This is similar to a FeatureGenerator, in that it does the same thing as the first step of PRA, but it does not actually produce a feature matrix.
Things this doesn't capture very well: - Recursion (if you want to demonstrate where PRA fails with respect to ProPPR) - Mutual exclusivity of certain relations (and maybe other kinds of relation metadata), though PRA doesn't do a great job exploiting this right now, anyway - Characteristics of actual data.