Performance

Loren Petrich’s lib uses an algorithm that seems to be in O(n^2). It is unpractical for n > 1000. (d3-voronoi runs in O(n) log(n).)

See http://bl.ocks.org/Fil/704bdbac80ab2e5d741d5fa3662bdf82

Three possibilities:

• 1. “I’m sorry, Dave, I’m afraid I can’t do that”: let geoVoronoi() log a warning if n > 500 (10s on my computer) and return an error if n > 2000 (2.5 min).
• 1. Switch to an approximate algorithm when n > 500 (If we accept bugs around the poles, we can piggy-back on d3-voronoi by using a specific cylindric projection.)
• 1. Find and implement a better algorithm.

[Fil]

Here’s a notebook that seems to work, built on top of d3-delaunay. https://beta.observablehq.com/@fil/geo-delaunay

Computations are several orders of magnitude faster than the current d3-geo-voronoi.

(The notebook is not that fast, because rendering is slow.)

[mbostock]

I added a mesh computation to my notebook, too. Edit: Ah, I see you’d already done this, too, and that I can simplify my code. 👍