Implement filled 2d density contours

[clauswilke]

The missing feature from 3.3.0, as described in #3846.

This is work in progress. Still to do:

• Implement geom_density2d_filled() analogously to geom_contour_filled()
• Usage examples
• Make nlevel work for geom_density2d_filled().
• Unit tests
• News item
• Fully document computed variables
• Improve documentation of stat_contour_filled()

[clauswilke]

Never mind. Apparently this works only from inside a stat.

library(tidyverse)

data(airquality)
airQ <- filter_all(airquality, ~!is.na(.))

ggplot(airQ, aes(x = Solar.R, y = Temp)) +
  geom_density2d_filled(alpha = 0.7) +
  geom_density2d(size = 0.25, color = "black") +
  geom_point(size = 1)

^{Created on 2020-03-07 by the reprex package (v0.3.0)}

[clauswilke]

@thomasp85 How's this?

library(ggplot2)

ggplot(faithfuld, aes(waiting, eruptions, z = density)) + 
  geom_contour_filled() + 
  facet_wrap(~waiting < 70)

set.seed(4393)
dsmall <- diamonds[sample(nrow(diamonds), 1000), ]
d <- ggplot(dsmall, aes(x, y))

d + geom_density_2d_filled() +
  facet_wrap(~cut)

d + geom_density_2d_filled(contour_var = "ndensity") +
  facet_wrap(~cut)

d + geom_density_2d_filled(contour_var = "count") +
  facet_wrap(~cut)

d + geom_density_2d(aes(color = after_stat(level))) + 
  facet_wrap(~cut)

d + geom_density_2d(aes(color = after_stat(level)), contour_var = "ndensity") + 
  facet_wrap(~cut)

d + geom_density_2d(aes(color = after_stat(level)), contour_var = "count") + 
  facet_wrap(~cut)
#> Warning: stat_contour(): Zero contours were generated
#> Warning in min(x): no non-missing arguments to min; returning Inf
#> Warning in max(x): no non-missing arguments to max; returning -Inf

^{Created on 2020-03-10 by the reprex package (v0.3.0)}

[thomasp85]

This looks good. The compute_layer approach was exactly what I had in mind...

It hurts a little that contour_var cannot be an aesthetic, but there is no way around that if it all has to happen in a stat...

Some part of me really wants to make the contouring happen in the geom so that anything that ends up as a raster hightmap after the stat can be contoured... but then we end up with the issues of colour scaling the output of contouring –– no singular best solution it seems

[clauswilke]

Some part of me really wants to make the contouring happen in the geom so that anything that ends up as a raster hightmap after the stat can be contoured... but then we end up with the issues of colour scaling the output of contouring –– no singular best solution it seems

If the geom could get the height data from the stat and the binning information from the scale then it might work. We could always in the future add such geoms in addition to geom_contour(), as I did in ggisoband (which conceptually does what you want, it just doesn't quite work right in all cases).

[clauswilke]

I think this is ready for review now. Closes #3846.

[thomasp85]

Is there any difference between stat_bin2d and the new stat_density_2d() with contour = FALSE. To me they seem superficially equivalent (at least in intend)

[thomasp85]

What is the purpose of this? Do we have any precedence for creating ggproto aliases?

[clauswilke]

This is needed to auto-generate documentation for aesthetics. We probably should do it more consistently across ggplot2. See here:
https://github.com/clauswilke/ggplot2/blob/743bad010638fdccaed4c0cfa0258e0879e1263c/R/geom-density2d.r#L9-L10

[thomasp85]

Ah... that seems bad, but is not really the fault of this PR

[clauswilke]

The relationship between stat_bin2d() and stat_density_2d() is equivalent to stat_bin() and stat_density(). One creates a histogram, by making regular-sized bins and simply counting how many observations fall into each bin. The other does a kernel-density estimation. Both are useful, and they are distinct.

[thomasp85]

You are right. I got bin2d confused

[thomasp85]

LGTM