Exceptions were R10, R11 and R12, with approximately 3% lower performance for the directly reduced variants, meaning that predictors chosen directly from the baseline variant varied only slightly compared to variants with intermediate training, even at higher reduction steps. Furthermore, even the ...