DSAverse

Initializing Sorting Algorithms...

Sorting

Trees

Graphs

Preparing interactive visualizations...

Back to Machine Learning

k-Nearest Neighbours

Classify a point by asking the k training points closest to it. Watch the vote, the decision map for k = 1 and k = 39, what changing the units does to “closest”, and why distances stop meaning much in a thousand dimensions.

Builds onTrain, Validation and Test Norms and Unit Vectors

Every distance and vote computed in your browser · seeded two-moons data

Act 1 of 6 · Distance · k = 1step 1 of 21

X[n, features] =[40, 2]distances[n] =[40]

Speed

Forty labelled points: blue and orange. k-nearest neighbours does no fitting at all — it keeps the points. To classify a new point (the star), it starts by measuring how far the star is from every one of them.

\hat y(q) = \text{majority label among the } k \text{ points nearest } q, \quad k = 1

k-Nearest Neighbours

Builds onTrain, Validation and Test Norms and Unit Vectors

Every distance and vote computed in your browser · seeded two-moons data

Act 1 of 6 · Distance · k = 1step 1 of 21

X[n, features] =[40, 2]distances[n] =[40]

Speed

\hat y(q) = \text{majority label among the } k \text{ points nearest } q, \quad k = 1

DSAverse

k-Nearest Neighbours

See also

k-Nearest Neighbours

See also