Cluster contours / segmentation

[Fil]

For a map / scatterplot where the value is categorical, it would be nice to derive contours for each category. A way to do this might be to:

1. apply a spatial interpolator to obtain a grid
2. for each category c, contour the function value === c with a threshold of 0.5 (or 0.7 to tighten it a bit)
3. done

(Also applies to quantitative data, by determining categorical bands.)

There are possibly better segmentation algorithms to consider.

This is currently possible to do (but very inefficiently) by running one contour mark for each category, with a value = 1 for points that belong to the category, and -1 for points that don't.