Exponential distribution family posterior probability function can be a logistic / sigmod form

Disclaimer: This article is a blogger original article, shall not be reproduced without the bloggers allowed. https://blog.csdn.net/wangfenghui132/article/details/78961760

robust logistic regression is strong, different distributions for the samples can get a pretty good result. Inside said Andrew Ng's courses, logistic function can be used to make the sample in line with the posterior probability function of exponential distribution family. Three years ago, how are their own can not imagine why, still holding a generalized linear model over and over again to see, I did not see a clue. Think about your own learning is really not enough knowledge of the system. Two days ago saw this definition, I understood why.
Manifestation of exponential distribution family reference this link: http://blog.csdn.net/saltriver/article/details/55105285. ligistic function is the reason expressed as a function of all posterior probability is very simple, a sample x, y = probability of 0 is:
the p-(the y-= 0 | the X-)
= the p-(x, y = 0) / the p-(the X-)
= the p-( X, Y = 0) / {P (X, Y = 0) + P (X, Y =. 1)}
= P (X | Y = 0) P (Y = 0) / {P (X | Y = 0 ) P (Y = 0) + P (X | Y =. 1) P (Y =. 1)}
=. 1 / {. 1 + P (X | Y =. 1) / P (X | Y = 0) * P (Y =. 1) / P (Y = 0}
P (Y =. 1) / P (Y = 0) is a constant
p (x | y = 1) , p (x | y = 0) for the same exponential family but parameters just not the same,
all can be p (x | y = 1) / p (x | y = 0) written in the form of an index, all the replaced after all becomes 1 / {1 + exp (-f (x | h1, η1 , h2, η2, p (y = 1) / p (y = 0)))}. wherein h1, η1, h2, η2 represent two exponential family parameters, p (y = 1) / p (y = 0)) it is often said, that is used next to WX of f (x | h1, η1, h2, η2, p (y = 1) / p (y = 0)) fit.
In the logistic regression, usually mentioned logarithmic probability, that such return is used in trying to fit the W * X log {p (x | y = 0) / p (x | y = 1 )}. Regardless of p (x | y = 0) in line with what kind of distribution, if it is exponential families, should have better performance only.