Hello,
I have a 12 variable exhaustive optimization I would like to run. I have created my own Custom Fitness Value for the optimization and everything appears to be running correctly.
My question is specifically about the optimization procedure. Is there a way I can tell the optimization or custom fitness value which variables are dependently grouped together? In theory my objective would be three smaller exhaustive tests and then taking the best setting from those.
Of the 12 variable there are 3 groups and these three groups need to be exhaustively tested. Each group is independent of the other two. Does this makes sense?
A setting of 1,2,3,4 vs 2,3,4,5 for the first four variables will have no impact on the output for the other 2 groups?
Hopefully my question is clear. Thanks for any help.
After writing this question out and reading it a few times I see that its not likely to find a viable answer. But I'll post anyways and see if anyone has a thought..
Exhaustive Optimization question
Re: Exhaustive Optimization question
Not an answer but some thoughts:
Custom fitness is of no need for exhaustive optimization, except for manually sorting / validating the result according to it. I mean, it has no impact on the result itself since we test every single combination anyway. While in genetic optimization it will orient the result...
If your groups of inputs are truely independant (no impact on each other), then you can optimize them one after the other. No need to do all at once.
But if there is dependancy between them, my best practice so far is an empirical approach: optimizing back and forth amoug input groups to slowy (and painfully) narrow to a balanced solution. The most difficult in this case is the starting point (chicken & egg dilemna): I have to arbitrary freeze one group of inputs in order to optimize the other.
Or you can always let the genetic engine do its cooking itself...
Custom fitness is of no need for exhaustive optimization, except for manually sorting / validating the result according to it. I mean, it has no impact on the result itself since we test every single combination anyway. While in genetic optimization it will orient the result...
If your groups of inputs are truely independant (no impact on each other), then you can optimize them one after the other. No need to do all at once.
But if there is dependancy between them, my best practice so far is an empirical approach: optimizing back and forth amoug input groups to slowy (and painfully) narrow to a balanced solution. The most difficult in this case is the starting point (chicken & egg dilemna): I have to arbitrary freeze one group of inputs in order to optimize the other.
Or you can always let the genetic engine do its cooking itself...
Re: Exhaustive Optimization question
Thanks very much for the reply... This is actually very helpful!Not an answer but some thoughts:
I had not thought about doing exhaustive one at a time! That is great help thanks.
I had thought about the Genetic recommendation but sometimes it just feels like I am getting too hands on and needed to be more mechanical. Nice to know that others are trying that approach!