Optimization:
Linear Regression
MATH2070: Numerical Methods in Scientific Computing I
Location: http://people.sc.fsu.edu/∼jburkardt/classes/math2070 2019/optimization linear regression/optimization linear regression.pdf
Find coefficients for an approximating formula to data.
Linear Regression Problem
Find the coefficients in a formula that will minimize the sum-of-squares error to given data.
1 Simple linear regression problem
In a previous class, we looked at a problem in which we wanted to fit n = 23 Ford Escort mileage (x) and
price (y) data values with a simple linear formula:
y = b + m ∗ x
We first normalized our data so that 0 ≤ x, y ≤ 1. Then we used gradient descent to estimate the values of
b and m:
1 bm0 = [ 0 . 5 , 0 . 0 ] ;
2 r = 0 . 0 1 ;
3 d x tol = 0 . 0 0 1 ;
4 d f t o l = 0 . 0 0 1 ;
5 itmax = 10 0 0 ;
6 [ bm, i t ] = g r a d i e n t d e s c e n t 2 ( @( x ) f o r d f ( x ) , @( x ) f o r d d f ( x ) , bm0 , . . .
7 r , dxt o l , d f t o l , itmax ) ;
Listing 1: Calling gradient descent2 for the Ford data.
This calculation came up with values of b and m for the formula:
y = 1.04802 − 0.796532 ∗ x
1