The current implementation of Tensor only works on a np.ndarray, and should thus be renamed accordingly.

We need a Tensor type that can accept a tensor from "any" framework; for now that will be numpy and pytorch.