Machine learning notes-talk about generalized linear models

The concept of generalized linear regression is derived from linear regression and Logistic regression, and many people are still confused and wonder why a generalized linear regression is a sudden one. As long as you know that continuous value prediction is using linear regression, discrete value prediction with Logistic regression is not OK? There are also relationships between concepts that are not clear, such as linear regression and Gaussian distribution, Logistic regression and Bernoulli distribution, connection functions, and response functions.

This confusion is understandable, in order to guide the Quick Start, from the point of view of practical problems to introduce the answer, but the probability hypothesis behind the explanation is insufficient, although the linear regression specifically opens a section to introduce the Gaussian distribution hypothesis, but many people mistakenly think that the purpose of this section is only to prove the minimum mean square error rationality, The Bernoulli distribution hypothesis for Logistic regression also needs to be explained.

The linear regression is based on the hypothesis of Gaussian distribution, and the Logistic regression is based on the hypothesis of Bernoulli distribution. If linear regression and Logistic regression cannot be understood from the perspective of probability, it is impossible to understand generalized linear regression by ascending one level, and the generalized linear model is to include other distributions in the same way, to extract the common points of these distribution models and to become a model so that other distributions such as polynomial distribution, Poisson distribution, Gamma distribution, exponential distribution, beta distribution and Dirichlet distribution can be calculated by step-by-step model.

Resources:

1, Http://cs229.stanford.edu/notes/cs229-notes1.pdf

Machine learning notes-talk about generalized linear models