REPLY Open Source Simulation Software (SD6814)

[SDMAIL Tom Fiddaman]

Posted by  Tom Fiddaman <tom at ventanasystems.com>

I'm not convinced that competition is the obstacle to SMILE/XMILE 
implementation. That presumes that SD software providers have a strong 
incentive to restrict access to a valuable network or installed base of 
models, or that they behave like the telecom industry (get people stuck 
on your product, then confuse and overcharge them). I don't think 
there's a strong case for either.

Vensim, for example, has effectively had an open model file format (the 
text .mdl) for a long time. The syntax is a little obscure, but 90% of a 
translation can be accomplished with just search & replace in a few 
minutes, especially with a simple model. There are automated tools 
(crude but effective) for Stella and DYNAMO equation translation into 
Vensim. Models are easy to translate; the diagram has traditionally been 
the sticking point, but that should be easy given today's tools. Ten 
years ago I succeeded in writing code to translate a Stella diagram from 
an exported PICT image.

The fact that these tools are hardly used suggests to me that there's 
little demand for translating models. I'm not really surprised by that. 
The core capabilities of SD software packages aren't very different, so 
there isn't ordinarily a compelling reason for the shift. What would be 
much more interesting is translation across modeling paradigms, for 
example from a spreadsheet to SD or from a mathematical programming 
language like GAMS to SD. However, there are significant obstacles to that.

In the first case, the problem is that spreadsheets are unstructured, so 
there isn't an equation set to translate. Manual replication of a 
spreadsheet is horribly painful (as I can attest, having just done one). 
Expose ( http://attunegroup.com/Expose/Expose.htm ) mitigates the 
problem, but has not taken the world by storm, though perhaps it should. 
I'm not sure the pain of translation is the reason for the lack of 
appetite for spreadsheet translation. More likely, it is due to the fact 
that the majority of spreadsheets are dynamically trivial and thus 
rather boring to translate.

In the second case, translation is hampered by the fact that SD 
languages may simply not be expressive enough to capture the content of 
the original. There's no sensible way, for example, to solve an integer 
program with continuous tools. I have translated a few equilibrium 
models into Vensim using the FIND ZERO function, but the result is 
usually less functional than the original (though perhaps more useful 
because it can be connected to other dynamics). In most cases, what is 
needed is not translation, but a structured way to exploit and compare 
results from both paradigms.

Returning to the open source SD question, I'm all for new tools and 
interoperability, but I'm not sure that those are what we need most. 
What I see as a greater need is open models. Those might come in two 
flavors:
- model components, like the Molecules ( 
http://www.vensim.com/molecule.html ), but extended and improved 
collaboratively
- general purpose topical models, including Jay Forrester's "20 models"
There are precedents for open models, including MARKAL and LEAP (both 
energy models with significant community backing). The payoff would be 
to get some modeling productivity enhancements that would make it easier 
to compete with established paradigms like general equilibrium. The 
trouble with such efforts is that there's a tendency toward bigger, 
rather than better models, so mechanisms need to be in place to 
encourage quality and transparency.

If we need new tools and interoperability to get those models done and 
make them usable, then great. Certainly it would help if SD software 
choice were not an obstacle to participation. However, I think the real 
constraint is quality control (per the minimum acceptable model standard 
conversation). Ensuring that models and components are worth sharing 
requires some combination of standards, social measures (ranking), and 
automated quality checks.* Tom Malone et al. at MIT are currently 
thinking about this challenge in a slightly different context.

Tom

*
Reality Check
http://www3.interscience.wiley.com/cgi-bin/abstract/114030779/ABSTRACT
Validation
http://www3.interscience.wiley.com/cgi-bin/abstract/20750/ABSTRACT

Posted by  Tom Fiddaman <tom at ventanasystems.com>
posting date  Wed, 12 Mar 2008 17:00:47 -0600