VRP Results with Optimal Path Completion

Here are results for the 4 vehicle, 8 targets case with low fuel, where the targets are grouped by twos into quadrants. DTMP uses a decision tree to categorize problem instances.

the difference is :
From the training data, the sub-problems associated with the optimal solution are saved at the corresponding leaves. When a problem instance is categorized at a leaf, all of the optimal sub-problems from previous training samples are tested and the ones resulting in the best value for that particular problem instance are added to the set of interesting sub-problems for the master level IP. Computation costs associated with this operation, only includes that of solving for sub-problems that are included in the set set of interesting sub-problems, and does not include all sub-problems that were tested.