You can change the MIP Strategy options for XPRESS by choosing XPRESS parameters from the Options menu and then pressing the MIP Strategies tab. This will display the dialog box shown below:
Figure 4.58: The MIP Stategies Tab in XPRESS Options Dialog Box
Option Name | MPL Name | Solver Param | ParamNr | Type | Default | Min | Max |
---|---|---|---|---|---|---|---|
NodeSet Selection | MipNodeSelect | NodeSelection | 8026 | list | 1 | 1 | 5 |
Breadth First | BreadthFirst | BreadthFirst | 8082 | int | 10 | 1 | MAXINT |
Node Selection Criteria | NodeSelCriteria | Backtrack | 8027 | list | 3 | 1 | 3 |
Target Objective | MipTarget | MipTarget | 7011 | real | 1e+40 | 0 | MAXREAL |
Relative Gap | MipRelativeGap | MIPrelStop | 7020 | real | 0.0001 | 0 | 1 |
Absolute Gap | MipAbsoluteGap | MIPAbsStop | 7019 | real | 0 | 0 | MAXREAL |
Variable selection | MipVariableSelect | VarSelection | 8025 | list | -1 | -1 | 6 |
Estimate deg. Mult | MultEstDgrade | DegradeFactor | 7010 | real | 1 | 0 | MAXREAL |
Default Pseudo Cost | DefPseudoCost | Pseudocost | 7015 | real | 0.01 | 0 | MAXREAL |
The Nodeset selection option has the following possible selections:
Local first search (1) | Choose between descendant and sibling nodes if available; choose from all outstanding nodes otherwise. |
Best first search (2) | Choose from all outstanding nodes. |
Local depth first search (3) | Choose between descendant and sibling nodes if available; choose from the deepest nodes otherwise |
Best first, then local first (4) | Best first is used for the first n nodes, after which local first is used. |
Pure depth first (5) | Choose from the deepest outstanding nodes. |
The number of nodes to include in the best-first search before switching to the local first search. The default is 10.
The Node selection option has the following possible selections:
Forest-Hirst-Tomlin (1) | If a target object function has not been set, choose the node with the best estimate target object function for global. If a target objective function has is been set (by the user or from a previous IP solution), the choice is based on the Forrest-Hirst-Tomlin Criterion. The Target obj. function used in "best estimate" node selection technique. This is set automatically after solving the LP relaxation unless set by the user. |
Smallest solution (2) | Always choose the node with best estimated solution. |
Smallest bound (3) | Always choose the node with best bound on the solution. |
The target object function for the global search (only used by certain node selection criteria). This is set automatically after an LP optimization routine, unless it was previously set by the user.
The relative MIP optimality stopping criterion.
The MIP search will stop if the relative optimality gap, ABS(best solution - best bound) / best bound, is less than or equal to this criterion value. The default is 0.0.
The absolute MIP optimality stopping criterion. The MIP search will stop if the absolute optimality gap, ABS(best solution - best bound), is less than or equal to this crietrion value. The default is 0.0.
This determines the formula used to calculate the estimate of each integer variable, and thus which integer variable is selected to be branched on at a given node. The variable selected to be branched on is the one with the minimum estimate. The variable estimates are also combined to calculate the overall estimate of the node, which, depending on the BACKTRACK setting, may be used to choose between outstanding nodes. The option has the following possible selections:
Automatic (-1) | Determined Automatically |
Min(up, down) (1) | The minimum of the 'up' and 'down' pseudo costs. |
Up + Down (2) | The 'up' pseudo cost plus the 'down' pseudo cost. |
Max(up, down) + 2 Min(up,down) (3) | The maximum of the 'up' and 'down' pseudo costs, plus twice the minimum of the 'up' and 'down' pseudo costs. |
Max(up, down) (4) | The maximum of the 'up' and 'down' pseudo costs. |
Down (5) | The 'down' pseudo cost. |
Up (6) | The 'up' pseudo cost. |
Factor to multiply estimated degradations associated with an unexplored node in the tree. The estimated degradation is the amount by which the objective function is expected to worsen in an integer solution that may be obtained through exploring a given node.
Default pseudo cost used in node degradation estimation, default is 0.01.