Scaling

Scaling the data and other constant parameters of the objective function so that the elements of the gradient are of approximately the same order of magnitude will help improve accuracy of the Hessian approximation. This can help a lot in obtaining convergence, and the results will have higher accuracy. This program illustrates. The output is

/home/sh/Dokumenter/octave/octave-forge/main/optim/doc/tip1.out
You can see that the scaled data gives a more accurate solution, using less than half the iterations.



Søren Hauberg 2008-04-29