Transferred from http://www.cnblogs.com/ModifyRong/p/7739955.html
1. Introduction
Logic regression is a very like to ask in the interview of a machine learning algorithm, because on the surface of the logical return form is very simple, very good grasp, but a question is easy to get confused. So in the interview when the first advice to everyone do not say that they are proficient in logical regression, very easy to be asked to pour, thereby reducing points. Here's a summary of some of the usual questions I've encountered as an interviewer and interviewed by someone else.
2. Formal introduction
How do you highlight that you are a person who has a good understanding of the logic of regression? That is to generalize it in one sentence. The logic regression assumption data obeys the Bernoulli distribution, by using the method of maximal likelihood function, the gradient descent is used to solve the parameters to classify the data two.
It actually contains 5 points. 1: Logical regression hypothesis, 2: Loss function of logistic regression, 3: Method of Logistic regression, 4: Purpose of Logistic regression, 5: How to classify logical regression. These questions are about assessing your basic understanding of logistic regression. The basic assumption of logistic regression is that any model has its own hypothesis, and under this assumption the model is applicable. The first basic assumption of logistic regression is to assume that the data obeys Bernoulli distribution. A simple example of Bernoulli's distribution is the flip of a coin, the probability of being thrown as positive is PP, and the probability of being tossed in negative is 1−p1−p. In the model of logistic regression, it is assumed that hθ (x) hθ (x) is a positive probability for the sample, and that 1−hθ (x) 1−hθ (x) is the probability of a negative sample. Then the entire model can be described as hθ (x;θ) =phθ (x;θ) =p The second assumption of logical regression is that the probability that the sample is positive is p=11+e−θtxp=11+e−θtx so the final form of the logical regression hθ (x;θ) =11+e−θtx