MaxSA - Example

SimAnn

Included in the package is a sample program simann.ox where a simple objective function according to Judge et al. is implemented.

The main program essentially looks like

#include <oxstd.h>      // Include the Ox standard library header
#include <oxfloat.h>
#import <maximize>
#import <packages/maxsa/maxsa>

main()
{
  decl vP, dFunc, dT, ir;
  
  println ("\nEstimating using MaxBFGS");
  ir= MaxBFGS(fnc, &vP, &dFunc, 0, TRUE);
  println (MaxConvergenceMsg(ir), " at parameters ", vP', 
           "with function value ", double(dFunc));

// Estimate using MaxSA, to find global optimum
  println ("\nEstimating using MaxSA");
  ir= MaxSA(fnc, &vP, &dFunc, &dT);
  println (MaxSAConvergenceMsg(ir), " at parameters ", vP', 
           "with function value ", double(dFunc), " and temperature ",
           double(dT));
}

On the output, you'll see information like

Estimating using MaxBFGS
Strong convergence at parameters 
       2.3545     -0.31919
with function value -20.9805

Estimating using MaxSA
Strong convergence at parameters 
      0.86480       1.2357
with function value -16.0817 and temperature 7.45058e-08

which indicates how MaxBFGS indeed did not get away from the local maximum, whereas the Simulated Annealing algorithm managed to get away to the global optimum.

Charles S. Bos

Department of Finance

MaxSA - Example

SimAnn