机器学习笔记（Washington University）- Regression Specialization-week one

1. Convex and concave functions

Concave is the upside-down of the convex function

and convex is a bow-shaped function

 

2. Stepsize

common choice:

as the iteration goes, we will decrease the stepsize against a fixed stepsize

alpha = alpha/t or alpha = alpha / (t ^0.5

and when the deriative is smaller than a set threshlod, we can stop the algorithm

 

3. Approach 1

set gradient=0, so we can solve for w0 and w1,

using those two equations:

and we can get that:

 

4. High leverage points

High leverage points mean that it ia at an extreme x value where there are no other observation.

And this point has the potential to change the least squares line since the center of x mass is heavily influenced.

An influential observation is one where the removal of the point siginificantly changes the fit.'

 

5. Asymmetric cost functions

This means the errors are not weighed equally between these two types of mistakes (too high and too low estimated price).