Xixax

Reminds me of an application I did last year. Wanted to watch and guage how long it would take various lifeforms to "randomly" evolve against each other into different ecological niches.

I set up things like metabolism ( very simplistic, rate at which they use energy, forcing them to find sources for it ), predatory capability, defensive capability, mobility on land, in the air and water etc. etc. etc.

Started with organisms that fed entirely on "natural" resources, things I fed into the system. After many thousands of generations I had some, what I thought, were incredible but "expected" results.

1) As more and more organisms multiplied and strained the natural resources, this led to greater selection pressure for mutations that allowed them to feed off each other instead, or to move to a new ecological niche ( had nothing on land or in the air to start )
2) As this process of feeding off each other exploded, so did variations in whether organisms were predatory or defensive in nature. Those that fed off of natural resources still were more likely to be defensive while those that fed off each other evolved predatory features.
3) The number of offspring something has was constrained by available resources. When they had too many offspring it hurt the ability for selection to find novel mutations because death was too "random", making these organisms evolution progress at a slower rate than their predators and competitors. Their breeding numbers were directly inverse, for the most part, with their energy needs and size.
4) I had decided at the end to do a trial run with features specific to sexual selection that had a negative effect on their general ability to move and found that this also regulated itself well. There were a few that had a great emphasis on this in spite of it's costs, but they were always in ecological niches where it was difficult for predators to move ( I had land divided into desert/forest/plains/high-grass plains/tundra/ice&snow/mountains ). I know that technically it's possible for a feature to be useful for survival and sexual selection at the same time, but I'm pleased with the results anyway.
5) Longevity was also self regulated by selection pressure. Mutations that lengthened it's life didn't necessarily help the entire population because it put too high a strain on resources cutting off available food for organisms capable of reproducing.
6) There were ( I stopped the automatic process and watched a few hundred generations of different species go by ) hundreds of "beneficial" mutations that never made it because of random luck, but enough still made it through to create a great deal of genetic diversity and cause life to spread across every ecological niche in the program.

My personal favorite event from the program was when awareness got to a certain level for land animals it was obviously the primary selective pressure for organisms to take to the air so that they could attack with little warning.

To be fair, all organisms were already reproducing sexually. I had a system that determined recessive and dominant genes in place, and tried to accomodate three main types of mutations: Point mutations, duplications, deletions. The object responsible for copying the gene and then meshing them together had variables that could be changed to alter the mutation rate, but of course there was a nice tight area where diversity was just great enough to provide change, but not so high that deleterious mutations killed off organisms.

If anyone is interested I can dig it up, but it only produces printouts, I never got to the stage of creating images based off the information ( I had toyed around with the idea of making a visual representation of it's abilities, but grew busy before getting to that point ).

Undercurrent

I wasn't suggesting actually replacing the GA with a hill climber, rather changing the fitness function to one that shows off the GA to better advantage.

Then you might want to look at some of the challenge problems at http://www.satlib.org/ to use as landscapes. You can fairly easily scale SAT problems from the really simple kind you're already solving to ones that will kill any GA. The population dynamics on some of them will definitely be interesting.

Amen-Moses

Just keep your evolving bit-strings well away from my air traffic control systems please, we have enough variables to worry about with pilots and controllers as it is.

Amen-Moses

NonHomogenized

That sounds really interesting. If you're willing to dig it up, I'd certainly be interested. It would mean I'd need to get a new printer, though. The old one I have sitting around sees little use... for a good reason. And I don't think I want to use my "free pages printed per semester" at the university for that. But, yeah, please, if you can find it, I'd love to have it... when I get my computer back up and running (my computer fried itself recently, so I'm waiting for RMAs and new components to arrive...) in a couple weeks.

Nic Tamzek

Say Rufus,

What would happen if you made offspring dispersal distance an evolvable variable? Say, in stable environments vs. commonly disturbed environments?

emotional

It did convince one erstwhile creationist here - me.



I didn't say "out of nothing", I said "out of no design", by which I meant that the organisms wouldn't be designed like watches by a watchmaker. The only design that takes place for evolution to occur is front-loaded design, and I believe that kind of design was made 15 billion years ago (yeah, you guessed it, I'm a theistic evolutionist).



But it's not watchmaking. See above.



Those are all unfair demands! As if they didn't know some front-loaded design (the materials and their properties) is necessary for evolution to take place.



I wonder what would happen if someone asked them to simulate Genesis creation. They'd probably simulate it by building a complex machine. But that would be a foul! The Biblegod didn't use any pre-existing matter to create. So here's where evolution has the edge: it can be simulated faithfully, while Genesis creation cannot.

CX

Any chance I could take a peek? I'd be fascinated to look at it.

Morat

What did you use for a random number generator? I've played with GA's a little, enough to seriously consider it as a topic for my Master's, and was looking for a better random number generator, as the standard C ones certainly aren't good enough.

I was looking at a few generally used for cryptography, thinking that if they were random enough to generate codes....

RufusAtticus

I'm not ready to release the code of the project; however, I will put up some of the important sections latter tonight.

RufusAtticus

Mersenne Twister