This keyword indicates activation function of neurons. The
input and output data are normalised (only training set is used to determine
range of values) to the default range of each function.
1/(1+exp(-x))
{0} -- the famous logistic activation function. The range of values
is [0.1, 0.9]. th(x) {1} Symmetrical
over 0, [-0.9,0.9] x/(1+|x|) {2}Symmetrical
and fast for calculations, [-0.9,0.9].