Does this rule still work
on data it has never seen?
/stat asks whether a difference is chance. These twenty lessons ask something else — does it hold on data you have not seen?
The standard of success is different, and so are the ways you get fooled. You can produce AUC 0.9 from data where y has no relationship to X — just select the features first. We turn the handle on each trap, building logistic regression, trees, PCA and ROC by hand in numpy.
How you'll learn
Split the data and never touch one side. Almost all twenty lessons are about protecting that single device.
No scikit-learn. Logistic regression, trees, PCA and ROC — ten lines of numpy each.
We manufacture AUC 0.9 from data with no signal at all — just select the features first.
/stat handles chance, this course generalisation, /epi causation. None substitutes for another.