I am making a custom optimization function and I want to somehow get the activation vector coming into the node.
As far as I understand:
p.data is the current value of the parameter
p.grad.data is the current value of the gradient for parameter
How do I get the actual data coming into the parameter?