[Coin-discuss] Open-source Modeling Languages

[Sebastian Nowozin]

Hello,

On Nov 20, 2007 4:07 PM, Kipp Martin <Kipp.Martin at chicagogsb.edu> wrote:

> There is very little of it right now, and there is not even an alpha
> version.  This is really on the back burner right now.

Thanks for the information, I can understand that the OS project has
very ambitious goals and that it takes quite some resources.

> However, I don't think OSmL will solve your problem. OSmL does not avoid
> the problem of representing the problem instance in XML format. OSmL is
> designed to produce the problem instance in OSiL format. Hence you still
> have to deal with an instance in OSiL format. OSmL is not a high level
> XML representation -- it generates the instance using XQuery and XPath.
> I am curious, why do you think users will shy away from XML more than
> say, MPS or nl formats?  Indeed I would argue that the XML is a lot more
> readable than MPS or nl formats.

Well, as for machine-interchange formats or programmatic creation, MPS
or XML are perfectly well-suited.  But comparing even a well designed
XML schema to a special-purpose modeling language falls short in my
opinion.  OSmL comes closer to this from a quick glance at the
example.

A language like AMPL or even the MathProg subset is quite natural and
easy to use.  Few people nowadays manually edit MPS files, let alone
.nl files, they are programmatically created from those high level
modeling languages.  OSiL looks to me like something inbetween, and
having exact semantics and syntax is great (you are right about
readability, but it is still a "low-level" readability, giving fewer
insight into the model structure than reading an AMPL model).  This
kind of XML will usually not be edited manually, I suppose?

Thanks for the OSmL link,
Sebastian