Is FEs the same as maximum iteration?
In most recent articles in meta-heuristic algorithms when the test benchmark mono functions are tested by proposed algorithms, the "maximum iteration" is not used in literature, instead "independent Run" and "maximum number of function evaluations, FEs" is seen. Does any one know these concepts?
Also about new versions of ICA algorithm that worked on assimilation part and angle Teta, does any one have its Matlab Code?
Some of the meta-heuristic algorithms uses population and generation. Therefore, the function is executed several times per iteration/generation.
The number of fitness evaluations is the number of times you evaluate a solution in the meta-heuristic algorithm. It is NOT the same as the number of iterations, since you can evaluate the fitness function more than once in one single iteration (as Ahmed says).
The reason why the maximum number of fitness evaluations is preferred to the maximum number of iterations is because each single fitness evaluation provides some information of the problem. Thus, limiting the number of fitness evaluations your are limiting the "amount of information" one algorithm can obtain from the problem. This way it is fair to compare two algorithms. When doing a comparison, all the algorithms should be run for the same number of maximum fitness evaluations. Otherwise, the algorithm evaluating the fitness function more times has, a priori, more chances to provide a better solutions, since it got more information of the problem.
Dear Francisco Chicano,
Thanks a lot. You're right. I got it. FEs is used in many recent scientific journals for comparison of new algorithms with previous ones.
Your idea means the following statement and formula, Isn't it?
Since the number of operators in each algorithm are different, the number of fitness function call or fitness function evaluation is different. In this case, we may say: The number of fitness evaluations (FEs) is equal to the number of fitness function calling in each iteration * the number of total iterations.
Regards.
Dear Farshid
Mr. Francisco Chicano is absolutely right about FEs. You can say FEs = Iterations*Dimensions (for each dimension a fitness function will be called)
The maximum number of function evaluations is regarded as a fair measure to compare algorithms in order to make sure that all algorithms sample the search space an equal number of times.
However, a recent study for PSO has showed that the number of individuals is important as well:
Andries P. Engelbrecht: Fitness function evaluations: A fair stopping condition? SIS 2014: 181-188
The FEs is the number of times that the objective function is evaluated and it is different from the number of iterations. Because in one iteration we have more that one evaluation of the objective function.
https://www.researchgate.net/post/Why_does_Function_Evaluations_number_significant_for_the_superiority_of_an_optimization_algorithm
- Badges
- Science topic