Re: [Swarm-Support] Taking screen shots and conducting experiments

[Darren Schreiber]

I'll take a crack at the "designing experiments" question since it is 
one of the things I am spending most of my ABM related time thinking 
about.

The basic way to evaluate a model that you have designed in Swarm (or 
another agent-based model toolkit) is to run it many times, varying the 
parameter settings and noting the changes in the dependent variables of 
interest.  There are a wide variety of ways to do this with software at 
this point, but one of my aspirations will be to develop a better set 
of standard tools people can use to evaluate their models.  The 
challenge in evaluating ABM's is that they often have non-linear 
properties.  Thus, studying the relationship between one independent 
and one dependent variable maybe insufficient to understand the 
behavior of the model.  This points to the problem of robustness.  If I 
find a relationship between X1 and Y1, does it still hold true if I 
vary X2 somewhat?  Or, is that relationship between X1 and Y1 just one 
of many possible relationships that are highly sensitive to the initial 
conditions or the setting of other parameters.

In my view, learning from ABM's is a lot more difficult than writing 
them.

        Darren




On Nov 8, 2003, at 1:00 PM, Steve Railsback wrote:

> > How could I conduct experiments with swarm....are there some web 
> > pages that expand on this....
>
> Not sure what you mean...how to design experiments? Software to 
> automatically run scenarios and replicates?
>
> Steve R.
>
>
> _______________________________________________
> Support mailing list
> address@hidden
> http://www.swarm.org/mailman/listinfo/support