Add symlog and compound scales

[danielnaab]

The motivation for this work is to produce Matplotlib-style "symlog" scaling, as described in #105.

The PR adds the notion of a "compound scale", a scale that delegates to any number of other scales. Its range and domains are inclusive to each provided scale. The symlog scale is responsible for instantiating the compound scale given a range or domain, with:

Log scale: [-Infinity, -1]
Linear scale: [-1, 1]
Log scale: [1, Infinity]

[curran]

This PR is missing documentation updates.

[danielnaab]

@curran Thanks, just pushed a commit with doc updates.

[Tillsten]

I am a regular user of symlog scales in matplotlib I think this PR is missing that the linear threshold and the scale factor of the scale is changeable. The former limits the linear domain and the latter determines the relative scaling of the log and the linear parts.

[danielnaab]

@Tillsten Right this isn't exactly what Matplotlib supports, it's a more basic implementation. If you have ideas how to integrate more flexibility, I can add to the PR.

[mbostock]

I’ve finally gotten around to implementing arbitrary transforms for scales in #156, to which I’ve added a (non-configurable) symlog transform. It would be pretty easy to make the symlog transform configurable if desired, and you can easily pass in a custom transform to a scale. There’s still some work to do over there, but I’d love feedback if you’re still interested in this feature. Thank you!

[ghost]

@mbostock Can you show me an example for configure a custom range (x) for symlog? I.e.

Log scale: [-Infinity, -x]
Linear scale: [-x, x]
Log scale: [x, Infinity]