Bots and Simulations > Evolution and Internet Sharing Sims

Zerobot evolution prize for conditional logic

<< < (6/17) > >>

Numsgil:
Yeah, I definitely think providing any sort of structure is cheating.

googlyeyesultra:
It's tempting to just use the program to generate a bunch of randomize dna's, and see if any of them are remotely close to conditional logic. Then just evolve those. Still, that would probably be considered cheating as well...

I'm a lot better at finding ways around the rules than actually doing the challenge.

cliftut:
I don't think that would be cheating... if you can come up with conditional logic in less than 50 years by randomizing stats, go for it! ^^; Although I don't think you would be able to tell if they were "close" to conditional logic, because either they have it, or they don't.

Once I get used to DB and understand some of this better, I might give this a shot. I have a slight bit of experience with a game that evolved conditional logic using neural networks, and I think I could use a similar approach to "teaching" them in DB as I did in the other game. I'm not sure if the other program had the conditional logic conveniently coded into the learning mechanism, though... probably did.

Anyway, this seems interesting, so I might attempt this after I've gotten acquainted with the program.

I think the best way to do it would be to pick what sort of conditional logic you want, then gradually work up to a situation in which they have to use it or die! Bwahaha!

Is manual bot-killing allowed? I'm not sure if it would help much, but I just want to know.

googlyeyesultra:
I think it is, ya. I'm taking a semi-serious shot at this. Also, one alternative to help "teach" the bots to do a couple of things while giving them energy: set some costs (like stores, movement, whatever) to a negative value (so the bot receives energy for doing them) and then put some other type of cost on (age, dna, whatever). This'll help evolve them towards including some .up and store commands in their dna.

Trafalgar:
So far the most evolved bots I've gotten fire constantly while moving forwards. This is not so good for evolution, however, as they kill each other off rather quickly. Plus I haven't evolved reproduction yet either.

I have tried another approach - starting with a massive genome of all 0s (15k dna length), with shepherd bots (who replace algae and do gain energy as veggies) who go around checking bots and rewarding them with energy based on their DNA length, among other things. Unfortunately, there are a few problems:

1. The bots are not spending enough energy since they're by and large doing nothing, so they're sitting at 29-30k energy all the time.
2. If I have constant-shooting bots in the sim, the shepherds wander into their line of fire and get killed (despite having maxed out shell and a large body). There's no time to react to the shot and flee because when they walk into the stream they get hit by 10+ shots at once. Shepherd death and replacement results in energy rewards not being given due to how I had to implement the only-give-energy-every-so-often system to get around robage and refage being useless after 30k cycles.

Hmm, perhaps having the shepherds trigger mrepro on the other bots if they're above 5k energy would be useful.

Navigation

[0] Message Index

[#] Next page

[*] Previous page