Prunning the tree
AnsweredHello
I am workig on an exact method which embeds with MILP. The idea is to set a heuristic solution into the branch and bound tree. I was able to update the incumbent to a very good point, however the bound still takes time. I need to fathom the tree in early. In this way, I need to get access to the previously fixed integer variables of the current node (that is parent node). How can I fathum the branch from the current node beyond?
Thanks
-
Hi Saeed,
I need to get access to the previously fixed integer variables of the current node
This information is not available.unfortunately.
If your custom heuristic is outperforming the Gurobi heuristics then perhaps it would be worth setting MIPFocus=3 so that the solver focuses on improving the dual bound.
- Riley
0 -
Hi Riley,
thank you for your response, First I tried using CPLEX, and although the cplex was unable to provide me the fixed integer variables over the parent's node, I simply used get.local_lower/upper.bounds method to retrieve such value, My first question is: Does gurobi support such method letting me to identify which one is fixed?
Next notation is, I defined a class of callback in gurobi for getting access to the model context where == MIPNODE, after ensuring that the given relaxed solution is optimal at the node, I used method called model.cbGetNodeRel; in this regard some of the integer variables are showing zeros and some ones, other are fractional. My question is: Can I interpret that the given set for zeros,ones are fixed ?Also, I am curious to know if any method exists letting me to invoke pre-solving during optimization process.
-Saeed
0 -
Hi Saeed,
Does gurobi support such method letting me to identify which one is fixed?
I think you're asking the same question in a different way. The answer is the same.
Can I interpret that the given set for zeros,ones are fixed ?
No, some will be fixed, some will just naturally be at bounds.
I am curious to know if any method exists letting me to invoke pre-solving during optimization process.
There is not.
I think you might have to consider using an open source solver like SCIP for these academic exercises. You can still tell SCIP to use Gurobi as the LP solver, but you'll have a lot more access to internals with SCIP.
- Riley
0
Please sign in to leave a comment.
Comments
3 comments