Start point automatic setting in nlinfit or lsqcurvefit.

Hello,

I have a problem finding good initial values for nlinfit or lsqcurvefit...

I found it is possible to automatically set the 'optimal' start point in 'cftool', but I couldn't know how it works.

I want to find an automatic way that both data-sets get fitted in an acceptable way without changing the starting points manually.

My fitting model is not linear model but 'powar-law' model. (A*k^b + c).

Any help is much appreciated!

Thanks!

Answer 1

[Matt J]

I'm assuming the independent variable in y = (A*k^b + c) is k? If you have a sample of y at k=0, that would be a natural estimate for c. Alternatively, if b<0, the asymptotic value as k-->inf would be an estimate for c.

Once you have c, you can approximately linearize the model

   log(y-c) = log(A)+b*log(k)

Note that the above is linear in log(A) and b, so you can do a linear solve

        p=[ones(length(y),1), log(k(:))]\log(y(:)-c)

        A=exp(p(1));
        b=p(2);