Regression

• A really simple form of regression is linear regression.
• With 2D data, all you are doing is building a best fit line for the data. (pg. 729)
• The best fit line is (traditionally) the one that minimises the mean square error. That is, it fits the data points such that the distance from the line of all the points is minimised.
• The line is y = mx +c; Russell and Norvig use w1 = m and w0 = c.
• w1 = (N(sum(xjyj) - (sum xj)(sum yj)))/(N (sum xj^2)- (sum (xj))^2)
• w0 = (sum yj - w1(sum xj))/N (Yikes does that work?)
• You can also do regression to curves, normal distributions and other functions.
• The key here is a good fit. That is small error.