I'd start with the simpler solution first: try putting all three trees in
an individual, and evolve that. If things go well, fine. If not, we can
talk about something more complicated.
On Fri, 4 May 2007, Bill Drozd wrote:
> I am intersted in evolving solutions to a agent based control problem where
> the solution consists of 3 GP trees. It seems that it would be possible for
> me to either approach this from a coevolutionary perspective and maintain 3
> populations of single-tree individuals, or have a single population where
> each individual consists of 3 GP trees. (All trees use the same set of
> parameters/function operators).
> What would be the difference between using these two approaches? It seems to
> me that given each of the 3 functions have a very different purpose,
> coevolution would be a more techinically accurate path of evolution.
> However, it also seems to be more computationally intensive.
> Any guidance would be greatly appreciated.