spearmanRho

function spearmanRho(
   hdIn,
   ldIn,
   opts?): MetricResult;

Defined in: metrics/correlation.ts:97

Rank correlation between original and projected distances — Shepard goodness.

Asks only whether distances kept their order, so a projection that stretches distances non-linearly but consistently still scores 1. That makes it a fairer global measure than stress for methods that deliberately warp scale.

• Needs: high-dimensional data and projection. No labels.
• Range: [-1, 1], higher is better.
• Cost: O(N² log N) time and O(N²) memory — it ranks every pair, so unlike most measures here it materialises them. maxPairs guards the allocation.

Parameters

Parameter	Type
hdIn	PointsInput
ldIn	PointsInput
opts	SpearmanOptions

Returns

MetricResult

See

Shepard, Psychometrika 27 (1962) https://doi.org/10.1007/BF02289630

Example

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

// Takes the points directly — it needs every pair ranked, so it does not read
// off `analyze` and it materialises the full N×N matrix.
spearmanRho(data, projection).value;  // 0.988

// Guarded by `maxPairs` (default 60e6, counting n²), so it refuses past
// n ≈ 7 745 rather than attempting the allocation.
spearmanRho(data, projection, { maxPairs: 2e8 });