Re: Data fitting with fmins

[Andreas Stahel]

Tweety <ftjp <at> gmx.de> writes:

> 
> 
> Am 19.07.2016 um 18:35 schrieb Andreas Stahel:
> > Tweety <ftjp <at> gmx.de> writes:
> >
> >> Dear group,
> > DELETED TEXT
> Dear Andreas,
> 
> thanks for the eplanation. I am still trying to figure out the tricks 
> you used. To be fair, I arbitrarily picked the dataset. I have others 
> which can be fitted much much better with a double exponential. The 
> devil is in the detail.
> 
> But thanks, this is solved now.
> 
> Best regards,
> Jan
> 

Dear Jan
Thank you for the good news, but I doubt that your problem is solved.
Have a look at the graphical output and try to restart you computation with
your supposed solution. Other codes can not determine the variances of your
parameters p(). This is a strong indication of a problem.
Compare with the fit of one exponential only.

Nonlinear regression is often more difficult than just throwing some
standard code at it. You have to proceed step by step and do visual
verifications, if possible.

Hope this helps

Andreas

P.S. please correct the typo in my code for the function f2 

%% now with two exponential, which will fail miserably
function res = f2(x,p)
   res = p(1) + p(2)*exp(p(3)*x) + p(4)*exp(p(5)*x);
endfunction