Empirical results

Abstract

This chapter presented empirical results comparing the performance of the analogical reasoner, Prodigy/Analogy, with the base-level problem solver Nolimit.

The extensive results obtained from a logistics transportation domain showed that:

• The analogical reasoner increased the solvability horizon of the base-level problem solver considerably. Within a CPU running time bound of 350 seconds, the complete set of 1000 problems was solved by Prodigy/Analogy, while only 458 of these problems are solved by Nolimit (see Figures 8.6 and 8.7).

• The cumulative running times for the analogical replay of the problems represent a speed-up of up 3.6 over the base-level problem solver, if only the problems solved both without and with analogy are considered. The speed-up increases to 5.3 if the problems not solved are also accounted for with the CPU time limit given to the base problem solver (see Figures 8.8 and 8.9).

• The solutions obtained by analogy are of equal or shorter length than the corresponding ones found by Nolimit for 82.75% problems (see Figure 8.10).

• When the retrieval time is added to the analogical replay time, Prodigy/Analogy still performs more efficiently compared to Nolimit for 293 problems out of the 458 problems solved by both configurations. The other 165 problems correspond to simpler problems for which Nolimit finds a solution to a problem in a shorter time than Prodigy/Analogy retrieves analogs for and replays them (see Figures 8.11 and 8.12).

• Finally the retrieval time suffers no significant increase with the size of case library (see Figure 8.13).