6.2. Performance comparison based on execution time
In addition to evaluating the performance of the algorithms in terms of the quality of the solutions obtained for a fixed number of evaluations, we have also carried out an evaluation in terms of computation time. Basically, we measure the computation time spent by both pairs of algorithms, MA − MAr and ILS − ILSr, to perform k iterations 6 (recall that both approaches, the classical and the restricted, follow the same path, and therefore, after k iterations the same solution will be reached). Particularly, we compare the computation time of the basic approaches with respect to the computation time of the restricted approaches that include: (1) the computation of the restrictions matrix at the beginning of the algorithm, and (2) a greedy local search under the restricted insert neighbourhood which, for each index, reads in the restrictions matrix the range of the positions to which the index can be moved.
It is clear that the execution time needed to perform a given number of iterations with the restricted approaches, ILSr and MAr, depends on the sparsity of the restrictions matrix associated to each particular instance. That is, the higher the number of zero-valued entries, the lower the checked neighbours, and therefore, the faster the greedy local search. In order to obtain a more general picture of the influence of the sparsity in the execution time, in this experiment, apart from the previously used benchmarks, xLOLIB and xLOLIB2, we have included additional instances with different sparsities. Particularly, we have considered the most common benchmarks for the LOP: LOLIB, LMC and MB. The number of iterations was set to k = 10, 000 without any previous experimentation. Results are displayed in Fig. 9 and Fig. 10.