Print

Print


On Nov 30, 2005, at 12:33 PM, Alberto Cuesta wrote:

> Now, to the meat: I work for the ITI (http://www.iti.upv.es), a
> research
> institute here in Spain. We are currently working with genetic
> programming using JEO, the java evolutionary computing library
> included
> with the DR-EA-M package (http://sourceforge.net/projects/dr-ea-m).
> I've
> been assigned the task to test both JEO and ECJ, and compare them,
> as we
> are open to change platform if we see it convenient.

I was at PPSN when DREAM was announced, and did a spiel for ECJ there
too, so my opinions on the matter are based on that time period, and
things may have changed for DREAM.

DREAM's basic notion was to build on a homegrown grid computing
system, treating each grid node as an island in a distributed EC
island model.  The EC portion of the code (JEO) was done largely in
spain by J. J. Merelo's group in Grenada, I think, and was supposed
to be a port or rewrite of the EO C++ toolkit.  At the time, JEO was
immature, and in that respect ECJ was much more featureful.  But
things may have changed: I've not checked in on JEO in a little while.

DREAM's primary nicety (the grid-computing environment) wasn't really
something that my group needed as we typically deployed on beowulf
clusters etc.  Plus it always struck me as strange that the grid
nodes were being used for islands rather than for master-slave
evaluation.  [Unrelated: we have a heavily revised version of the
master-slave evaluator in-house that handles dynamic failures and new
nodes, among other nice things].

We also had some discussion about porting ECJ to sit on DREAM at
PPSN.  It wouldn't be all that hard to do now with the new modular
version of ECJ actually.

Sean