Lesson 14 — Mean squared error

Mean squared error: punish big misses

For number predictions, the standard error measure is MSE (mean squared error): for each example square the difference between prediction and truth, then average. Squaring makes every error positive and makes one big miss count much more than several small ones.

# preds [2, 4], truth [3, 4] -> ((2-3)**2 + 0) / 2 = 0.5

Lower MSE is better; 0 means perfect.

Why it matters: MSE is the "loss" that most regression models minimise during training, and the number you use to compare two models.