Topic 1's closing lesson introduced underfitting and overfitting informally, through a picture: a straight line failing to capture curvature, and a wiggly high-degree polynomial chasing noise. This note gives that picture a name and a precise mathematical decomposition — bias and variance are not vague intuitions but two distinct, additive contributors to expected test error, and understanding which one dominates a given model's failure is the single most useful diagnostic in this entire unit.
In this lesson
- Learning objectives
- Generalisation error, formalised
- Bias and variance, defined
- The trade-off and the U-shaped curve
- Worked example
Before you start
- Overview: Learning Theory and Generalisation
- Locally Weighted Regression and a First Look at Bias and Variance
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Financial Machine Learning syllabus.