机器学习课程笔记（二）

机器学习课程笔记（一）

第四课

Multivariate linear regression:

Matrix:

Gradient descent:

Feature scaling:

Between -3 and 3 or -1/3 and 1/3

Mean normalization:

purpose: making gadient descent run faster

Conform:

cost function J(θ) must decrease after each test

Error correction

1. incesase:

Problem: Learning rate is too large or code is wrong
2. circulation:

Problem: Learning rate is too large

Polynomial regression:

feature scaling is extremely important

set different feature value to fit the data in order to get a better model

Normal equation method

elample:

set derivative equal to zero

solution:

To make clear the derivation process, refer to this piece of passage — Click Here
If you use nomal equation method, there is no need to use feature sacling

Comparation