Hey @TomAugspurger, how opposed would you be to introducing small deviations from the scikit-learn interface to PCA like adding a flag in dask-ml.PCA to avoid requiring this centering?

That would certainly be a bad default, but if I know that is going to happen downstream from a scaling operation in a pipeline that already does the centering, I would like to have the option to avoid the extra operation. Is that reasonable or are you trying maintain something closer to perfect consistency?