Fig. 3

Learning curves. The graphs above show that, as expected, performance of the prioritization method on each dataset is an increasing function of training set size. Since the total number of available positive instances varies significantly between datasets, not all sizes could be tested for each dataset