The third part of this notebook takes you through a tour of the classical way of assessing significance through (t-)tests. In machine learning applications with a lot (sometimes millions) of features, these tests have limited utility. If individual feature significances are to add up to something meaningful over these numerous features, each significance has to be really strong (its reciprocal must be two orders of magnitude more than the number of features). Yet, the reasoning is very elegant, and will help you understand linear regression a lot better.
This section has three problems, please submit them on Laulima.