[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
fest discussion
From: |
Doug Donalson; |
Subject: |
fest discussion |
Date: |
Tue, 24 Feb 1998 18:48:48 -0800 (PST) |
Is anyone interested in putting together and discussion
group on model testing and verification at SWARM fest?
Here is my "take" on it with examples from my research.
One of the most difficult and important aspects of complex
modeling is that of model testing and verification. It
falls into two broad catagories, verifying that the model
is bug free, and determining the effects of what I call
model infrastructure.
Two examples of the first catagory:
Test individual components:
Ex: I have an agent that in the absence of all other
effects dies off exponentally so I turn off all
other effects and confirm that all the individuals
die off and that the half life time matches the
death rate.
Compare with simpler models:
Ex: I have a "base line" delay differential equation
model that provides stability boundries and
equlibrium values for a set of parameters. I
developed a second model that is the analog of
the DDE but with the quantization effects associated
with requiring the number of individuals to be
an integer. This is the Stochastic Birth/Death
or SBD formulation.
If I force my complex model to be "well mixed"
spatially, then it should match the results of the SBD
for the same system size and parameter choices. The
SBD model can then provide the parameter space
where the quantization stochasticity has a small
effect on the system dynamics. This allows a
set of parameters or "parameter space" in which
my complex model should have very similar results
to the DDE model.
Bottom line: I find a parameter space that if my complex
model is working correctly it should provide results
that match those of an analyticaly (or at least
numerically) tractable model.
The second catagory has to do with the effects that the
actual model implimentation may have on the results.
This is basically worrying about artifacts from model
structure.
Ex: Any time you design a model around a lattice
structure you by definition quantize space.
Does this "minium size" of one square have
a significant effect on the results?
Ex: When using a synchronous schedule how does the
ordering of events within a time step affect
the outcome. For example, if it was a predator/prey
simulation, if predation always came before prey
birth you would potentially expect a different
outcome then if births came before predation.
I know this is long winded, but these are key issues if
we are to believe and understand the results of complex
models. Is anyone interested in tossing around problems
and solutions in these areas? Let me know if anyone is
interested.
Cheers,
Doug Donalson
***************************************************************************
* Doug Donalson * Office: (805) 893-2962 *
* Ecology, Evolution, and Marine Biology * Home: (805) 961-4447 *
* UC Santa Barbara * email address@hidden
* Santa Barbara Ca. 93106 * *
***************************************************************************
* *
* The most exciting phrase to hear in science, the one that *
* heralds new discoveries, is not "EUREKA" (I have found it) but *
* "That's funny ...?" *
* *
* Isaac Asimov *
* *
***************************************************************************
==================================
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.
==================================
- fest discussion,
Doug Donalson; <=