Experiments
We have performed several experiments in order to compare the performance of Ant Colony Optimization and Consultant-Guided Search algorithms.
Before starting the experiments, we have tuned the parameters of our competing algorithms. Tuning a metaheuristic is a difficult and time consuming task, but, fortunately, there are a few tools that help in automating this process.
One of them is based on the ParamILS method, which executes an iterated local search in the parameter configuration space and it is appropriate for algorithms with many parameters, where a full factorial design becomes intractable.
Another one is based on the F-Race method, which sequentially evaluates candidate configurations and discards poor ones as soon as statistically sufficient evidence is gathered against them.
Constructive metaheuristics like ACO and CGS can be combined with local search procedures in order to improve the solution quality. In the first set of experiments we have run the algorithms without local search and in the second set we have run the algorithms with 3-opt local search.