[ Home  | Tools  |  Links/Other  |  Performance World ]

PAVER - GAMS Model Translation Web Submission Tool (GMS2XX)

• AlphaECP
• AMPL
• AmplNLC
• BARON
• CoinFML
• CplexLP
• CplexMPS
• Dict
• FixedMPS
• GAMS (scalar)
• Jacobian
• Lago
• Lgo
• LindoMPI
• LINGO
• MINOPT
• NLP2MCP
• ViennaDag
• ALL (this creates scalar versions of all supported languages, listed above)

Creating a Self Contained Model:

            gams mymodel dumpopt=11

CONVERT Options

CONVERT has options that impact the translation. For example, for a cone model the option CONEREFORM creates an NLP rather than a scalar cone problem. Please refer to the CONVERT manual on the GAMS web page for all options.

In the GAMS model that you are going to submit, you can add options to the CONVERT option file in the following way:

            ...

            $ontext >> convert.opt
            conereform 1
            objvar myobjvar
            $offtext

            solve mymodel min obj using lp;

Background Information

The GMS2XX translator is transforming GAMS models into a very simple internal scalar format. A model that is passed on to a solver can be seen as a linear list of variables, X1 to Xn, a linear list of equations or constraints, E1 to Em, plus the relationship between these variables and equations represented in some form. This internal format of a single model can then be written out in many different formats. With GAMS as output format, the scalar model consists of

• a list of declarations of the variables named X1 to Xn with extra declarations for the subsets of positive, integer, or binary variables,
• assignment statements for the bounds and initial values,
• a list of declarations for the equations named E1 to Em,
• the symbolic form of these equations.

All operations involving sets are unrolled, and all expressions involving parameters are evaluated and replaced by their numerical values. Since there are no sets or indexed parameters in the scalar models, most of the differences between modeling systems have disappeared. The GAMS format is therefore very easy to transform into the format of another language.

The proprietary parts of an industrial model are usually hidden in the names of sets, parameters, variables, and equations; the names of set elements, the numerical values of parameters, the structure of the symbolic equations, and the relationships between these items. During the translation of a user-written GAMS model into scalar format, most of this information is lost. All that is left is the size of the model, the structure of the Jacobian, and some linear and nonlinear relationships with numerical coefficients. The developer of a model will often not even be able to recognize his own model. Industrial users with an interest in improving the solvers they use themselves can therefore in many cases release the scalar version of their model to a model library like MINLPLib.

Example

For questions or comments please contact the Performance World managing editor.