[Coin-standards] An argument for public domain modeling infrastructure

[Leonardo B. Lopes]

On Thu, 11 Apr 2002, Alan King wrote:

> 
> I think the last point in Irv's message is the most important.  A new
> modeling standard will be useful only if a significant part of it is in the
> public domain.  The problem with AMPL, GAMS etc etc is that their
> respective developer communities are kept artificially small by nature of
> the licensing.   Developing new features has to have a strong business
> model, which is very difficult in such a low revenue field like
> optimization.  So progress grinds to a halt.

It is interesting that this idea of a modeling standard keeps coming up
over and over again. That is not what I see as the outcome of this work.
It would be nice if we had that, but I don't even think it is possible to
develop a standard for modeling. This, BTW, is my opinion and I don't even
know if my advisor agrees with it. And a standard that can be modified by
vendors (like what Irv proposed) is not a standard at all. Maybe a lot of
vendors wish that the ampl and gams licensing terms were different; that
is reasonable, but to me that is a separate issue from this one.

Modeling takes place in too many heterogeneous environments for there to
be a single solution to the problem. If what you need to do is describe
algebraic structure, the only possible step up from AMPL is writing on a
whiteboard (somehow I think people won't be surprised to hear me say
that). But different models require different things, often beyond what
any modeling language can -- or should -- provide.

I like the idea of a modeling infrastructure. That is different than a
modeling standard. This is what I feel we are trying to set up. A standard
*for instances* is part of the foundation for that infrastructure. It
allows a developer to use a modeling language or library for what they are
good at; then manipulate the generated objects, send them to solvers, use
integration tools, or whatever the situation demands.

> 
> I suspect that the pace of new features in AMPL and GAMS and any of the
> other proprietary modeling languages is probably one new major (like
> nonlinear, stochastic, NL files, etc) feature per decade.  For instance,
> Stochastic-MPS standard has been published since 1987 (ie 15 years) and no
> modeling language has incorporated it.
> 
> Any arguments?  Do we believe that this represents a reasonable pace of
> progress for our field?
> 

Clearly it is not a reasonable pace of progress. There are lots of reasons
for it, though. Maybe progress in modeling languages is more of a symptom
than a cause. BTW, Soon we will have some workable implementations for SP
coming out, from Mitra's group, and one that I am myself working on (when
I'm not writing these emails:)).



========================================================================
Leonardo B. Lopes                                       leo at iems.nwu.edu 
Ph.D. Student                                              (847)491-8470
IEMS - Northwestern University              http://www.iems.nwu.edu/~leo