A multi-dimensional array of data.
Most functions accepting a
Tensor object can alternatively accept a NumPy array.
You can use the arithmetic operators
to perform operations between two Tensor objects.
The value of a Tensor node can be fetched in a graph calculation
by adding its name in the output_node_names parameter for the function call.
This will add an item to the output dictionary in the calculation result object,
whose key is name. The item’s value will be a dictionary
with the “value” of the Tensor as a NumPy array.