Skip to content

README.md

Latest commit

 

History

History
80 lines (61 loc) · 3.27 KB

README.md

File metadata and controls

80 lines (61 loc) · 3.27 KB

test/BoxSolver

A tester which provides very comprehensive tests for BoxSolver, a CDASolver for extremely simple problems where each ColVariable can be dealt with separately subject only to bound and integrality constraints and a linear or quadratic Objective. In fact, is there are any other kind of Constraint in the Block, BoxSolver plainly ignores them and solves the corresponding separable relaxation. However, an interesting feature of BoxSolver is that it at the same time minimizes and maximizes the Objective. The tester also provides some tests for any CDASolver able to handle Linear Programs (such as MILPSolver and its derived classes CPXMILPSolver , SCIPMILPSolver and GRBMILPSolver), as well as for some of the mechanics of the "core" SMS++ library.

Given the number of variables, a single random "box-only" AbstractBlock with separable Objective (a FRealObjective with either a LinearFunction or a DQuadFunction inside) is constructed and solved with a BoxSolver. The AbstractBlock has the following properties, each with fifty-fifty chances:

  • a min or max Objective;

  • a fully linear Objective or one with also quadratic coefficients, in which case the Objective is not necessarily convex for a min problem or concave for a max one, as BoxSolver can deal with it;

  • all continuous variables or some integer ones; however, when there are integer variables and DIRECTION_TEST = 0 (see below) the Objective is selected to be convex for a min problem and concave for a max one, as a MILPSolver may argue otherwise);

  • guaranteed to be always feasible or not;

  • guaranteed to be always bounded or not.

According to the value of the macro DIRECTION_TEST, we then:

  • either compare the get_opposite_value() with the get_var_value() obtained by reversing the sign of the Objective (that must be equal);

  • or compare with the results of an appropriate Solver (e.g., a :MILPSolver.

The AbstractBlock is then repeatedly randomly modified (on bounds and objective, as there is nothing else) and re-solved several times, the results are compared.

The usage of the executable is the following:

   ./BoxSolver_test seed [wchg nvar #rounds #chng %chng]
   wchg: what to change, coded bit-wise [3]
         0 = bounds, 1 = objective 
   nvar: number of variables [10]
   #rounds: how many iterations [100]
   #chng: number changes [10]
   %chng: probability of changing [0.6]

A batch file is provided that runs a large set of tests (which are anyway very quick, in particular if DIRECTION_TEST > 0) with different sizes and seeds of the random generator; all these passing is a good sign that no regressions have been done for the tested modules, and in particular for BoxSolver.

A makefile is also provided that builds the executable including the MILPSolver module and all its dependencies (hence, obviously, the core SMS++ library), although the latter could be switched away when testing with DIRECTION_TEST > 0.

Authors

  • Antonio Frangioni
    Dipartimento di Informatica
    Università di Pisa

License

This code is provided free of charge under the GNU Lesser General Public License version 3.0 - see the LICENSE file for details.