colour/fill Delaunay plot similar to this voronoi plot in R

Question

using the iris dataset as an example,

I am trying to create plots for Voronoi and Delaunay partitions.

I can do this just fine for Voronoi PLots using ggplot and ggforce packages

#voronoi plot with filled polygons
ggplot(iris, aes(Sepal.Length, Sepal.Width)) +
  geom_voronoi_tile(aes(fill = Petal.Length))

Now I would like to do a similar plot but with Delaunay triangulation, where the triangles are also filled like the polygons in the Voronoi plot above, however, I can't work out how to fill/color the triangles by a third variable

# Delaunay triangles
ggplot(iris, aes(Sepal.Length, Sepal.Width)) +
  geom_delaunay_tile(aes(fill = Petal.Length))

I think this is a bug. The data created in `StatDelaunayTile$compute_group` only has four columns - no fill or PANEL column - whereas the data created with `StatVoronoiTile$compute_group` has those columns. I guess somehow StatDelaunayTile does not inherit aesthetics in the same way and I am not sure if this is desired. — tjebo, Dec 26 '22 at 16:58
@tjebo yea I saw this too, but I have looked to other ways of plotting this without relying on ggforce, but could not yet find an alterative solution that works either — Myriad, Dec 26 '22 at 17:12
@tjebo thanks I did also file an issue related to this on ggforce GitHub page a while ago, but did not get an answer https://github.com/thomasp85/ggforce/issues/292 — Myriad, Dec 26 '22 at 17:15
I guess Thomas will be super busy with other stuff. I'm sure there must be a fairly straight forward fix as I feel this is related with required/default aesthetics that are set within the Stat and Geom. These are things that have very often cause me big headaches, and I'm usually hoping for wizards like AllanCameron or Teunbrand to show me where the problem is — tjebo, Dec 26 '22 at 17:19
@tjebo You are too kind. Actually, the method I have sketched out below should be easy enough to incorporate into ggforce. — Allan Cameron, Dec 26 '22 at 17:22
@AllanCameron thanks so much for this, would it be possible to instead fill the triangle with a gradient that presents the color of each vertex, say if I had a color value for each vertice — Myriad, Dec 26 '22 at 17:33

Allan Cameron · Accepted Answer · 2022-12-26T17:55:27.227

In Voronoi tiling there is a 1:1 correspondence between the points and the tiles, so one can easily map a fill aesthetic to the tiles. Each row in your data frame will have its own tile and therefore can specify a fill, an alpha, etc.

This is not the case in Delaunay tiling, where each row in your data frame describes a single vertex, and each tile therefore requires 3 rows of data to describe. These 3 rows have different values of Petal Length, and there is not a 1:1 correspondence between rows and tiles.

The obvious option is to average the Petal Length of the three vertices, but it seems that ggforce does not have an option to interpolate the vertices this way. However, we can use the deldir package in the same way ggforce does to achieve this ourselves:

library(ggplot2)
library(deldir)

tri <- triang.list(deldir(iris$Sepal.Length, iris$Sepal.Width))
do.call(rbind, lapply(seq_along(tri), 
                      function(x) {
  data.frame(Sepal.Length = tri[[x]]$x, Sepal.Width = tri[[x]]$y, 
             Petal.Length = mean(iris$Petal.Length[tri[[x]]$ptNum]),
             tri = x)
  })) |>
  ggplot(aes(Sepal.Length, Sepal.Width)) +
  geom_polygon(aes(fill = Petal.Length, group = tri))

Additional

To gradient fill your triangles, probably best to draw blank triangles over a 2D interpolation:

library(interp)
library(ggforce)

interp(iris$Sepal.Length, iris$Sepal.Width, iris$Petal.Length, 
              duplicate = "mean", nx = 1000, ny = 1000) |>
  interp2xyz() |>
  as.data.frame() |>
  setNames(names(iris)[1:3]) |>
  ggplot(aes(Sepal.Length, Sepal.Width)) +
  geom_raster(aes(fill = Petal.Length)) +
  scale_fill_continuous(na.value = NA) +
  geom_delaunay_tile(data = iris, fill = NA, color = "#00000050")

colour/fill Delaunay plot similar to this voronoi plot in R

1 Answers1