flekschas / regl-scatterplot

Scalable WebGL-based scatter plot library build with Regl
https://flekschas.github.io/regl-scatterplot/
MIT License
194 stars 24 forks source link

feat: KDBush worker and spatial index exposure #178

Closed flekschas closed 5 months ago

flekschas commented 5 months ago

This PR introduces worker-based spatial indexing and exposes the index for reuse.

Description

What was changed in this pull request?

Why is it necessary?

Fixes #174

tl/dr: Exposing and (re)using spatial indices is useful when performance is key as it allows to drastically reduce the draw time by skipping the expensive spatial indexing. For instance, imagine you only want to change the z/w coordinates that are used for defining the point color, opacity, or size. Previously, every time you called scatterplot.draw(points) would trigger a re-indexing of the x/y point coordinates even though the did not change. Similarly, one could now precompute the spatial index and load it upfront. In both cases, scatterplot.draw(points, { spatialIndex }) can be dramatically as the spatial indexing is skipped.

Caution: Please be advised that when you pass a spatial index to the draw call, regl-scatterplot assumes the index conforms with the point data. Regl-scatterplot does not actual verify that the two conform! So only use this feature if you know what you're doing!

Example

In the example below you can see how one can reuse the initial spatial index to save redrawing time upon follow up draw calls where only the z/w coordinates change.

const x = (length) => Array.from({ length }, () => -1 + Math.random() * 2);
const y = (length) => Array.from({ length }, () => -1 + Math.random() * 2);
const z = (length) => Array.from({ length }, () => Math.round(Math.random()));
const w = (length) => Array.from({ length }, () => Math.random());

const numPoints = 1000000;
const points = {
  x: x(numPoints),
  y: y(numPoints),
  z: z(numPoints),
  w: w(numPoints),
};

const scatterplot = createScatterplot({ ... });
scatterplot.draw(initialPoints).then(() => {
  // After the initial draw, we retrieve and save the KDBush spatial index.
  const spatialIndex = scatterplot.get('spatialIndex');
  setInterval(() => {
    // Update Z and W values
    points.z = z(numPoints);
    points.w = w(numPoints);

    // We redraw the scatter plot with the updates points. Importantly, since
    // the x/y coordinates remain unchanged we pass in the saved spatial index
    // to avoid having to re-index the points.
    scatterplot.draw(points, { spatialIndex });
  }, 2000);
})

In the video you can see that this leads to a roughly 6x performance increase in Firefox when rendering one million points.

https://github.com/flekschas/regl-scatterplot/assets/932103/d7200b11-97cc-4346-bcf7-d68c3755b0d6

Checklist