We should consider what to do with datasets that have NaN values, as the current transformations do not accept them.