ABM toolkits

[Marcus G. Daniels]

Much to my despair, I don't have much time anymore to to play with
nifty software.  However, today, I'm stuck waiting on the CD-ROM
burner.  So rather than spend the entire blasted day doing annoying
SDG administrative tasks, I decided to download the latest and
greatest offerings from Brookings and the University of Chicago.

I'll throw up my a few observations, and maybe folks in
the community who have more experience can throw in their observations
or some factual information.  I get the feeling there isn't a clear
picture of who is good at what and why they are doing it.

Ascape

The first thing that jumped out at me were the packaging 
improvements.  It's point and click all the way.  No futzing with
batch files, CLASSPATH, and all that.  There are a bunch of demos
related to agent-based modeling research at Brookings; in particular,
the Sugarscape stuff seems to be organized to relate to their publications.

>From the API, Ascape seems to me to have put of energy into making it
easy for agents to interact, especially on cellular automata spaces.
Common needs like agent's subjective awareness (what's around me) are
inherited, which ought to make agent's smaller, i.e. more robust and easier
to code.

Considering it is Java, graphics in Ascape seem, for lack of a better
word, `tight'.  Press a button something happens right away, reorder
windows the refresh occurs right away.  No gunk left on the screen, or that 
sort of thing.

Repast

I also downloaded the new release of Repast (1.1). 
The first thing that jumped out at me was the nicer looking control panel.
Actually, it looks just like Ascape!

Repast doesn't quite seem to be as responsive to me as Ascape.  Of
course, that's pretty much a useless remark.  But, hey, so long as
we're cloning each others' GUIs, I guess it is in some sense it is
appropriate.  So there!  :-)

Repast has a more general scheduling engine than Ascape.  It's a lot
like Swarm's, but uses some different idioms.  And of course, Repast
and Ascape are `pure java', which in principle means code portability
is less of a problem, i.e. it is easier to make demos on the web,
migrate code, and that sort of thing.

Finally, Repast has a nice multilayer raster display, where you can
turn different layers on and off.  (In heatbugs, for example, you can
see the bugs, the diffusion space, or both.)

My summary

Ascape has strong features for doing lockstep/temporally-flat, CA-like
simulations.  It includes several fairly interesting/complex models.

Repast is a respectable Java-based simulator implementation with
interfaces highly similiar to many in Swarm.

Swarm's major downside (in my opinion, which is not necessarily the
SDG opinion), is that we don't use garbage collection.  Swarm has a
well-tested multi-level discrete event scheduler, and relatively good
software integration features (e.g. HDF5, R).  Swarm also does not
depend on any proprietary software, even in the case of Java.

License-wise:

  Ascape is freely usable (not sure about redistributable)

  Repast has an `advertising clause' type license

  Swarm is GPLed, but we'll provide alternative licenses to those that 
     need more latitude.

                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.