On Thu, Nov 11, 2010 at 11:00 AM, James Bergstra
wrote:
> How hard would it be to write a hessian function in tensor.basic next to
> the grad function? It would use scan internally for now, and later if we
> can figure a faster way to do it, so much the better.

I always get confused about this thing. The hessian as the jacobian can be applied to sparse matrices as well right? As such is not tied to tensor, so maybe it should be outside of tensor ? Like grad is, or dot ?

In the same line of thought, scan works on tensors, should scan be inside the tensor folder ?