Basically, I failed to trying to implement individual global MINLP solvers (alphaBB, ANTIGONE, BARON, Couenne, LindoAPI, and SCIP) in C++/Xcode/Mac system. On the other hand, I realize that GAMS already integrate a lot of solvers. So I am thinking of using C++/Xcode/Mac system to construct models and let GAMS solve that. What would be the standard practice in this respect? Thank you.
Standard practice tends to be one of:
- formulate the model in GAMS and solve it in GAMS
- formulate the model in C++ and solve it in C++
In either case, you can read data from a text file and use it to populate your model instead of hard-coding inputs.
GAMS does have a C++ API. At one point, I've used it, and based on that experience, I wouldn't recommend it, simply because mixed-language programming tends to be challenging to maintain. Nevertheless, if you wanted to generate data in C++, and then use the GAMS API to import that data into GAMS, you could do that as well.
Some of the solvers you cite are not available from C++. For instance, BARON interfaces with GAMS and MATLAB, but does not have a callable C++ interface. ANTIGONE also has a GAMS interface, but no publicly available C++ interface. (Perhaps you have a version directly from the authors?) alphaBB is an algorithm for solving MINLPs, but I don't know of software called alphaBB that solves MINLPs.
I don't have any experience using Xcode as an IDE for software development, but at minimum, you should be able to get Couenne and SCIP to work on a Mac by compiling them and linking your software appropriately. If you are having trouble with this step, I'd recommend decoupling data generation and MINLP model formulation. Generate the data using a C++ program, and then read the data in GAMS and formulate your MINLP in GAMS, then call GAMS to solve your model.