[FEA] Adding constraints and variables to an existing LP problem (C/C++ API)

[ktp-max]

Hi all,

Thanks for your great work in developing cuopt.

I would like to ask if you could provide a feature in the C/C++ API for expanding an existing LP problem by adding additional constraints and variables and/or modifying existing constraints and variables (changing a matrix coefficient, an objective coefficient or a constraint/variable bound).

There seems to be something in the python API regarding adding constraints/variables, but I have not seen anything related in the C API. For the latter, it seems that the only way for modifying an LP problem is to delete the existing instance and build/solve from scratch the augmented one. In the C++ API, there are several get-type functions in the class optimization_problem_t which return non-cost references to containers e.g. :

rmm::device_uvector<f_t>& get_constraint_matrix_values();

Could these be used for augmenting the existing problem? Would that be safe?

Most simplex solvers support problem modifications since these provide runtime benefits if the primal/dual of the initial problem stays feasible in the modified instance. It would be very nice to have this also in cuopt, if possible.

Thanks,

C.