Best penalty lower than best individual in genetic algorithm

Hi,

I am doing an optimization with the generic algorithm, using some integer parameters. The best penalty I get for many generations is lower (~ -5 million) than the fitness function of the best individual (~ -4 million).

Can you help me interpret this? I thought that the penalty value was the fitness function for feasible individuals and the fitness function of the population plus some penalty for non feasible individuals, so my understanding was that the fitness function of the best individual would be lower than any best penalty of the entire optimization.

Thanks

Answer 1

[Alan Weiss]

I believe that your interpretation of the meaning of the penalty function is correct, so I, too, am puzzled by what you report.

If you can reproduce your problem, I suggest that you contact technical support, and possibly file a bug report.

Alan Weiss

MATLAB mathematical toolbox documentation

[Jorge Mallo]

Thanks for your answer Alan, will file the report.

Flow up question: I was thinking in not using integer parameters from the point of view of the generic algorithm optimization, and then inside the cost function covert float parameters to integer using floor or ceil. Any drawbacks to this approach?

[Alan Weiss]

You are free to do what you like with the genetic algorithm. If you have linear or nonlinear constraints, though, you might find that your problem becomes infeasible with your approach. And I am not sure that the cost function is the only thing you would need to change to get integer outputs where you want them (what about mutation and crossover?).

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[hanane ensam]

Hello, please, could you give me the interpretation of the graph (penalty value in function of generation) and the exact definition of penalty value

[Alan Weiss]

The penalty function for integer ga is described here.

Alan Weiss

MATLAB mathematical toolbox documentation