[Scilab-users] Problem with function fitting

[Serge Steer]

Your problem comes from the fact that for some values of n the computed
value of e are complex values (some negative values to a non integer
exponent. This can be made obvious with the following function

function e = G(n,z)
  xm = z(1)
  ym = z(2)
  e = ym - f1(xm,n)
  if ~isreal(e) then pause,end
 endfunction 

The solution depends on the problem you want to solve. for example one
can replage G by
function e = G(n,z)
  xm = z(1)
  ym = z(2)
  e = ym - f1(xm,n)
  e=real(e*e')
endfunction

Serge Steer

Le 17/12/2014 18:53, Maihem a écrit :
> Hi,
>
> I have problem with fitting function to measured data using scilab. I tried
> scilab function datafit(), lsqrsolve(), leastsq() but none returns me a
> proper result or can't perform calculations.
>
> //I have measured data:
> xm;ym
> 2.0;99.9449173761
> 1.0;99.8097145719
> 0.5;97.9769654482
> 0.25;36.4046069104
> 0.1;1.4872308463
> 0.071;0.5207811718
> 0.063;0.3705558338
>
> //Script used to fit data
>
> miny=min(ym);
> maxy=max(ym);
> minx=min(xm);
> maxx=max(xm);
>
> Zz=[xm;ym];
>
> //Function describing problem (I want to find best fit by finding n()).
>
> function g1=f1(x, n)
> g1 = miny+((maxy-miny) ./((1+(n(1).*(x ./n(2)).^n(3))+(1-n(1)).*((x
> ./n(2)).^n(4)))).^n(5))
> endfunction
>
> function e = G(n,z)
>   xm = z(1)
>   ym = z(2)
>   e = ym - f1(xm,n)
> endfunction 
>
> n0 = [1 ; 1; 1; 1; 1]
> [n0_opt,err] = datafit(G,Zz,n0) 
>
> After calling function scilab returns me error:
> Variable returned by scilab argument function is incorrect.
>
> I tried to find solution for this problem in scilab mailing lists and tried
> some posted scripts and advices but I can't localize problem.
>
>
> Łukasz
>
>
>
> --
> View this message in context: http://mailinglists.scilab.org/Problem-with-function-fitting-tp4031546.html
> Sent from the Scilab users - Mailing Lists Archives mailing list archive at Nabble.com.
> _______________________________________________
> users mailing list
> users at lists.scilab.org
> http://lists.scilab.org/mailman/listinfo/users