Average Nearest Neighbour Index

A classic, fully interpretable test: compare the observed mean nearest-neighbour distance to what you’d expect if the points were scattered at random.

What it means

R = observed mean NN distance / expected mean NN distance under complete spatial randomness:

• R < 1 — clustered (points closer together than random)
• R ≈ 1 — random
• R > 1 — dispersed (more evenly spaced than random)

A z-score gives significance (Clark & Evans). The primitive returns the index, the z-score, and a plain-language interpretation field.

GPU approach

A composition: the GPU does the O(N²) nearest-neighbour search; ANNI adds the cheap CPU summary (mean, the CSR expectation, the z-score) on top. Nothing new runs on the GPU — it reuses an existing primitive, which is the point.

Usage

import { anniGpu } from 'intraspatial/gpu/spatial/anni';

const r = await anniGpu(xs, ys, { area: width * height });
// { index, zScore, interpretation: 'clustered' | 'random' | 'dispersed', ... }

Status: ✅ implemented and validated (random / clustered / dispersed fixtures) · src/gpu/spatial/anni.ts