MPL for Windows Manual

Change CPLEX MIP Strategy Options

You can change the MIP strategy options for CPLEX by choosing CPLEX parameters from the Options menu and then pressing the MIP Strategy tab. This will display the MIP strategy options for CPLEX in the dialog box as shown below:

The Node selection option is used to set the rule for selecting the next node to process when backtracking (proceeding back through the tree when a node is infeasible or cutoff). The Best Bound method works better for many problems; but Depth First may find a good solution deep in the tree faster. Depth first will also use less memory.

• Depth first The depth-first search strategy chooses the most recently created node.
  
• Best bound The best bound strategy chooses the node with the best objective function for the associated LP relaxation (default).
  
• Best estimate The best estimate strategy selects the node with the best estimate of the integer objective value that would be obtained from a node once all integer infeasibilities are removed.
  
• Best estimate alternate Alternate best-estimate search.

• 0 Best estimate node always selected
  
• 1 Best bound node always selected
  
• >1 The interval at which the best bound node will be selected from the tree.

Variable Selection

The Variable selection option is used to set the rule for selecting the branching variable at the node which has been selected for branching.

Minimum infeasibility	The minimum infeasibility rule chooses
			the variable with the smallest fractional 
			value. It may lead more quickly to a first 
			integer feasible solution, but will usually 
			be slower overall to reach the optimal integer 
			solution.

Automatic		Branch variable automatically selected. 
			This option allows CPLEX to select the 
			best rule based on the problem and its 
			progress (default).

Maximum infeasibility	The maximum infeasibility rule chooses
			the variable with the largest fractional value; 
			It forces larger changes earlier in the tree, 
			which tends to produce faster overall times to 
			reach the optimal solution. 

Pseudo costs	        Pseudo costs causes variable selection based 
                        on pseudo-costs which are derived from pseudo-
                        shadow prices.

Strong branching	Strong branching causes variable selection 
			based on partially solving a number of subproblems 
			with tentative branches to see which branch is the 
			most promising. This strategy can be effective on 
			large, difficult MIP problems.

Pseudo reduced costs	Pseudo-reduced costs causes variable 
			selection based on pseudo-costs which are derived 
			from pseudo-shadow prices. 

MIP Probe

Determines the amount of variable probing to be performed on a problem. Probing can be both very powerful and very time consuming. Setting the value Probing level 1-3 can in some cases result in dramatic reductions or dramatic increases in solution time on particular models.

No probing
Automatic	
Probing Level 1
Probing Level 2
Probing Level 3

MIP Priority order

A priority order assigns a branching priority to some or all of the integer variables. Variables with priorities will be branched on before variables without priorities. Variables with higher priorities will be branched on before variables with lower priorities (when the variables have fractional values). To switch off this function, make sure the box is unchecked.

Use MIP priority order	  When this option is checked, will use the priority order
                          (if it exists) for the next mixed integer optimization.
Do not generate	
Use decreasing cost	
Use increasing bound range
Use increasing cost per coefficient count	

Branch Direction

The Branch direction option is used to decide which branch, the up branch or the down branch, should be taken first at each node. For some problems, directing the algorithm to always branch up or down can improve performance.

Branch down 		Branch down (restricted to lower value).

Algorithm select	The algorithm selects the branch direction (default).

Branch up		Branch up (restricted to higher value).