that "Non-linear least-squares fitting is an art!"
You're now learning just how true that statement is.
Anyway, your problem is not with the initial values, it's with the
scaling of the parameters themselves. You've missed an important lesson
from "help fit" there.
> a = 1e-7
> b = 10
> k = 0.5
> g(x) = a/(1+b*exp(-k*x))
> fit g(x) 'ss_nici_raw_processed_201.txt' u (1/$1):($2/800**3) via a,b,k
> plot g(x),'ss_nici_raw_processed_201.txt' u (1/$1):($2/800**3):($3/800**3) w
> errorbars not
Hmmm... so why rescale your y axis so heavily, if all that achieves is
giving your "a" paramter a magnitude that's wildly different from the
others?
a = 50.0
fit g(x) 'ss_nici_raw_processed_201.txt' u (1/$1):2:3 via a,b,k
would have been equivalent, quicker to type, and with the additional
benefit of actually using your y errors for the fit.
|
