next up previous
Next: Experimental Setup Up: Storing and Indexing Plan Previous: An Empirical Test of

Experimental Evaluation of the Complete System

 The experiments reported in this section tested the full DERSNLP+EBL system using the dynamic multi-case storage and retrieval strategy described in Section 3. The aim was to evaluate the replay system in a more complex domain. Our hypothesis was that performance would improve over problem solving experience as more negative interactions are discovered and stored. In addition, we predicted that DERSNLP+EBL's method of storage would result in a low library size and low retrieval costs.

The Logistics Transportation domain [40] has become somewhat of a benchmark in the CBP literature. A scaled up version was therefore chosen for this purpose. We tested large multi-goal problems drawn from the domain shown in Figure 4 scaled up to first 6 and then 15 cities. This size of domain is unusual in the current literature.