Model3: Softmax Regression

Now that we've solved the problem of binary classification, let's talk about the problem of multiple classification

1.Overview of multi-categorical problems

What do we do when binary classification expands into multiple classifications？

2.Softmax regression

Softmax regression is used to solve multi-classification problems and is a generalization of logistic regression.

 example：

In summary:

or： 

 3.Cost function

(1)Cross-entropy loss function

The cross-entropy loss function is also a generalization of multi-class.

(2)The gradient descent method is used to find the minimum

We are able to calculate its minimum value mathematically.

We calculate its gradient：

This gradient is then used to update the parameters.

(3)Regularization

Thus：

4.Neural network with softmax regression

(1)Principle

Outputs a column vector with ten values, a1 to a10, representing the probability of each class. 

(2)TensorFlow code

(3) Improve the code