Playground
Every overlap-removal method in the library, over datasets chosen to show where each one shines and where it struggles. Points are coloured by where they started, so you can see how much of the original structure survives.
…
Colour is the point
Section titled “Colour is the point”Each point is tinted by its original position — hue by direction from the centre, depth by distance. A layout that keeps neighbours together keeps the colours smooth. Scrambled colour means the spatial structure was lost, which is the one thing no summary number shows you directly.
Where they differ
Section titled “Where they differ”Try Elongated: Hilbert must pad to a square and squashes it, Gilbert matches the shape. Then switch to NMap-squared on the same data — its cells stretch with the region rather than staying square, which is the difference between equalising area and equalising shape.
Try Integer lattice or Single column: NMap refuses them outright, GridFit takes them
in its stride. Try Heavy duplicates with CMDS: coincident points have an ideal distance of
zero, so the weight 1/d² is infinite and every position comes back non-finite.
Reading the numbers
Section titled “Reading the numbers”Distinct cells turns red the moment two points share one — that is the property every method here is supposed to guarantee, and the ones that miss it are named on their own pages.
Mean displacement is in cell pitches, not pixels or fractions of the plot, so it means the same thing across methods that answer in different units. The histogram beneath the plot shows the whole distribution: a method can post a respectable mean and still fling a handful of points clear across the layout, and that tail is what a reader actually notices.