General > Off Topic

Finaly 3.0 is somewhat out (and stuff)

<< < (2/13) > >>

Light:
I think trying to create a sort of map is impractical and wouldn't work in reality, but some sort of group bot behaviour would be an interesting area to investigate not sure how much of an impact it would have on improving bots in the league though.

Numsgil:
practicality aside, a group hunting behavior would allow smaller bots to take down larger bots.

Could help shore up the sudden increase in heavy bodied top bots.

Botsareus:
Ok I finsihed my experiment trying to evlove bots using 2.35db , I pinned a nonmutating bot agenst a mutating virsion of itself:

cond
  *.nrg 20000 >
start
  50 .repro store
stop

cond
start
 -1 .shoot store
  10 .up store
stop
end

^ that is fomiliar to you guys I post that BEEEEP* a lot. , the above was used as starting Dna.

Results did not make sense:

If bouth robots were givin the same starting values: the one that did not mutate won

If  robots with mutation was givin high papulation in the Beginning: It won but the only thing different was the mutation rates.

Conclusion:

The Dna system and mutation is still too limited to produce working and  unique code.

Based on that ^ now I will try robots that have many many generations on one repro, the idea is that enough changes in the dna will acumilate in one go , to produce some kind of new better robots.

Numsgil:
There are some flaws in your experiment:

1. Time.  90% (99%?  Maybe 99.99%) of mutations are BAD.  Or at best have no effect.  So it will take alot of time for a mutation sim to produce something interesting.

Try letting it run for 3 days straight.

2.  Conditions - You are trying to judge how well a bot evolved based on how well it fairs against another bot in F1 starting conditions.  But the bot didn't evolve to win under F1 conditions, it evolved to survive in its current settings.

After 30 000 cycles the energy levels are going to be different than they are at the start of a standard F1 match.

Basically, you'll end up with a bot that is designed to survive in F1 conditions after 30 000 cycles.  So if it doesn't do as well as an unmutated version in F1 conditions at 0 cycles, don't be so surprised.

Try adding some unmated version to the mutation sim.  See which one fairs better on the mutated one's home turf.

Botsareus:
"Try letting it run for 3 days straight." ROFL , no really man , I did simulators before; my simulators produce resolts a lot faster... , I am working on a sim right now (not really - I am fead up with vb right now) anyway, when I get that sim finished I'll give you the source code. (unless it does not work witch I dought)

Navigation

[0] Message Index

[#] Next page

[*] Previous page