I have seen MINPACK but I don't want to calculate the Jacobian Matrix and so
on...
I want to minimize the system of 2 parameters:
f(x,y) - fo
g(x,y)-go
(with f and g non linear functions)
with
x> xmin
y>ymin
In Matlab, I use to do:
Ko = [x_init; y_init];
K = lsqnonlin('mymatrixfuntion',[xmin xmax; ymin ymax],Ko);
with my function defined as below:
____________
function rest = mymatrixfunction(K)
x=K(1);
y=K(2);
rest = [f(x,y) - fo; g(x,y)-go];
______________
If you have limits, the easiest way is successive
range division in half - for 1D anyway.
I assume there is a 2D version of this. Not hard to
write something that may work.
e.g. pick an x, with this x look for min sqd err between
the y -limits by succssive div range by 2, then use this y
and go to do the same with x. Would it converge - and how do you
know when ? Good question!
The problem with non-linear lsfit - is that you don't
know if there's more than one error "pothole". You
could get trapped in a shallow pothole, while there's
a crater "just over there ..."!
If you know there's theoretically only one, it's easy.
Chris
Yes, I've seen the same advises in
http://irfu.cea.fr/Projets/COAST/methodes_numeriques_MINI.pdf
It seems to be the Amoeba (amibe in French) optimisation
Well I'll write it....
Thanks
>
> Yes, I've seen the same advises inhttp://irfu.cea.fr/Projets/COAST/methodes_numeriques_MINI.pdf
> It seems to be the Amoeba (amibe in French) optimisation
>
> Well I'll write it....
>
> Thanks
... or for example copy the engine from Numerical Recipies, F77,
chapter 10.4 and 10.7: http://www.nrbook.com/a/bookfpdf.php
Svend
I have just opened the file again using FireFox 3.0.5. Just prior to
downloading of the pdf-file you have to accept an automatic
instalation of an Adobe file plug-in. Check your settings allow plug-
ins etc.
Svend