Darwinbots Forum
Bots and Simulations => Evolution and Internet Sharing Sims => Topic started by: EricL on July 04, 2006, 01:44:39 AM
-
One way to think of the dynamic cost multiplier is as a measure of organisim efficiency in a sim. If I set the dynamic heterotroph populatiuon target to some number, say 500 and the sim initally stablizes at a cost multiplier of around 1 but after say, a few million cycles the cost multipler has crept up to say 2, then one way to look at this is that the 500 organisms that have evolved over that time are twice as effecient as their ancestors in that they form a stable population that can survive and reproduce in an environment twice as costly. Assuming sufficient nrg resoruces in the sim, if one were to set the costs back down to the inital stablization point were their ancestors population stablized, there would presumedly be a population explosion of the more recent, more effecient organisms.
I'm interested in hearing from anyone who has wtinessed this in an evo sim and what charactistics evolved. Did generation time decrease? Did the organisms evolve some behavioural trait which assists in aquiring nrg?
Thanks.
-
just some random thoughts ...
a stream of consciousness thing ...
I thank you for your indulgence in advance.
humor me, eh?
One way to think of the dynamic cost multiplier is as a measure of organisim efficiency in a sim.
yes.
and in order to do any of this 'looking' ...
this examination via empirical results ...
one must first be able to minimize the effects of multiple variables ...
of 'outside' influences ...
to ensure that what one is seeing is indeed the effect of
whatever variable one is 'studying' and/or manipulating.
I think we are closing in on being able to do that.
this is the 'lab' part of the DB experiment, eh?
being able to first create/develop a 'stable' environment
that is at equilibrium ...
then introducing a change ...
say of an additional organism or bot ...
then observing how the efficiency is altered over time ...
where the new equilibrium settles ...
allows us a window into the dynamics involved in the
system as a whole.
of course this is for our own benefit ...
our own gathering of knowledge ...
which hopefully will then enable us to better set up darwinbots
as a viable environment ... for being a real evolution sim ...
which we can then 'let go of' to do it's own thing ...
to not control ... hands off ...
to work at evolving species and/or an 'organisim' on it's own ...
'within' the initial paramerters we must 'impose' at the start ...
so that 'we' can still make sense of the results ...
so that doesn't appear to be chaos to 'us' ...
so we can recognize the 'order' that is there.
so this ability to experiment and tweak/fine tune ...
is simply preparation ...
building/creating a viable system/program ...
which can respond without either stagnating and dying out ...
or cascading out of control.
it is like Heisenberg said, eh? ...
we can't help but influence the results in some way.
let the experimenting continue.
-
I want to get there too eventually, but it requires a bugfree program. I don't know how many times I've started over with a new sim of zerobots and after awhile something mess it up. So here I go again. I'm gonna start a new evosim right now and I was thinking about starting with mazes and bozes... (I mean boxes) ..that drifts around. Hopefully some interesting bots will evolve this time.
-
I want to get there too eventually, but it requires a bugfree program.
I'm trying my best bud. Crashing or other bugs which can impact long run sims are always my top priority.
Things are improving. At leat the bugs are in new areas, primarily because sims could never get to those areas before. The corrupted save sim issue a few versions back for example was never hit before because no one ever tried to re-load a sim with that many mutations before. Similarly, I bet if you exceed 32000 shots in the current version, it will crash. No one has ever had a sim with that many shots before. We are blazing new ground.
-
Well, I'm running with mazes and boxes on the screen and I have them absorb the shots. So it will run faster and there wont be so many shots. A nice feature. I'm getting close to 460000 cycles and everything has worked smoothly so far. Knock on wood.
The shapes gui could use a little fixing though, so it works better to make changes on the fly. Settings doesn't seem to stay put, but..Uh, that's another topic.
-
Bump.
Just bumping the topic. Still interested in hearing from anyone whose has seen CostX rise over time in an evo sim.
-
I don't know where to find this 'dynamic cost multiplier' but in answer to the question:
"Did the organisms evolve some behavioural trait which assists in aquiring nrg?"
A couple of examples spring to mind.
1. Some of my bots when I first started where improved in Evo sims, by spotting an improved behaviour or adaptation that appeared and changing the original bot.
2. Not a 100% but pretty sure that the 'sever birth tie' was first seen in a evolution sim, again quickly gaining control and increasing the overall population. I'd have to go way back and check the old forum.
Other mutations, such as reducing the .repro lvl or discarding a gene you thought you needed but didn't, jump out as positive evo mutations I've seen before.
The reason I don't use Evo sims now to improve bots is that the starting bot is usually much better designed than it used to be. To see a major advance in a bots design during an evolution sim you do need the bot to have faults it can improve on after all.
Evolution is enviroment driven if I remember correctly, perhaps designing a bot for one end of the scale and then releasing it in an enviroment at the other end of the scale would show better mutations.
-
I don't know where to find this 'dynamic cost multiplier'
bear in mind that all of Eric's work has been on the development of 2.4 since Numsgil moved on to working on the C++ version.
The dynamic cost multiplier doesn't exist in 2.37.6