pythorch（五） Softmax理解

pythorch（五） Softmax理解

 

>>> import torch
>>> import torch.nn.functional as F
>>> from math import *

>>> a=torch.tensor( [-2.1441,  1.8928, 0.45])
>>> F.softmax(a)
tensor([0.0141, 0.7975, 0.1884])
>>> x = pow(e, a[0])
>>> y = pow(e, a[1])
>>> z = pow(e, a[2])
>>> x/(x+y+z)
0.014077570436341351
>>> y/(x+y+z)
0.7975006948005181
>>> z/(x+y+z)
0.1884217347631406