Theory of Intelligence
MACHINE LEARNING•Classic Algorithms•Ensemble Learning•3.6 Stacking

Introduction

Instead of a simple vote, train a model (a blender) to learn the best way to combine the predictions of several other models.

timer1 min read•format_list_bulletedConcept 1 of 3•forumLesson Discussion

Introduction

A voting classifier combines models with a fixed rule: majority vote or average. But maybe one model is very good for certain kinds of examples and another for others. Maybe one is always slightly too optimistic. A fixed vote can't learn any of that.

Stacking (short for stacked generalisation) asks: why not train a model to learn how to combine them?

Picture a manager with a team of advisors — an economist, an engineer and a lawyer. Over time, the manager learns whose advice to trust for which kind of question. The advisors are the base models; the manager is the blender (also called the meta-learner).

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
3.5 Gradient Boosting