Genetic Optimizer: Population Count
Author: akuzn
Creation Date: 12/4/2013 9:13 PM
profile picture


Good day.

Does population count in GA depends on real range and step of parameters?
For example if i use as parameter type of signal [0..3] population still be as set in settings (100 by default) or it can use less population?

profile picture



No, population count is static does not vary with the number of strategy parameters. Note: GA won't reprocess duplicate parameter sets, thus saving processing time.
profile picture


But wich step is used?
For example:
Please log in to see this code.

I m not sure but i suppose that it will be something like that:
Range = 30-2 = 28;

StepFitness = 28 / 100 = 2.8. and GA will produce population of 100 anyway. In fitness funciton computing step in parameters will be 2.8 and with (int) if needed accordingly to value type.
Certainly overall number of steps and Fitness computing wont be 100, but seems only population count 28 is needed.
So overall computing time could be less.

I ve implemented some GA and EV codes in strategy and it seems to me less population for small parameters range gives time economy and good result.
I suppose you have realised extension convenient to any even scientific conditions. But seems to me strategies require less precision.
Just trying now to tune GA and WFO computing time.

If i use in GA settings biggest range of parameters used in strategy but less than 100 will it be right decision?
Tell me if im wrong.
profile picture


According to the GA manual:

No formal rule exists for finding the best combination. We recommend starting your optimization using default values for up to 200K runs required.
profile picture


I dont know how your ga is realised. Just trying to get more usefull informartionthan thatn can be read in manual.
profile picture


for such a small range on a single parameter. why bother with GA. Shouldn't exhaustive evaluation provide you with enough computations to find your best solution?