Advice to a Swarm Beginner...

[Darren Schreiber]

A Swarm newbie recently wrote me asking for advice on getting 
started.  I figured that this may be of general interest and perhaps 
could encourage others to share their collective wisdom on embarking 
on Swarm modelling.  I appreciate any feedback or 
insights--especially as I hope to develop a Swarm course in the not 
too distant future (assuming dissertation, Ph.D, and gainful 
employment.)

        Darren




-------------------------------------------------

Welcome to the world of agent-based modeling! And, thanks for the 
complements.  I will be glad to share what I have learned since I 
know how much I benefited from the work and aid of others.

There are two categories of advice I would start with:

1) Modeling

2) Programming

Modeling

	Some of the modeling issues are generic to social science. 
But, in many ways they become more important agent-based models. 
This is a new method and if you want it to sell, you have to do a 
much better job than I think would be required if you were running 
regressions or just extending some well known results in game theory. 
The other reason these considerations are important is that 
agent-based modeling can be so powerful that the modeler can be lure 
into visions of granduer that fail the principles of parsimony.
	Most important principle of modeling -->  Keep It Simple 
Stupid.  The famous KISS principle means that you should kiss any 
complications good bye until you have really beaten your initial 
basic model to death.  I think it is great to have a vision of what 
you want to prove with your model on the 10 year time horizon. 
Honestly, I actually have a Grand Unified Theory of Politics that 
underlies and unifies my work in political cognition and mass party 
formation.  But, if I am going to get to that point it will be with 
very discrete and unambitious little steps.
	A great idea is to take well-developed principles from your 
field of interest and try to prove the obvious and well established 
results.  I think my work has accomplished this goal.  Now, it isn't 
ground breaking so far because I have only told audiences what they 
already knew.  But, I have gotten there a different way and 
agent-based modeling allows me a lot of flexibility to go beyond my 
first destination point.
	The other thing that "classical" approach will do is to 
reduce the fear and anxiety of your audience.  You are bringing a new 
method that they know may be a new hurdle or threat to their 
scholarship (all new paradigms suffer this).  And, computational 
models have a very "black box" feel to them.  Be transparent about 
borrowing existing theory, be transparent about simple and intuitive 
models.  I believe that once you can do a good job with the classics, 
you will build your own skills and confidence enough to get more and 
more ambitious.

Programming
	The KISS principle is a boon to a programmer as well.  On my 
second modelling project, I wrote a very detailed project development 
plan with a timeline and small increments between steps that would 
allow me to check my progress and have lots of attainable goals. 
This meant that within a few days of my first go, I had something 
that worked.  It was trivial and uninteresting and only printed a 
little text and had absolutely no substantive interest.  But, it had 
the virtue of running and being easy to debug when it wasn't.
	I got that first bit of code up and all the other incremental 
improvements by borrowing liberally from more experienced 
programmers.  I tried to reduce the amount of original code so that I 
could minimize my work and maximize productivity.  I downloaded every 
single piece of SWARM code I could find, read them, and cannibalized 
them.  I also made liberal use of the SWARM support listserve.

Anyway, I think those are my two most useful pieces of advice I would 
give to a beginner.  I am going to post this advice on the Swarm 
modeling listserve and I am sure that others will chime in with their 
insights.  Feel free to email again.


        Darren


--

_____________________________________________

                 Darren Schreiber
                  Attorney at Law
                 Graduate Student
             Political Science, UCLA
                address@hidden
        http://www.bol.ucla.edu/~dschreib


                 ==================================
  Swarm-Modelling is for discussion of Simulation and Modelling techniques
  esp. using Swarm.  For list administration needs (esp. [un]subscribing),
  please send a message to <address@hidden> with "help" in the
  body of the message.
                 ==================================