"SIG for Open Standards in System Dynamics"

[Robert Muetzelfeldt]

In his posting of 6 Jan 2011 on the XMILE thread, Bob proposed that:
"If there is sufficient interest we can add a forum specific to model interchanges issues to the discussion boards. This could even be done as a Special Interest Group activity. Perhaps this could be the Special Interest Group for Open Standards in System Dynamics."

There was little response to this, nor indeed to the XMILE thread itself, despite some promptings from various people. I have a strong interest in the value of open standards in modelling, and I'm frustrated that so little progress has been achieved within this community. Those of us interested in this need both a discussion forum (somewhere) and - as I have expressed in the XMILE thread - some form of collaborative space (a wiki) for gathering together ideas etc.

I'd like to strongly support Bob's idea of setting up a forum on this site. However, if there is insufficient desire within this group to do that, then I'll be happy to set up one somewhere else - probably as a Google group.

Cheers,
Robert

[Leonard Malczynski]

Robert,
There may be some value in differentiating between open modeling standards and open modeling software standards aka XMILE.
SD has a very nice, succinct modeling standard of stocks and flows (okay, and an auxiliary or two).

The vendors have been differentiating their products via interface, functions, data connectivity, model size, etc.
I believe that they will continue to do so, after all, that is how they compete.

Believe me, I too would like to have zero switching costs when bringing a model from Vendor A into a tool from Vendor B.
How the community gets competitors to come to agreement is a challenge and possibly not in the best business interest of the software firms.
In addition, making a product conform to open standards has to have a payoff.
The work required to conform may take valuable resources away from other improvements to the software.

I believe that there may be some indication of the difficulty and consequences of this if we look at the RTF standard for word processing and how that market ended up.
Len

[Bill Braun]

Len writes that changing standards may not be in the best interests (real or perceived) of software publishers, and may distract resources from other more desirable (by some measure) improvements.

Len, I think your statement is spot on so far as it goes. But isn't the history of standards in software a very fluid one? They change periodically. For example, I cannot read models constructed in newer formats. If formats change anyhow, targeting a common standard for the next change ought not be much different from a proprietary one developed under the cover of darkness.

Hasn't such an open standard been largely adopted by makers of word processing software (XML)? Ignoring opinions of how well any one program is doing it, it appears to be a common target.

I wonder (truly wonder) if there is a marketing mental model that convinces software companies that once they win a customer they have to lock the customer in for fear of losing them? I could make the argument that an open standard would actually influence companies to attract and keep customers as a function of innovation, not giving them the option to rest on laurels because customers face considerable switching costs.

(I'm not slamming the folks making software, nor implying they do not put enormous effort into their products. I only ask if it might be more so should innovation be the dominant basis for competition.) It also strikes me that with the introduction of the System Dynamics Case Repository (http://cases.systemdynamics.org) the SD community as a whole could greatly benefit from an open standard (in terms of increasing its acknowledgement by business (and other sectors) as a tool-of-choice).

Bill

[Thomas Fiddaman]

From the other side of the fence (software publishing), I don't think the competitive mindset is at the heart of lack of progress on standards. I'm sure that mindset exists in some places, but it's really not compelling, at least to us with Vensim. The competition is not really ithink and Powersim; it's spreadsheets and a hundred other tools. Judging by Google volume, Matlab and Mathematica are 100x the size of SD tools, and Excel is another 100x bigger. Interoperable file formats create a collective advantage for us little guys.

isee (iThink/STELLA) is to be congratulated for actually making this happen. For us, progress has been held up by some more pressing architecture questions that need resolution first. But, we still think it's a good idea.

My hypothesis is that there's actually not much demand for moving complete models from one platform to another. Certainly we don't see much demand for it, though that could be a chicken & egg problem (it's hard, so no one bothers to ask). I think the real value from open code is going to be at the level of Molecules of structure. For that to work, there may be some infrastructure needed in addition to an interchange format - a social site for sharing structures, perhaps a micropayment system for some cases, an embedded documentation format, etc.

Quality control may be the biggest hurdle. How do you know that a molecule you pick up on the web is any good, and that you're using it appropriately? Vensim's Reality Check equations make it possible to embed some tests in a structure, but I'm not sure they're readily translatable. They're not widely used, which means that modelers may need to adopt a new mindset when creating content for sharing.

[Leonard Malczynski]

Tom,
Your argument was one I also made a few years back.
In fact the community is so friendly that models (or at least molecules) seem readily available.
Some are so small that manual translation is not that difficult and is a great learning experience.

I agree, the molecule level is where the action should be. The Vensim Molecules are useful and small enough to translate.
Someone may have done that on their own (is anyone listening?) and make them available to all of us.
I suppose that putting the translation burden on the vendors is the easy way out.

I agree with your quality control note.
Many of us (or is it our management) seem to have a belief that models that come from others are well built.
Perhaps it is the institutional provenance that misleads us!

We have seen several situations of 'benchmarking' an SD model to either another SD model or another methodology.
The burden seems to be placed on the SD modeler to explain why she can't, as silly as it seems, match output from the external models!

I'm glad this thread is at least rolling over in its sleep!
Len

[Jean-Jacques Lauble]

Hi everybody

Concerning the utility of investing into the possibility to ease the translation from one language into another, one can consider the different needs of people interested into this feature.
From my own experience I can identify two main usages.

First usage: mainly the possibility to translate public models published for instance in the SD conference.
In this case, the utility is mainly informative and mainly technical as in 99% of the case I have no practical knowledge of the problem studied. There is no real stake and there may be many models concerned. In this case, the utility is to increase the number of models that can be studied.

One can in this case, argue that often models are available in a player format or in an original non recordable version (like the I Think one).
But the main interest in a model is the capacity to analyze it, and if for instance I download an I Think model, I may run it, but not knowing how to use the analytical tools in I Think, the interest is quite limited. Learning how to use the main analytical tools in I Think and Powersim is not possible, because it is only by practice that one learns to use them.
The analytical tools are mainly, sensitivity analysis, reality checks (available only in Vensim), gaming, partial simulation, optimization, synthesim and the possibility to override the behavior of the variables, runs compare etc… .
Another feature that I use more and more, is making batch analysis, by driving all the preceding tools from a VBA program and making automatic analysis (mainly during the night). I do not know how to do that using another tool.

At least, one can too argue that not knowing practically the problem, the need to make a deep analysis is restricted. But there is still a need, although not critical.

Another remark concerning that first usage, is that it would be more useful for the SD conference papers, to verify that they conform to some basic rules like dimension consistency and in many cases, physical mass balance, which is unfortunately very often not done. I personally have no interest to study a model with unit errors or with no mass balance verification when it has rather complex physical flows.

By making this remark, I suggest first concentrating on easier goals to achieve and highly rewarding.
But apart from this last remark, building translating tools makes sense.

Second usage:
The second usage, which I have nearly never encountered, but that can be a concern to other people, is finding a model whose problem is close to mine, and that I could get some practical usefulness out of it. In this case, the interest is much higher and will justify investing into a manual translation for many reasons.

First translating a model from one language into another is highly instructive.

Due to the stake, it will be safer to do a manual translation as it will be more secure. It will eventually spot some modeling errors.

The time needed to translate will be marginal, compared with the time spent with the analysis of the model.

To resume that second case, the utility of translating tools is week.

One last remark is that most of the publicly available models are more or less academic or have been simplified to fit to the academic word (limitation of the documents, data needed, size of the paper, interdiction for commercial reasons to publish the full model etc…).
They are then more often from the first type of usage than from the second.

Regards.
Jean-Jacques Laublé.

[Travis Franck]

I would like to see an open and common format but a reason other than moving between SD modeling packages. A common format would speed up innovation for getting models out of the SD packages and onto other platforms, such as the web, iOS, Android, Java, etc.

First, I understand that many of the vendors have solutions for some of these platforms. But none have really good solutions for many. Instead, what would be really great is a workflow that allowed SD modelers to use their SD software of choice, and then save to a common format, and then convert that model into a target platform of choice. Take a model and convert it to a programming language like C or Java or Javascript. That would allow for rapid, visual development in Powersim, Vensim, iThink, with all the great development tools they provide (RCs, unit checking, etc.). But then modelers could do more innovative things with the models. Flight simulators, tablet apps, educational tools, etc. Many more possibilities, many of which I'm sure I can't even imagine.

A common file format would allow the community to develop a single conversion tool that would convert that open format to another platform. Additional target platforms could then be added by the community. And vendors would only have to support the common file format.

I have searched recently for conversion tools and, so far, have found several different attempts to convert a specific file format to a specific target. Sometimes people were redoing the work done by someone else. Having a single conversion tool would like help every one, and the community as a whole.

[Robert Muetzelfeldt]

Thanks for the thoughtful responses to my original posting.

In fact, it was not my intention to spark a discussion on the pros and cons of a an SD markup language, whether vendors would want to invest in developing one, etc. Instead, my original question assumed that there is a group of people out there who are interested in discussing the development of such a language (rather than whether there should be one), and I was asking where that discussion could be hosted.

I asked the question (rather than just starting a forum somewhere else) because I do feel that this forum would be the ideal place to host it, especially given Bob's floating of that idea in his words which I quoted. But it seems that this feeling is not shared by members of this forum, indicating that the right course of action is for those of us (assuming it's not just me...) who want to discuss the practicalities to go off somewhere else.

This creates a bit of a problem for me personally, because (sorry, don't know how to express this better) I am not really a System Dynamics person. Rather, I am an ecological modeller who has developed software (Simile) which happens to have an SD capability. This is to a large extent a difference of culture - there is a lot of overlap between "ecological models" and "System Dynamics model" - but none-the-less real for all that. However, I have for many years found it odd that the SD community has not moved towards a common model-representation language (or rather, has not turned the on-off discussions over this time into a community endeavour). Hence my setting up of the XMILE wiki during a previous iteration, to try to help things along. And I would again be happy now to join in with a community initiative to work through these issues.

I'd like to make three further points:

First, I would really encourage anyone interested in this discussion to check out SBML - the Systems Biology Markup Language (http://sbml.org/Main_Page). Systems Biology, which is largely concerned with the modelling of biological pathways, emerged a long time after SD - indeed (some time after the first mention of an SD interchange language...). But quite soon the community (largely the developers of similar-but-slightly-different software tools) got together and designed a common data model and an XML realisation. Significantly, this has spawned the development of over 200 (!) software tools capable of reading and writing SBML- almost exactly reflecting Travis's thinking in the most recent posting in this thread. Also significantly, the BioModels web database now holds some 820 models expressed in SBML, along with web tools for viewing and simulating these models, and generating code in various languages (again reflecting Travis's ideas).

Sure, there are differences between the two fields, but there are also significant similarities. For example, the reaction concept which underpins SBML is also a cosmetic overlay on top of a set of ODEs, just like SD; and some of the SBML-compliant tools are commercial. So I don't think we should dismiss the SBML experience outright.

Second, and again reflecting Travis's posting, I think that it is better to think in terms of a declarative model representation language rather than a model interchange language. Once we have an open, standard, XML-based language, then all sorts of things will flow from that. Anyone can develop an XSLT transform (for example) which can display/process an SDML model in some novel way, and put it on the web - totally changing the culture away from the exchange of models between "the big three". For an example of what I have in mind, please look at http://www.itfutures.ed.ac.uk/robert_muetzelfeldt_synopsis.htm and the associated presentation.

Third, we are starting to see the emergence of various new kids on the block when it comes to SD software, including web-based ones. Maybe not individually, but collectively these are game-changers. Being typically free-to-use and small, the inclination to buy into a common language is much greater.

[Steven Adler]

I think an open standards based initiative to develop an XML markup language for System Dynamics is a great idea. The SD vocabulary is rather small, which has enabled many small vendors to build limited integration without open standards. A standards drive would make it easier to port models between vendors and integrate other modeling disciplines as well. It would help the industry to grow and as a participant and leader in many past open standards initiatives, I'd be happy to help with this one.

[Steven Adler]

I hope all are stil interested in creating a SIG for Open Standards in System Dynamics. This industry needs an open XML standard that describes SD models. I agree with Tom that there probably isn't a huge market for model portability between tools, since modelers invest a lot of time in learning to master each tool. But I believe that there is an untapped market for model re-use, portability between modelers, and the construction of interchangable model components.

How does one go about setting up this SIG? Its time to move this idea forward from concept to reality.