As the title says, the indices of a split are determined before pre-split hooks. This will cause errors if samples are dropped during a pre-split hook, as the indices (which were determined before the hook) no longer align with the DataManager (which runs the split after).
I'm denoting this more as a warning for now; this will be resolved soon(tm) when I re-write the data splitting hook structure to better align with how the library actually runs (being inner vs. outer, train vs. test)