Hi all,
following Problem. I'm trying to classfy xy coordinates as above (output +1) or below(output -1) a parabola. For this I am trying to use a neural network with tanh() activation functions and set the treshhold at 0 ( so if the net outpus a 0.0001, thats +1,0.000 and smaller is -1).
How exactly do I calculate the error/the adjustment to the weights when doing this? Just use the derivative of tanh() in combination with the difference between my (binary) output and the (binary) target?
[link][comment]