I built a neural network in tensorflow which should predict the probabilities of an event to happen. As output function I therefore passed softmax. I tried to customize the softmax function so that the probabilities of all (three) classes sum up to 1.05 and the minimum (maximum) probability is 0.02 (0.99). I have the following piece of code so far but I am not sure whether this is correct
def custom_activation(x):
return K.softmax(x)*(1 + 0.05)
Does anyone know how to implement this?