The sign flipping just makes sure all the corresponding components in U and V^T are on the same side of the axes and doesn't affect the correctness of the results. Currently we are not using it, the transformed data was required which shouldn't be necessary. Currently, our pytests are comparing against scikit-learn without respect to the sign, which is still valid, however we should correct the sign flipping and add it back in at some point just so that we can guarantee exact match w/ sklearn. The benefit here is to prevent confusion from users when they compare the results side-by-side.