CAn discuss this in the meeting tomorrow. Scipy/sparse doesn't support any ufuncs, so is there a point of using this huge dataset of only 2d tensors?