Rank and/or quantile transform

[mbostock]

It’d nice to have a map transform that converts quantitative values into ranks (1, 2, 3, …) or the equivalent p-quantile in [0,1]. This would be an alternative to the quantile scale.

[Fil]

Should this be part of d3-array?
https://observablehq.com/d/c1eb24e6680046d7

[mbostock]

I figured we would build it and see if it’s useful before we decide.

One nuance on implementation: if the same value is repeated, then they should have the same rank, and ranks should be skipped. So for example [3, 1, 3, 4, 2, 0] is mapped to [4, 2, 4, 6, 3, 1] (assuming 1-based ranking). Also I think all non-comparable values would need to be replaced with NaN (or undefined?).

[mbostock]

Also if we use d3.sort we will need to fix d3/d3-array#217 such that all undefined values (null, NaN, and undefined) are moved to the end.

[Fil]

I've added a "same value" test (with a test on ! a <= b to make sure two Date objects representing the same date have the same rank).

I'm not sold on starting at 1. I see the cosmetic appeal, but it makes this slightly more difficult to, for example, use the rank to lookup a value in another array "🥇🥈🥉"[rank].

[mbostock]

I’m not convinced we want 1-based ranking either. That was just my first thought… we can start with 0-based if you like.

[Fil]

The inherent difficulty of writing this "simple" function makes me think it should be in d3-array. And it's already nice to use in Plot:

[Fil]

pull-request for d3.rank d3/d3-array#228

[mbostock]

Added in #570.