Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Evolving Warriors

30 views
Skip to first unread message

Will Varfar

unread,
Sep 25, 2002, 6:34:35 AM9/25/02
to
Question about corewars etiquette really;

I am beginning to evolve warriors with my own twisted algorithms.

Is it 'ok' to cross-pollenate my genepool with real live other-peoples
warriors? I plan to routinely introduce real live warriors from the
archives and rather than just use them to benchmark 'fitness', allow
them to procreate in the same way that my evolved warriors do, even
procreating with my evolved warriors.

Will this piss the anonymous doners of these genes off? Would I have
to maintain a geneology?

Best regards,
Will.

HiEv

unread,
Sep 25, 2002, 10:04:25 AM9/25/02
to

It would surprise me if it pissed anyone off. As long as the end
product wasn't basically a one-off clone of the "gene donor" then it
shouldn't be a big deal.

However, you'll probably find that if you do this you'll end up with a
lot of "mules." (If you cross a horse and a donkey you'll get a mule,
which is usually sterile.) When you interbreed two dissimilar "species"
of corewarriors the result will almost always be less fit than the
general population unless it is nearly identical to one of the parents
(and even then it's iffy.) This is because the parts are almost always
too dissimilar to work together properly in the offspring.

A better strategy might be to start off with several "species" that only
breed amongst themselves but fight all the species. So while the
members of Group A will fight against members of Groups A though J plus
some other unchanging benchmark warriors, they will only breed with
other members of Group A. Using speciation this way has been shown to
help increase the pressure for innovation.

However you could seed each "species" with slight random mutations of
one existing corewarrior and then let the genetic algorithms take it
from there. This would also make it easier to report who the "Eve" was
for each of your groups if you wanted to.

These are just a few suggestions, and I could be wrong, so please let us
know how it turns out! (Good or bad.)

--
The difference between intelligence and stupidity is that intelligence
has its limits.

Dave Hillis

unread,
Sep 25, 2002, 1:19:51 PM9/25/02
to
howard...@hotmail.com (Will Varfar) wrote in message news:<5e314d3b.02092...@posting.google.com>...

> Question about corewars etiquette really;
>
> I am beginning to evolve warriors with my own twisted algorithms.
GOOD!! The more of us the better.

> Is it 'ok' to cross-pollenate my genepool with
snip

> Will this piss the anonymous doners of these genes off? Would I have
> to maintain a geneology?

Please don't put them on the evolved hill at Source Forge. Other
than that, IMO no one will mind. (You can use the "test" option to see
if they would have made it onto the evolved hill.)
I'll make a couple of suggestions. 1) Don't bother keeping track
of geneology, but do keep carefull track of which handwritten warriors
you add. After several generations it will probably be depressingly
easy to determine paternity because you'll have a population of
slighlty mutated clones of one or two of them, possibly stripped of
their decoy and boot. If you put all the published warrior's object
code in one file it makes it easier to use text-search to match lines
from the evolved warrior. And without checking you run the risk of
anouncing that you have "evolved" Porch Swing.
2) I've spent more time than I'd like to admit dropping very
strong warriors into a population and hoping that would squeeze out
the last 2% performance from them. What I haven't done but should have
is take some fairly weak published warriors with some real room for
improvement and see if anything good happens. If I can't improve a bad
warrior, I'm not going to improve a good one.
Using a GA just to improve a specific part of a warrior is a
different story.
Dave Hillis

Will Varfar

unread,
Sep 25, 2002, 6:08:44 PM9/25/02
to
> do keep carefull track of which handwritten warriors
> you add. After several generations it will probably be depressingly
> easy to determine paternity because you'll have a population of
> slighlty mutated clones of one or two of them, possibly stripped of
> their decoy and boot.

Dave, your concerns are very real. I do have some simple
anti-stillborn code in mind, to cull really bad children really
quickly, but even with my computers working flat out I am going to be
searching a haystack, and risk finding only 'real' warriors.. uggh

I am planning to work at a meta level rather than a
instruction-by-instruction level, so my evolved warriors are going to
look a lot more like their parents than otherwise anyway

0 new messages