liveBook · Manning

Normalized mutual information

As explained in Timeless Algorithms, The Foundational Ideas

Normalized mutual information is a normalized measure of the association between cluster assignments and known categories.

Normalized mutual information (NMI) evaluates whether cluster assignments reveal known categories. It measures how much knowing an observation’s cluster reduces uncertainty about its true category, complementing geometric assessments of cluster separation. section 5.3.4

Fruit clusters

If 12 fruits—four apples, four bananas, and four pears—are clustered into three groups, each containing only one fruit category, knowing a fruit’s cluster removes all uncertainty about its category. This illustrates an ideal outcome for NMI. section 5.3.4

How Timeless Algorithms, The Foundational Ideas covers it

How deep it goes
Defines NMI through reduced category uncertainty but does not provide its formula in the available explanation.
Examples
Uses 12 apples, bananas, and pears divided into three category-pure clusters.
What it assumes
Assumes known categories against which cluster assignments can be evaluated.
What you can do after
Evaluate whether cluster assignments reduce uncertainty about known categories.
How it connects
Connects clustering evaluation with entropy, purity, and geometric cluster assessment.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Timeless Algorithms, The Foundational Ideas. The section links point to the parts of the book it draws on.