Galapagos Gene Pool: Force unique values?

Hey all, I'm trying to use the Gene Pool to generate a series of integers which will act as index values (to pull items from a list). However, each Gene slider acts independently and consequently many duplicate values are produced from the Pool.

Is there a way to force all sliders to turn out an unused value?

Or, is there another way I can achieve this concept and be able to use it for Galapagos solving?

Thanks!

Replies to This Discussion

Permalink Reply by Hannes Löschke on February 25, 2014 at 5:05pm

You can do that with just one slider (gene)

- create a series of integers as long as your list

- jitter the list with 100% strength (make seed input a gene)

- shorten the list to the first n elements you want.

Permalink Reply by Mark Pothier on February 25, 2014 at 6:40pm

Hi Hannes, thanks for the reply. Here are my concerns:

If Galapagos is only controlling the seed slider of the Jitter component, I would think that it would lose the ability to understand how its specifically optimizing the fitness value, since ultimately every combination is just another random shuffle. In order for it to search for the best combination of indices, I think it would have to have individual control over each value drawn. Does this make sense?

Permalink Reply by Hannes Löschke on February 26, 2014 at 4:10am

You are right, jitter will just randomly scramble the list. I'm not entirely sure how Galapagos would handle such cases but I was thinking that at least the best random cases would be kept and others generated and tested.

To introduce a second parameter that determines the start position of the sublist. So for the same seed value, Galapagos could test similar lists with smaller variation.

Finally you could just add up a few genes with MassAddition. Use the partial sum output as an increasing series with variable spacing. Then you could remap that series to fit the range of indices of your list to pick from.

Permalink Reply by Daniel Kolling Andersen on May 27, 2015 at 7:08am

Hi Mark

Did you ever come up with something for this problem?

Permalink Reply by Mark Pothier on May 27, 2015 at 6:38pm

Hi Daniel,

Unfortunately no...I believe I just ended up accepting a small percentage of duplicate values for the purposes of what I was doing. I don't think that was ideal though for evolutionary solving....

Permalink Reply by Daniel Kolling Andersen on May 28, 2015 at 3:33am

Alright, thanks anyway.

Permalink Reply by Oak Island on June 8, 2015 at 3:51pm

Hi. A little late but I found something that may be what you need: an optimization tool to genomes with elements without repetition GAvilan http://narcea.es

This tool is free for 15 days, this time can be enough to solve your problema
I used the tool and it seems to work pretty well, do not know if adapted to your needs

Also has an example to do that, but slow, with only galapagos.

Permalink Reply by Camelia on June 29, 2017 at 3:22pm

Hi,I have this problem (what Mark said) and this link(http://narcea.es ) does not work!!could anyone help me?

Permalink Reply by Ewan Branda on February 10, 2019 at 11:10am

A rather late reply to an old thread, but it may be helpful to some. You can do this using a Random Key variant of the genetic algorithm. This is used in job scheduling problems (see https://bit.csc.lsu.edu/~jianhua/random-key.pdf and https://campus.exactas.uba.ar/pluginfile.php/92564/course/section/1...) Don't use index values as your genome. Instead, the genome should be a list of random floats between 0 and 1. To generate an individual (phenotype) simply sort your list using the genes as sort keys (and select a sublist if desired). This guarantees uniqueness. Bean (1994) explains it very clearly.

RSS