Callback for customized branching
Answered
Is there any callback function or other method to impose a customized branching rule at a MILP search node, or to impose a priority for the selection of the branching variables ?
I refer to something as the branch callback by cplex or xpress, or the branch priority file by cplex.
-
Gurobi doesn't offer a branch callback. However, you can set branching priorities with the BranchPriority variable attribute.
0 -
Thanks for pointing to my attention this attribute. Just another question: can we set this value dynamically (I mean, at each node of a B&B tree, inside a callback) or we have to fix it when defining the model ? I know that Gurobi modifies the model at the presolve phase, before starting the B&B, so I have some doubt about the effect of a change of this parameter in a callback.
To be more clear we are working with some machine learning mechanism and we want to experiment their effect in guiding the B&B. As you know some method aims to produce a ranking of the variables which is independent of the evolution of the tree, so a prefixed priority list is ok, but other, and more interesting methods, tries to adapt the rule to the evolution of the tree. In this case we need to adopt a MILP solver that allows to define a possibly different ordering at each node.
0 -
The BranchPriority attributes must be fixed when defining the model, so you unfortunately won't be able to use them to implement a dynamic branching strategy.
0
Please sign in to leave a comment.
Comments
3 comments