Linear Regression

Linear Regression

PREVIOUS: Introduction

Linear Regression

For linear regression, we want to find Y = WX + b that gives us the lowest L value.

Y is our Model
W is a matrix of weights
X is a matrix of our inputs
B is our bias
L = (Y - y)^2, with Y being our desired output and y being our predicted output.

Ordinary Least Squares

Data and Application
Model
Optimization Problem
Optimization Algorithm

$m=\frac{\sum(x - \bar{x})(y-\bar{y})}{\sum(x - \bar{x})^2}$ $b = \bar{y} - m\bar{x}$

See link

Polynomial Features

What about when the curve isn’t linear? Polynomial curve? Taylorization of any function? Overfitting?