0

I have been trying to understand the softmax, and came up with below simple example.

def simpleSoftmax(allValues):
    return np.exp(allValues) / np.sum(np.exp(allValues), axis=0)

Invoke

simpleSoftmax([3,2,4])
array([ 0.24472847,  0.09003057,  0.66524096])

In this case 0.66 has higher probability. Understood.

Now, this shall be done like

(3/9)*100 = 33.33
(2/9)*100 =  22.22
(4/9)*100 = 44.44

Now if we see 44.44 takes higher value, and which results same as softmax.

I am sure there is something interesting behind this softmax with respect to legacy averaging. However i dont understand what is that going to make difference between these two ways?.

Whoami
  • 13,930
  • 19
  • 84
  • 140
  • see https://stackoverflow.com/questions/17187507/why-use-softmax-as-opposed-to-standard-normalization – Dennis Soemers Dec 22 '17 at 13:59
  • 1
    Possible duplicate of [Why use softmax as opposed to standard normalization?](https://stackoverflow.com/questions/17187507/why-use-softmax-as-opposed-to-standard-normalization) – Maxim Dec 22 '17 at 15:20

0 Answers0