Manual Tuning Overview

Solving a mixed integer problem usually involves five stages, depending on how hard the problem is:

1. Presolve.
2. Solving the initial LP relaxation.
3. Heuristics.
4. Cutting.
5. Branch-and-bound tree search.

This guide addresses those five parts of the solving process separately in the following sections.

How to tune the Xpress Optimizer for a particular problem depends very much on what your goal is:

Find any solution.

For some problems it can be hard to find that first integer solution. The emphasis in such a case will often be on heuristics and on guiding the branch-and-bound dives away from potential conflicts.

Find good solutions quickly.

If the goal is to find better solutions quicker, without necessarily providing any bound on how good the solutions are, the emphasis will usually be on heuristics and a fast branch-and-bound search of the solution space. Note that since release 8.12, the Optimizer has a special mode with a focus on primal heuristics. To activate this mode, set HEUREMPHASIS=1. We advise trying this first before conducting a long tuner run or trying various heuristic controls individually.

Solve the problem to optimality.

The Optimizer is by default tuned towards finding an optimal solution to a problem. Default settings balance heuristics for finding good solutions with the necessary raising of the best bound to prove optimality.

You can customize the Tuner for your individual goals. The TUNERMETHOD control allows using different ground sets of controls that the Tuner evaluates. The TUNERTARGET control allows setting different measures by which individual solves should be compared, including the time, the gap, the final solution value.