Re: Simulating Individual Behavior

[Jan Kreft]

OK, I agree with Paul and Mike that it can be very tricky to determine the
parameters for individuals. Bacteria as individuals, e. g., are simply so
small that only the latest high-tech methods could give estimates for the
parameters you would have to use in a model (and that costs a lot of
timemoney). Almost always microbiologists measure the population average
and disregard the heterogeneity of the population. 

But I can see another way to solve the problem. If you start with a
conceptual model with all the parameters and mechanics you need and only a
rough idea of actual values (from population average measurements) you can
run the sim and compare the output with the real system to be modeled.
Then you optimize the parameters to achieve the desired output. Perhaps
call it back-calibration. (Is there an official term for it?) 

For that purpose, you need a parameter manager (see previous discussion
with "parameter" in the subject line) to allow a search through param
space in an evolutionary fashion. As far as I can tell, Gecko's param
manager is (the only one?) suited for that task. 

BTW, does IbM hurt also? If so, sorry for that ;-). Point being is that
that's the term used in the literature and if I want to search for new
literature, I rely on this keyword. Therefore I would want to use this
keyword myself sometime. 

Cheers,

Jan.

Mike Brown wrote:

> I think Paul's question raises two important points - though they may be more
> about the organization of science than about the inherent difficulties of
> SWARM or any other ABM (Thanks, Chris. I like that acronym better too.)
> Ecologists are concerned with aggregations of individuals; they have not had
> to know as much about individual species as, for example, zoologists.
> Similarly,  microeconomists focus on the behavior of individual firms and
> macroeconomists on the aggregate behavior of the economy, etc.
> 
> ABM create two problems. First, the ecologist and macroeconomist have to
> "ratchet down" and study the behavior of individual entities. Moreover, they
> have to know enough about individual behavior to determine which specific
> behaviors might be relevant to the question under investigation. While this is
> not an insurmountable problem, it does demand a new focus for researchers.
> 
> Second, there are some disciplines where data on the behavior of individual
> "agents" simply has not been studied. To take a bad example, look at
> economics. Macro studies the behavior of the aggregate, and micro the study of
> the firm -- but who has been looking at the behavior of the consumer? We have
> been able to make a "rational actor" assumption for so long that we have not
> bothered to collect data about the real-life behavior of induividual
> consumers.
> 
> For these reasons, I think Paul is very right -- modeling and validating the
> behavior of individual entities can be very tricky. 
> 
> Mike
> 
> Paul Fishwick wrote:
> 
> >A key problem with IBM is not that it is necessarily
> >computationally prohibitive, but that not enough data are
> >available to calibrate the model...
> 
>  Paul - why is data calibration more of a problem for IBM
> models than for other modelling technologies? What sorts
> of problem domains are you particularly concerned about?
> 
>  I can allow as how it would be difficult to calibrate
> an IBM model of a forest precisely so that each tree-agent
> is parameterized via data taken from its respective
> real-tree in the forest. Yet, all modelling technologies
> make do with some degree of approximation. A reservoir-flow
> model of tree-species interaction in a forest would simply
> treat all of the trees of each species as "one big tree" 
> of that species with respect to some data (such as  concentration, 
> nutrient uptake, waste-production, and etc.) while ignoring
> other data (such as spatial distribution, variety within the 
> species, and etc.) This will be justified for certain questions 
> about forest dynamics, but not for others, and might make more 
> sense for some problem domains than for others. 
> 
> Thus, there is data and there is data. All modeling technologies
> must pick and choose among the data, and one always has to 
> focus on some reasonable subset of data.  You seem to 
> be suggesting that this is fundamentally more of a problem
> for IBM models than for other modelling technologies
> - can you elaborate?
> 
>   ...and, please!, it almost *hurts* to use the acronym IBM!
> could we use "ABM" for Agent-Based Models? I think it fits
> better anyway, as an agent in this class of models is not 
> always an "individual" in the common sense of that term....
> I know the term has some historical precedent for models
> in this class, but the acronym IBM induces a certain amount 
> of, shall we say, cognitive dissonance, no? (not that the 
> acronym "ABM" itself is inviolate with respect to prior 
> cognitive content....but, still!.....)
> 
> 
> Chris Langton





                  ==================================
   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.
                  ==================================