# Neural network symbolic inputs and weights

If I create a neural network like, for example, so:

net = NetGraph[{LinearLayer[3], LinearLayer[5], LinearLayer[1]}, {1 -> 2 -> 3}, "Input" -> 2];
net = NetInitialize[net];


I can then evalute its output at a set of numeric inputs:

net[{1, 1}]


When I do this with symbolic inputs, however,

net[{x, y}]


I get an error:

"NetGraph: Data supplied to Input was not a length-2 vector of real numbers (or a list of these)."

Is it possible to view a symbolic representation of the outputs as a function of symbolic inputs (and potentially symbolic weights)?

Note: I know I could reimplement neural nets without using the convenience functions to achieve this. My question is, to avoid reinventing the wheel, can I use NetGraph / other neural net functions to do this?

• I don't believe this is supported at the moment. – Carl Lange May 18 at 14:00
• – Carl Lange May 18 at 17:51