neighborhoodHit

function neighborhoodHit(
   ldIn,
   labels,
   k?,
   knn?): MetricResult;

Defined in: metrics/labelled.ts:88

Share of each point’s visual neighbours that carry its own label.

Answers “if I look at a point and the ones drawn around it, do they agree?” — close to how a viewer actually reads a coloured scatterplot.

• Needs: projection only. Labels required.
• Range: [0, 1], higher is better. The chance level is the largest class’s share of the data, not 0.
• Cost: O(N²·D).

Parameters

Parameter	Type	Default value
ldIn	PointsInput	undefined
labels	readonly unknown[]	undefined
k	number	20
knn?	Int32Array<ArrayBufferLike>	undefined

Returns

MetricResult

See

Paulovich et al., IEEE TVCG 14 (2008) https://doi.org/10.1109/TVCG.2007.70443

Example

import { neighborhoodHit } from "@saehrimnir/sickle";

// Takes raw labels, not a Clusters object.
neighborhoodHit(projection, labels, 20).value;  // 1 — every neighbour shares a class
// Chance here is the largest class's share, not 0.