Re: GA & GP

[glen e. p. ropella]

> I guess another difference is that GP does not use mutation, AFAIK.

In the implementations I've seen, it does, but very minimally.
The main workhorse is crossover.  And it is hard to imagine
what "mutation" should mean in a GP environment.  Given
an S-expression, do you mutate by chopping off a branch?
Cut two branches off and swap them?  Randomly generate a
new branch?  Or, maybe, swap terminals with other ones chosen
from your terminal set?

While the subject is up... Has anyone thought about doing a 
GP library for Swarm?  I want to; but, I have no idea when 
I'll get time?  What might it consist of?  I would assume
we could have methods defined in the agents that called
evolved C functions.  Or we could use the createActionCall
passing a pointer to an evolved function.  The problem lies
in compiling those functions once they're ready to run.

Or, it might involve a minimal lisp or s-expression interpreter...

glen