To recapitulate, our experiment indicates that simply storing ``all'' the homogeneous rules, as performed in our implementation, did not improve the prediction accuracy of BruteDL. In fact, certain aspects of BruteDL concerned with limiting search to only best rules really precluded our implementation from storing all rules.
These experiments suggest that storing only the homogeneous rules that are the best for at least one training examples is a rather effective method in terms of both the prediction performance and the efficiency of the learning system, though the reasons for this are not fully understood. Loosening this restriction, as we implemented it, does not improve the performance of BruteDL.