Guide colorsteps creates an ugly, large NA box

[clauswilke]

Technically another issue against my PR #3864. I made this figure yesterday and I thought the guide looks ugly. Why is the box representing NA so large? Should NA be shown at all by default? Should it maybe be shown with a little gap? Is this correct at all?

library(ggplot2)

set.seed(4393)
dsmall <- diamonds[sample(nrow(diamonds), 1000), ]

ggplot(dsmall, aes(x, y)) +
  geom_density_2d_filled(aes(fill = after_stat(level_high))) +
  scale_fill_viridis_c(guide = "colorsteps")

^{Created on 2020-03-09 by the reprex package (v0.3.0)}

[clauswilke]

For comparison, this is what it looks like with guide = "legend".

library(ggplot2)

set.seed(4393)
dsmall <- diamonds[sample(nrow(diamonds), 1000), ]

ggplot(dsmall, aes(x, y)) +
  geom_density_2d_filled(aes(fill = after_stat(level_high))) +
  scale_fill_viridis_c(guide = "legend")

^{Created on 2020-03-09 by the reprex package (v0.3.0)}

[thomasp85]

I’m pretty sure this is due to the recent fix to make NA work with binning. In my point of view, NA shouldn’t be shown in the legend at all. I’ll look into a fix

[yutannihilation]

Hmm, this might be my oversight, but the CRAN version seems to behave the same... I'm a bit confused. Is it intended to use *_viridis_c(), not *_viridis_b(), for binned scales?

library(tidyverse)

sessioninfo::package_info("ggplot2", dependencies = FALSE)
#>  package * version date       lib source        
#>  ggplot2 * 3.3.0   2020-03-05 [1] CRAN (R 3.6.3)
#> 
#> [1] /home/yutani/R/x86_64-pc-linux-gnu-library/3.6
#> [2] /usr/lib/R/library

data(airquality)

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

ggplot(airQ, aes(x = Solar.R, y = Temp)) +
  stat_density_2d(aes(fill = after_stat(level)), geom = "polygon") +
  scale_fill_viridis_c(guide = "coloursteps")

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

library(ggplot2)

set.seed(4393)
dsmall <- diamonds[sample(nrow(diamonds), 1000), ]

ggplot(dsmall, aes(x, y)) +
  geom_density_2d_filled(aes(fill = after_stat(level_high))) +
  scale_fill_viridis_b(guide = "colorsteps")

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

[clauswilke]

It looks to me like with the binned scale, we're binning twice, once to make the contours and once to assign the colors. Notice how we have more contour bands than colors.

library(ggplot2)

set.seed(4393)
dsmall <- diamonds[sample(nrow(diamonds), 1000), ]

ggplot(dsmall, aes(x, y)) +
  geom_density_2d_filled(aes(fill = after_stat(level_high)), color = "white") +
  scale_fill_viridis_b(guide = "colorsteps")

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

[yutannihilation]

Thanks, I see.

[thomasp85]

Yeah. There are some stats that provide the binning themselves because they need access to the unbinned data. These should not be mixed with the binning scales