r magnitude training upper limit plotted against r magnitude at which the F₁ score drops to a fraction of the F₁ score (shown in the legend) without a r magnitude training limit. Below this limit 50% of the data were used for training. A linear relationship is maintained for stars and quasars, while galaxies show a non-linear change between r magnitude of 19 and 20 for F₁ score fractions less than 0.9 (also seen in Fig. 10).