Re: Evolution by Computer: 1000aa in 1163 generations.

On Fri, 12 Dec 2008 19:28:36 -0800, seanpitnospam wrote:

<snip>

Not even close to the real thing. Congratulations - - you've just
reproduced Dawkins' "Methinks it is like a weasel" algorithm all over
again.

That's what you asked for. "Highly specified", remember?

1)  Real evolution doesn't work towards a specific goal, and this does.

That's a really big problem.

For you, yes.

You used a template-matching algorithm
where every single character match was defined as selectable in your
algorithm.

That's correct -- because you didn't specify a fitness function, I chose
one of my own.

What's your fitness function, Sean?

The problem with this notion is that selection is not based
on an improved function, but on an improved character match to a
pre-existing sequence.

That's an improved function in your toy model of evolution, unless you'd
like to specify a more realistic one.

Of course, such template-matching evolution
cannot be used by biosystems to find novel functionally beneficial
targets sequences with unknown locations in sequence space.

Correct! Nor does evolution "find novel functionally beneficial target[]
sequences". It finds sequences that improve overall fitness. *Your*
model of evolution fails to take this into account.


For
example, how on Earth would a bacterium use such an algorithm to evolve
a flagellar motility system?

Wow, you really just don't get it. Evolution doesn't solve motility
problems. It changes fitness. If the result of that is motility,
great. If not, that's fine, too.

As soon as you introduce "specified functionality", you're off dealing
with a toy model of evolution -- it still works, of course, as "methinks
it is like a weasel" shows. The interesting bits happen with nontrivial
fitness functions, which I don't think you're going to be able to provide.


In short, this algorithm of yours doesn't remotely resemble the
mechanism of RM/NS. You've only got half the problem going here - the
RM part. You haven't remotely modeled the NS part of the equation.

Sure I did -- that's in the "Reproduction" section.

<snip>

4)  This is a tiny population size, but then, 1000aa isn't that big of
a
deal.

That's because you didn't model functional targets in 1000aa sequence
space. Your model is a template-matching model - not remotely
realistic.

Oh, it's not remotely realistic, but that's because it's evolving toward
a specified target. The sequence space was quite accurate -- a 1000aa
sequence of 100 possible genes each.

<snip>

You've produce Dawkins' template-matching "simulation" which in no way
reflects the actual mechanism of RM/NS when it comes to finding
functionally beneficial targets in sequence space. That's your
fundamental error here. Try again . . . and better luck next time.

Hey, it's your model, not mine. If you can tell me how to do "highly
specified" without specifying anything, I'd love to hear about it.

Or you could code up your own model....

.

Relevant Pages

  • Re: What do you know that the rest of the world doesnt?
    ... assumption - that sequence similarities necessitate common decent via ... In the simplest sequence evolution ... Designer didn't specify at design-time (though, perforce, He had to ... those functions that have a greater minimum threshold requirement. ...
    (talk.origins)
  • Re: Cohens paper on byte order
    ... the external data model to a bit stream, ... comparable standards do specify external data ... sequence, and which therefore does not provide guidance ... standard character codes). ...
    (sci.crypt)
  • Re: What do you know that the rest of the world doesnt?
    ... assumption - that sequence similarities necessitate common decent via ... deliberate design. ... In the simplest sequence evolution ... Designer didn't specify at design-time (though, perforce, He had to ...
    (talk.origins)
  • Re: Public disclosure of discovered vulnerabilities
    ... > a FEW examples that have had hundreds of postings, with no consensus ... > Sequence points, including how a tree is mapped to a linear model, ... > and extending C for asynchronicity. ... and bad ways to specify and implement these. ...
    (sci.crypt)
  • Re: Rand() with base
    ... >> same sequence of PRNs? ... But it doesn't specify which seed has to generate ... No, the Standard does not ... to specifying the 'rand' algorithm, which the Standard doesn't do (and we ...
    (comp.lang.c)