Hi, generally the max no. of generations are fixed based on some trial runs. However it depends upon the population size, preciseness of definition of objective function and constraints, use of real valued or binary valued chromozomes, method of selecting chromosomes for reproduction, crossover type (single point/multipoint), crossover rate, mutation rate, etc.

Thank you for your response.

I want to know if a solution that needs a large number of iteration for convergence, is acceptable or we have to find a solution with fewer iteration numbers, considering that the problem have a large number of constraints and variables.

hi, dear friend. I agree with Mr.Vijay Gs. this matter depends on your other conditions and your problem. In addition, I think if you work a problem a large number of constraints and variables , the best work is try  by several numbers or it's an old saying "Interpolation" but not exactly Interpolation.

Dear Farhad Adelnia

It varies from problem to problem. For selection the best population size for any metaheuristic optimization algorithm is to do repeated runs by varying the population size and select the one corresponding to which are algorithm gives the best fitness value.

Regards

Mahamad Nabab Alam

Thank you for your answers. 

I know that it depends. Also, I select the best one by which the algorithm converges to a good answer. But, this best value for generation size for a particular case is 5*10^4. I want to know that if this large number for generation size is acceptable. 

In other words, I want to know that if the solution which converges to a good answer in such number of generation size is acceptable.

Thank you

Dear Prof. Adelnia,

                                 I think that to compute the number of offsprings which in a String Theoretic Quantum Mechanical Linear GA Framework i.e. involving actual computation with conservation of information and nanoscale energy input, such as please refer to my coal mining papers on my RG page the maximum number of iterative steps is six for the meta game i.e. the game/algorithm to set up the iterations. Regarding the actual number of offsprings for the objective function iteration it would depend on the Higgs-Bosonic resolution of the computing machine since for example there would be approximations which would depend on the current flow and approximation steps for the iterations. My son Sandipan Mallick, Nick Hamburger and I have dome some Quantum Industrial Physics computations with these using analytical physical chemistry with weakly radioactive nanomaterials constructed econophysically in design, but that is all at working paper stage and some results have been submitted to the Nobel Prize and Nemmers Prize Committees and to Com Hem, American Wire Co. and Siemens Ltd. where these technologies will be used for manufacturing Nano Wires and Nano Circuits and also Systems Integrated Furnaces for the materials conversion. I hope this answer helps you in some ways. The work is painstaking and nerveracking and involves decades of hard work from very close quarters with high degree of politically charged confidentiality, is built on the analysis of nanomaterials some of which do not exist at present but have no substitute as legally affordable means of development through legal and scientific  methods of trade and commerce and "quantum" mercantilism. Earl Chair Prof. Dr. SKM QC EPS Fellow (In) MES MRES MAICTE

Roughly speaking it is not the number of generations what counts.  You should rather consider the number of objective function evaluations instead.  This, in turn, is a product of population size times number of generations, if only the population size is not too small,. see my article at RG: https://www.researchgate.net/publication/216486175_Biology_Physics_Small_Worlds_and_Genetic_Algorithms?ev=prf_pub

There you will find, among other things, the raliable estimate of a minimal population size.  I bet you'll be amazed how small this number might be. On the other hand, GA's  are 'sentenced' for success --> https://www.researchgate.net/publication/1958790_Amazing_geometry_of_genetic_space_or_are_genetic_algorithms_convergent?ev=prf_pub

And going back to your last question.  GA will not tell you whether its final solution is, or is not acceptable.  You have to be the judge.  Of course, the numerical value of fitness will be very helpful but, again, it is you who designs the fitness function.  Anyway, 50 thousands of generations looks like an overkill.

Chapter Biology, Physics, Small Worlds and Genetic Algorithms

Article Amazing geometry of genetic space or are genetic algorithms ...

Farhad, there's no such thing as a maximum number of fitness evaluations. EAs, like other metaheuristics, is an anytime algorithm so the longer you run it the better the solutions you get. Actually, under some weak conditions (elitism + ergodic mutation) you are guaranteed to find the optimal solution asymptotically. Then again, this is not what you would do in practice. Basically, the goal is spending your computational budget in the best way so your algorithm has to take a decision at every step on whether continue running with the current population or re-start it (completely, as if it was a new run of the algorithm, or partially, re-utilizing some of the solutions found). You can take this decision based on different factors: number of evaluations without improvement, convergence of the population, etc. You need to find out what works best for your problem. Do not think too much about absolute numbers, I have worked on problems for which millions of fitness calls were done. As long as this is computationally affordable, it is no problem.

Hi, I would suggest you set a maximum number of adjacent iterations where the fittest solution value is not moving.

Simple difference between generation and iteration in evolutionary algorithms?

Dear Raghav Parouha,

Replace the word generation with population (set of chromosomes) and iteration with single evolutionary cycle. This way things may become more clear and easy to understand.

According to the problem and methods the no of iterations vary