Linear Model for classification

1. Linear Model for classification

Solving classification problems we have seen three of the above: linear classification problem, linear regression problem solution classification, logistic regression. Next, look at the differences and connections between the three.

The difference between the three is mainly the difference between the error function:

The same diagram shows the following:

Why is uper bound effective? Theoretical proof:

Three comparison of algorithms to solve classification problems:

2. Stochastic Gradient descent

Use logistic regression to solve the two classification problem: one is the batch gradient descent, need batch processing data, one is the random gradient descent. corresponding to PLA and pocket, respectively

　　

3. Multiiclass via Logistic Regression

3.1 One vs All

　　

3.2 One vs one at a time

is the 22 combination. See the two solutions to the logistic separately. Then the majority vote.

　　

In addition, anything similar to the one that can be obtained from a 01-worth classifier can be used to solve a multi-classification problem.

Linear Model for classification