liveBook · Manning

Mode

As explained in Machine Learning System Design

Majority vote is an aggregation method that selects the label or decision supported by more than half of the votes.

When labelers disagree about a binary classification, majority voting combines their judgments into one label. It can improve accuracy over a single labeler, but correlated errors—such as several people struggling with the same noisy sample—can limit the gain. section 6.2.4

Three labelers

If each of three labelers is correct 70% of the time, a majority vote is correct 78.4% of the time. One cost-conscious rule accepts two agreeing votes; if they disagree, it adds three more votes and uses the majority. section 6.2.4

How Machine Learning System Design covers it

How deep it goes
Works through a three-labeler accuracy calculation and an adaptive voting rule, then cautions that labeler errors may be correlated.
Examples
Calculates 78.4% accuracy for three labelers who are each correct 70% of the time.
What you can do after
Apply it to binary labeling, accepting two agreeing votes or collecting three more when they disagree.
How it connects
Connects labeling to binary classification, pairwise comparison, and Condorcet’s jury theorem.
Cover of Machine Learning System Design
liveBookmanning.com
Read Machine Learning System Design on liveBook Get the book at manning.com

Also explained in

Compare how 8 books explain mode →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.