L1, L2, cross-validation, learning curves, and deciding what to do next from evidence rather than instinct.
8h · 7 lessons · 1 challenge
Before this: Classification & Logistic Regression
"The model underfits, so what do you do?" is the most common diagnostic question in an ML interview, and the answer is a procedure, not a guess. This module teaches that procedure: learning curves, validation strategy, and the specific action each diagnosis implies.
Regularization is derived as a prior, which connects it back to the probability track and forward to weight decay in deep learning.
Training and validation curves, one prescription: more data, more capacity, more regularization, better features, or stop. Fifteen cases, and the game shows you the outcome of your choice.
Worth 400 XP.
