stat_density_2d() and geom_density_2d() appear bugged/produce weird output

[AEgit]

Some changes introduced between ggplot2 version 3.2.1 and version 3.3.1 (I did not check version 3.3.0, but I reckon it has the same problem) have lead to weird outputs when using stat_density_2d() and geom_density_2d().

Here a minimum reproducible example:

1. Dataset:
   problem.df.RData.zip

2. Code:

# #Load ggplot2
library(ggplot2)

# #Load dataset
# setwd(readClipboard())
# save(problem.df, file = "problem.df.RData.zip")
load(file = "problem.df.RData.zip")

# #Recreate problem
problem.facet <- ggplot(data = problem.df, aes(X, Y))
problem.facet <- problem.facet + geom_point(color = "lightgray", alpha = 0.1)
# #If we use geom_density_2d we already see a difference between different ggplot2 versions
problem.facet + geom_density_2d(aes(colour = Facet.Groups)) + facet_grid(. ~ Facet.Groups)

# #If we use stat_density_2d() different ggplot2 versions result in different plots which look even weirder
problem.facet <- problem.facet + stat_density_2d(aes(fill = stat(nlevel)), geom = "polygon") +
  facet_grid(. ~ Facet.Groups) + scale_fill_viridis_c()
problem.facet
# #Check the differences in "Group3", "Group4", and the "NA" column
# #Based on visual inspection of the data (see also the geom_density_2d() plot) the behavior of
# #ggplot2 version 3.2.1 should be the correct one

Using the same dataset and the same code, ggplot2 version 3.2.1 produces the following two figures:
ggplot2_321_geom_density_2d.pdf
ggplot2_321_stat_density_2d.pdf

Using the same dataset and the same code, ggplot2 version 3.3.1 produces the following two figures:
ggplot2_331_geom_density_2d.pdf
ggplot2_331_stat_density_2d.pdf

Using the same dataset and the same code, ggplot2 version 3.3.2 produces the following two figures:
ggplot2_332_geom_density_2d.pdf
ggplot2_332_stat_density_2d.pdf

Notice, how weird the colouring/breaks in the newer ggplot2 versions look like ("Group3", "Group4", and "NA" column). Also note, that the left lower peak in "Group4" appears to have disappeared in the newer ggplot2 versions using stat_density_2d() (it is still visible, however, if we use geom_density_2d()). Since the issue appears in both version 3.3.1 and 3.3.2, it is probably not connected to these (solved) problems (?):
#3976
#4004

Is this intended behaviour? To me the output looks like either the result of changed defaults (I know, that version 3.3.0 changed quite a bit here relative to version 3.2.1) or a bug.

[clauswilke]

geom_density_2d() and stat_density_2d() were essentially broken before 3.3.2, so I would consider the old behavior wrong. Please read through the current documentation, specifically the new argument contour_var, and go over the examples provided.

If after going through all this material you still think there is a problem, please submit a new issue, using an existing data set in your reprex (e.g. diamonds as in the documentation) and please prepare your reprex using the reprex package.

[AEgit]

@clauswilke I will try to follow your advice. Note, however, that I am not sure, whether I will be able to reproduce the issue using an existing dataset in reprex (which I originally planned), since the problem appears specific to datasets. In the above example the plots for the "Group1" and "Group2" columns stay the same across all versions, while the "Group3", "Group4", and "NA" column change (see attached plots). This indicates, that the issue will only appear under very specific circumstances (the uploaded dataset + code allows to reproduce all figures I posted above, assuming that the respective ggplot2 library is installed).

But I will see what I can do and then come back to you.

[AEgit]

@clauswilke Here you can find a standard reprex of the same issue:
#4114

Hope it helps!