Hi,
I am using CMAES to maximize the posterior likelihood of a large scale model, after repeated unsuccessful attempts with csminwel. CMAES did a great job, and converged to a value of around 3300. But when I used the mode as a new initial guess, I was able to go down by 50 points. I have two questions:
-
How can I control CMAES settings? For instance, to deal with local maxima, I was thinking to increase the population size and/or have multiple restarts.
-
If anyone else has been playing with this algorithm, suggestions on which options to change to improve its performance (in terms of finding the global max) are very welcome.
Anyway, I think CMAES is an excellent algorithm, thanks for adding it to Dynare!