ggMarginal ignores expand argument

[wpetry]

In ggMarginal, the marginal plots use expand = TRUE even if the user has specified p to have expand = FALSE. This leads to a misalignment of the data between the main plot and margins.

[daattali]

Can you please add a small fully reproducible example code

[wpetry]

Sure thing, @daattali

Load packages & generate data

library(tidyverse)
library(ggExtra)

theme_set(theme_bw() + theme(panel.grid = element_blank()))

set.seed(503234)
dat <- tibble(x = rbeta(5000, shape1 = 5, shape2 = 1),
              y = exp(rnorm(5000, mean = 0, sd = 2)))

Using the default expand = TRUE works—the density plots cut off in line with the data. This is expected behavior.

p <- ggplot(dat, aes(x = x, y = y))+
  geom_point()+
  scale_y_log10()+
  coord_cartesian(xlim = c(0, 1), expand = TRUE)
ggMarginal(p, type = "density")

Using expand = FALSE misaligns the density plots. The lines should extend the whole axis extents, and this is particularly visible in the apparent zero density around x=1. This is unexpected behavior.

q <- ggplot(dat, aes(x = x, y = y))+
  geom_point()+
  scale_y_log10()+
  coord_cartesian(xlim = c(0, 1), expand = FALSE)
ggMarginal(q, type = "density")

I was able to fix it very crudely by adding a logical argument called expand to the function and replacing top[["coordinates"]][["expand"]] <- expand and right[["coordinates"]][["expand"]] <- expand. This clearly isn't the right way to do it, otherwise I'd send in a PR. Ideally an automated check of the original plot p would be run (including custom adjustments, e.g., scale_x_continuous(expand = c(0.5, 1)), and the top and right marginal plots would be adjusted accordingly.

Thanks for maintaining this great package.

[daattali]

Great, thanks for the detailed report and examples. I'm on mobile and rushing, but is this related/duplicate of #128 ?

[wpetry]

They're very closely related for sure (sorry for not referencing #128), but I think this may be a separate (or subset) of that issue that would be easier to fix.

[daattali]

Thanks for reporting. Feel free to take a look and tackle it in the proper way if you find some time!

…

On Fri, Jun 21, 2019, 17:20 Will Petry ***@***.*** wrote: They're very closely related for sure (sorry for not referencing #128 <#128>), but I think this may be a separate (or subset) of that issue that would be easier to fix. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#145?email_source=notifications&email_token=AAHIQFDCR2LYOQ34CG73VH3P3VA2XA5CNFSM4H2T5XHKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODYJTUSA#issuecomment-504576584>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAHIQFDODCDF3OYQQ445RWLP3VA2XANCNFSM4H2T5XHA> .

[wpetry]

I thought I had a solution for the CRAN version, but I see that the GH version has switched to R6. I'm afraid this is a bit out of my depth.

[crew102]

I'd be happy to integrate your solution into the current codas of ggExtra, so long as the fix you're talking about isn't too hairy.

[crew102]

Hi @wpetry , I looked into this and it basically amounts to #128. ggMarginal() currently doesn't handle the cartesian coordinate system well (i.e., it doesn't work well for plots that were built using the coord_cartesian() function). The reason why it seemed like there was a specific issue with the expand argument is b/c the coord_cartesian() call in your example didn't actually change the look of the plot. If you change the values for xlim in your example, you'll see that ggMarginal() fails to account for the cartesian coordinate system, even when expand = TRUE. In other words, the problem isn't the expand argument per se, but rather the inability of ggMarginal() to account for coord_cartesian().

library(ggplot2)
library(ggExtra)

set.seed(503234)
dat <- data.frame(
  x = rbeta(5000, shape1 = 5, shape2 = 1),
  y = exp(rnorm(5000, mean = 0, sd = 2))
)

p <- ggplot(dat, aes(x, y)) +
  geom_point() +
  scale_y_log10() +
  coord_cartesian(xlim = c(0, .5), expand = TRUE)

ggMarginal(p)

Created on 2019-06-29 by the reprex package (v0.2.0.9000).

[daattali]

Thanks for looking into it. Closing this as duplicate