Such polarity differences are examples in which small differences in conformer energies can have significant effects on molecular properties. Since experimental properties reflect a Boltzmann-weighted average of multiple thermally accessible conformers, even small differences in conformer energies have large effects on populations involved in property prediction,  as recently discussed with conformer and polarity effects on solvent viscosity.\cite{Vo_2019}
Machine Learning Batch Evaluation 
An advantage for ML and force field predictions is the ability to batch evaluate by loading all conformers of a molecule at once and evaluating them as a batch opposed to evaluating one at a time, as with conventional quantum chemistry methods. Table \ref{362588} indicates the median sequential times from Table \ref{310223} and median time per single point in batch evaluation. Speedups range ~70-170 times faster for both force field and ANI methods. We note that while the ANI methods improve performance in batch evaluation, traditional force field methods do as well, with similar speedups.