Re: [Repast-interest] RePast 1.3 released

[Nick C. <ni...@sr...>]

Hi,

On Tue, Apr 24, 2001 at 09:01:46AM -0600, Wim Hordijk wrote:
> > > Nit: I found that I needed to include both the repast.jar and
> > > the colt.jar file in my classpath.  Are others finding this
> > > to be the case?
> > 
> > I can't reproduce this. That is, the following works for me:
> > 
> > java -cp repast/lib/repast.jar:/home/nick/jbproject/classes 
> > uchicago.src.sim.jiggle.JiggleModel  
> 
> I found that I need to include repast.jar and colt.jar in the classpath
> when compiling the source code of my model, but for running the model
> including just the repast.jar is sufficient. Dunno why though.

Oh, I see. I think I misunderstood Owen's original message then.
I suppose "java" reads the manifest, but "javac", "jikes", etc.
don't. The error makes sense then as you're most likely to 
actually refer to a class in colt.jar directly in your model code.
As opposed to classes in the other jars which are referenced via 
some class in repast.jar.

> > > I'm hoping we can use repast as the standard
> > > analysis tool so that we can verify each other's studies.
> > 
> > If this pans out, let me know how the verification works out.
> > There's always talk of reproducing each others models to 
> > duplicate the results, but I'm not sure how much of it
> > actually occurs.
> 
> We've been having discussions about this here at SFI. On the one
> hand it would be nice to have a standard modeling environment so
> we can verify each other's models, but on the other hand, it would
> actually be better if someone's model is verified in *another*
> modeling environment. If the results are still the same, you can
> be pretty sure your model is OK, cos you've tried it on two
> independent platforms. If I reproduce your repast model in repast,
> we still don't know if there's maybe something funny in repast that
> causes certain results. Re-implementing it in, say, Swarm would be a
> better test. So, I'd say it would probably be better to have some kind
> of standard "higher-level" or pseudo agent-based modeling language,
> one that is implementation and platform independent. People can then
> exchange their models in this high-level language, and implement it
> in their favorite modeling environment. It's like giving someone an
> algorithm in pseudo code. It shouldn't matter in which language I
> implement it (C, Fortran, Java,...), it should still work the same.
> If not, I know I made a mistake somewhere, or there is some language
> dependent problem...


I agree completely. In fact a few people are tired of me saying
this -- "what do you mean implement it with two toolkits, I have
enough trouble with one!" As part of an IDE for creating network
type models that I'm currently working on, I've been thinking about
some sort of neutral xml-based model description language. (I need
to look at the IMT stuff here as well). The ide will store models
in this language and then a bytecode or source-code generation
"plug-in" will do the real work. That's the longer-term goal,
anyway. At the moment, its probably tied too closely to repast, but
not so closely that the repast bytecode - code generation part
couldn't be disentangled. The tricky part here is specifying
behavoir. At any rate, this model descripition language is
intended to function as your high-level language mentioned above.

Nick

-- 
Nick Collier
Social Science Research Computing
University of Chicago
http://repast.sourceforge.net