The softmax function maps a real vector 
to a vector of positive
components that sum to 1. For 
, its components are
 |
Every component is positive and the components sum to 1. Adding the same constant to every component of 
leaves the result unchanged. For 
, the softmax function can be expressed using the sigmoid
function. It is widely used to convert a collection of real-valued scores into
probabilities for a categorical distribution.
