About Mr.David's post; Coalescence

Dear All and Mr. David Rutten(if you are there.)

I have question, about detail of algorithm. In his(your) post "Evolutionally Solver: Coalescence" (http://ieatbugsforbreakfast.wordpress.com/2011/03/05/evolutionary-s...), Galapagos' algorithm consists of 3 techniques; 'crossover coalescence' and 'Blend Coalescence' and 'Preferenced-Blend'. But he didn't mentioned about how to combine these. There is similar question in selection article too, how to combine 3 methods?

I am now trying to write a thesis and code, about this.

If possible, could you teach me the detail of here?

Regards

Ken

Replies to This Discussion

Permalink Reply by David Rutten on September 8, 2013 at 2:06am

Hi Ken,

Galapagos currently applies all methods at random. You cannot change that behaviour. It used to be editable in older versions of the Galapagos UI, but I took it out of more recent versions as it seems to be a pointless user setting.

David Rutten

david@mcneel.com

Tirol, Austria

Permalink Reply by ken ho on September 10, 2013 at 11:17am

Thank your immediate reply.

Regards
Ken

Permalink Reply by ken ho on September 26, 2013 at 1:34am

Dear Mr.David

I hope I am not bothering you. I have one more question.The question is
'When you decide to finish calculation in Galapagos?'

I am now making my own simple GA in processing.
Then,it is so difficult to know which is the very best optimum answer.
There is no proof the last answer is not local optimum.
I don't know how to manage.

Thanks in advance, regards
Ken

Permalink Reply by David Rutten on September 26, 2013 at 2:18am

Hi Ken,

that's the problem with this approach, you don't know you've found the best answer until you've tried all possible permutations. Since that is ridiculous (there's often simply no time to perform a brute force search).

But that's ok, a GA (or any other stochastic algorithm) doesn't promise to find the best answer. Technically the only thing that is guaranteed is that no matter when you stop it, you will always get some answer, and the longer you're willing to wait the better that answer will become.

There are several ways to decide when to stop, Galapagos uses the first 3 of these:

Time limit. The solver is allowed to run for N minutes.
Target value. Perhaps you're not looking for the best answer, perhaps any answer that yields a fitness above 53.9 is good enough.
Stagnation. If you haven't been able to improve on an answer for X generations or N minutes, assume you're done.
Genetic variability. If all the genomes are extremely similar, you've lost the variability you need for further exploration. At this point you can either throw in the towel or start again, making sure you distribute the genomes for the next solver run away from all the genomes of previous runs. That way you're guaranteed to start exploring new areas that may hide undiscovered fitness peaks.

David Rutten

david@mcneel.com

Tirol, Austria

RSS