Jacobin matrix of the NN's output respect to the input
1 view (last 30 days)
Is there a convenient built-in funciton for the NN can get me the Jacobin matrix of output respect to input? For instance, # input (feature) = 4, # output = 3. Then this function gives me a 3*4 matrix.
I found one called defaultderiv, but this gives the Jacobian of errors with respect to the net's weights and biases, which is not what I want.
Even though, by knowing the weights, biases, active function, I could form a symbolic equation fed into jacobin( ) function to obatin what I need. But it seems a bit messy.
Thanks for your answer