@vizcrush/transform

Fast typed-array primitives: radix sort, min-max normalization, range filtering. The boring-but-essential pieces that show up in every data pipeline.

Import

import { sortBy, normalize, filterRange } from "@vizcrush/transform";

sortBy(data, keys?, descending?)

Sort a Float64Array in place (or by separate keys) using a radix sort (linear-time on fixed-width floats).

const data = new Float64Array([3.1, 1.4, 1.5, 9.2, 6.5]);
const sorted = await sortBy(data);
// Float64Array [1.4, 1.5, 3.1, 6.5, 9.2]

Sort one array by another's keys:

const values = new Float64Array([10, 20, 30, 40]);
const keys = new Float64Array([3, 1, 4, 2]);

const sorted = await sortBy(values, keys);
// values reordered to match keys ascending: [20, 40, 10, 30]

Descending:

const sorted = await sortBy(data, undefined, /* descending */ true);

Cost: O(n) — radix sort runs in linear time on fixed-width floats. ~3-5× faster than Array.prototype.sort on a million elements.

normalize(data, range?)

Min-max normalize to [0, 1] (default) or a custom range:

const data = new Float64Array([10, 20, 30, 40, 50]);
const normalized = await normalize(data);
// Float64Array [0, 0.25, 0.5, 0.75, 1]

// Custom range:
const scaled = await normalize(data, [-1, 1]);
// Float64Array [-1, -0.5, 0, 0.5, 1]

Useful for:

• Color mapping in heatmaps and density grids
• Feeding ML models that expect normalized inputs
• Coordinate normalization before spatial indexing

filterRange(x, y, xMin, xMax)

Extract the subset of (x, y) pairs where x falls within [xMin, xMax]. Returns separate x and y typed arrays.

const result = await filterRange(x, y, /* xMin */ 1000, /* xMax */ 2000);
// { x: Float64Array, y: Float64Array } for points with 1000 ≤ x ≤ 2000

Use case: viewport zoom. When a user zooms a time-series chart, you don't want to downsample the entire dataset every frame. Filter first, then downsample the filtered slice:

async function onZoom(visibleMin: number, visibleMax: number) {
  const sliced = await filterRange(x, y, visibleMin, visibleMax);
  const display = await lttb(sliced.x, sliced.y, canvas.width);
  chart.update(display.x, display.y);
}

Performance reference

Operation	100K	500K	1M
sortBy	5 ms	28 ms	60 ms
normalize	0.3 ms	1.5 ms	3.0 ms
filterRange (50% viewport)	2 ms	11 ms	19 ms

See also

• @vizcrush/downsample — typically follows a filterRange call
• Algorithms reference