Dear Ox-users,
While working with a not-so-smooth likelihood function, I needed an
optimization algorithm which would be more robust than the gradient-based
methods.
I came upon the Simulated Annealing algorithm as implemented by Bill Goffe
in Gauss and Fortran, and translated it into Ox. With his permission, I
published the result on a web page
http://www.tinbergen.nl/~cbos/software/maxsa.html
from which you can also download the code:
http://www.tinbergen.nl/~cbos/download/maxsa.zip
The interface is made to mimick the MaxBFGS interface closely, such as to
allow for an easy change between the two. Examples are on the webpage.
Note that the robustness of simulated annealing does come at a price:
Convergence can be quite a lot slower.
Greetings,
Charles Bos
|