Theory of Intelligence
MACHINE LEARNING•Supervised Learning•Linear Regression•3.5 Learning Curves

Introduction

Plot training and validation error as the training set grows, and use the shape of the curves to tell underfitting from overfitting — and whether more data will help.

timer1 min read•format_list_bulletedConcept 1 of 3•forumLesson Discussion

Introduction

Cross-validation tells you how good a model is. Learning curves tell you why it's not better, and what to do about it — in particular, the expensive question: "Will collecting more data help?"

A learning curve is like tracking a student's scores as they study more practice questions:

  • If both practice and exam scores stay low however much they study, the student needs a better method (more capable model), not more questions.
  • If practice scores are perfect but exam scores lag far behind, they're memorising — more varied practice questions (more data) and less cramming (regularization) will help.

Discuss theoretical concepts, mathematical formulas, and architectural foundations.

Supports inline code with `backticks`
No comments yet. Be the first to start the discussion!
arrow_backPrevious Lesson
3.4 Polynomial Regression