Overfitting and Underfitting

71 min
Block 9 — Generalization
Objective
define overfitting and underfitting by the relation between two risks — the empirical risk on the fitting sample and the true risk under the distribution — rather than by the informal slogan "the model learned the noise"; connect both conditions to the capacity of the hypothesis space and to its effective, not nominal, size; produce the classic U-shaped test-error curve from real measurements on three different capacity knobs; build learning curves and read high bias against high variance from their shape; separate the validation curve from the learning curve, which answer different questions; demonstrate memorization by fitting random labels; place the interpolation threshold and double descent as a measured caveat to the classic U; expose the slower failure in which the validation set itself is overfitted by repeated selection; recognize underfitting hidden behind a metric that has saturated; and read the signature of each condition off a table of train, validation and test numbers.
Estimated duration
70 minutes
Prerequisites
chapters 026 (train/validation/test split), 028 (data leakage), 029 (fit, predict, predict_proba) and 030 (error, loss and metric)
Associated quizzes
031.1-quiz-generalization.md to 031.7-quiz-capacity-and-regularization.md

Preview — the rest of the lesson is for enrolled readers.

Already enrolled with a code?

Your access is tied to your account, not to this link. Sign in with the same email you used in class: your course is waiting, no need to enter the code again.

Sign inNo account yet? Create one
This lesson is part of the “Generalization, Overfitting and Underfitting” module

The first modules of the course are open to everyone. For the rest you have three options: buy this course once and for all, subscribe, or enter the code handed out in class.

Are you a student on this course?

The code is tied to your account: sign in or create an account and it will be applied automatically when you come back.

No account yet? Create one