softmax

converts scores to probabilities

$P(y_i)=\frac{e^{y_i}}{{\Sigma }_j{e}^{y_j}}$ ensures output is between 0 and 1, and sum of all scores = 1