Plot method for data_tabulate

[etiennebacher]

Close #289

Minimal working version so far (no docs, etc.). It is such a short step to convert the table into a plot that I think we should restrict the number of args to a minimum here. Otherwise it will just be quicker for the users to make a fancy plot by themselves than go through all possible args (e.g the docs of sjPlot::plot_freq() are a bit overwhelming 😅). I have two "important" args in mind for now:

• keep or remove NA? if we keep them we show raw counts and percentage, otherwise only the valid ones
• show or hide counts and percentage on top of bars?

@jmgirard what do you think?

library(see)
library(datawizard)
library(patchwork)

x <- data_tabulate(mtcars, select = "cyl")
plot(x)
#> [[1]]

x <- data_tabulate(mtcars, select = c("cyl", "gear"))
my_plots <- plot(x)
wrap_plots(my_plots)

[bwiernik]

A few comments:

1. we should include error bars by default
2. I think we should apply theme_modern() as the default theme for all plots in see
3. We should relabel "NA" to "(Missing)" and have an argument for the label to use in addition to whether or not to inlcude.

[etiennebacher]

library(see)
library(datawizard)
library(patchwork)

x <- data_tabulate(mtcars, select = "cyl")
plot(x)
#> [[1]]

plot(x, remove_na = TRUE)
#> [[1]]

plot(x, na_label = "Missing values")
#> [[1]]

[bwiernik]

Looking good! Error bars?

[etiennebacher]

@bwiernik did you have something like this in mind? Default is to add error bars (I just took the code from sjPlot::plot_frq())

library(see)
library(datawizard)
library(patchwork)

x <- data_tabulate(mtcars, select = "cyl")

plot(x, remove_na = TRUE)
#> [[1]]

plot(x, remove_na = TRUE, error_bar = FALSE)
#> [[1]]

^{Created on 2023-06-02 with reprex v2.0.2}

[etiennebacher]

Remaining tests failures are not caused by this PR, they are also on main

[strengejacke]

I think this can be merged?

[bwiernik]

I'm making some tweaks

[bwiernik]

okay, I made those tweaks. The major one is to use the Wilson CI for proportions rather than the normal approximation -- this avoids getting CI bounds that escape [0, 1] or intervals collapsing to nothing at p = 0 or 1

[bwiernik]

this is ready to merge assuming checks still pass

[bwiernik]

@IndrajeetPatil do you know why the R-CMD-check-strict on ubuntu is taking forever to load all of the dependencies?

[strengejacke]

The automated checks for the see package are a pain, because there are a lot of minor issues, which yield an error (on purpose), but which are difficult to iron out in order to have a "clean basis" for tests.

[IndrajeetPatil]

@IndrajeetPatil do you know why the R-CMD-check-strict on ubuntu is taking forever to load all of the dependencies?

Possibly due to this issue: r-lib/vdiffr#137

[bwiernik]

That looks like it