Bots and Simulations > Evolution and Internet Sharing Sims
Mutation Sims
EricL:
FYI, I'm working on local multi-instance sharing as we speak. That will be a highly parallel way to lever multiple CPUs.
Zinc Avenger:
I've been trying to run decent evo sims for a while now, my major breakthrough came when I completely disabled Point mutations. This allows each mutation to be evaluated for the entire life of the bot, so a bot that is "improved" by a mutation is not going to get a negative Point mutation to kill it off before it has a chance to pass on the goodness.
I'd also recommend setting an oscillating Mutation Rate, with 1 cycle at 16x and then 5 (or so) at 1/16x. This way there is a one in six chance of a high-mutation reproduction and a five in six chance of a low-mutation reproduction. This allows the dna to do some mutating, but if a bot is successful then five sixths of its offspring will be relatively unmutated. I find this strikes a nice balance between getting serious mutations and letting natural selection reward the fit.
Testlund:
That works if you run a sim with any of the default bots. I usually have point mutation set to 10 000 000 when running with those, but in my evosim where the bots started with only 13 zeros in their dna I had to set point mutation to 1 000 000 for anything to happen at all. This gets changed by delta mutation over time though. Zerobot sims are more challenging I think. I whould recommend giving it a try. But one needs to have patience because it can take weeks before a functioning bot evolves. I'm at 13 million cycles but only got 39 offsprings. It's running 24/7 on my old computer.
EricL:
I wonder whether one can conclude from this thread that 4 billion years ago, external, radiation induced mutations were critical to evolving the first replicators from the amino acid soup, but then once the basic mechanisms for evolution are in place - reproduction, etc. that external point mutations become the least interesting type of mutation and can even be detremental....
BTW, the posts above w.r.t. scaling on multi-cpu machines no longer apply. Teleporters in 2.42.8 allows for multiple DB instances to cooperate on a single OS image, allowing as linear scaling on multi-cpu machines as the underlying OS allows. Way cool.
Zinc Avenger:
Interesting theory, I've always seen Point mutations as harmful, but I'd never considered them as the original driving force in starting the whole thing off. It certainly sounds plausible! I guess Point mutations would become even more harmful once you start getting multibots/multicellular organisms, nobody wants their left kidney to suddenly start moving around on its own or start to produce venom
On a similar note, I've been trying to "vaccinate" a bunch of not-top-rate bots like C. Ancestralis against some of the best by creating a weaker strain of predators (remove .repro and add randomised conditions to a few of the more important genes) and defining them as veggies with a very low pop cap and very very low rate of energy input. It requires a lot of supervision (killing off the predators if they get a little too successful, adding in the occasional batch of non-veg zerobots with high initial energy when the sim is getting a bit low on energy and occasionally repopulating the predators), and although I haven't had any dramatic successes they're starting to move in the right direction.
On your recommendation Testlund I shall prepare a zerobot evo sim. Does anyone have any recommendations for a good setup for this? I'd hate to evolve something useful only to have the entire thing go extinct on me minutes later!
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version