plot_predictions and labs dictionary

[brueckmann]

Hello,
great package, love it!
Just a minor something I realized:

I cannot add labs via dictionary to plot_predictions. Please see the minimal reproducible example below.

library("marginaleffects")
library("ggplot2")

# Set labels by variable name instead of aesthetic from https://ggplot2.tidyverse.org/reference/labs.html?q=labs#null
cardict  <-  c(
  disp = "Displacment",
  # Not in use
  cyl  = "Number of cylinders",
  mpg  = "Miles per gallon",
  wt   = "Weight (1000 lbs)"
)

# Works in ggplot
ggplot(mtcars, aes(mpg, hp, colour = wt)) + geom_point() +
  labs(dictionary = cardict) +
  theme_classic() # just to see that it did sth.


mod <- lm(mpg ~ hp + wt, data = mtcars)

# and plot_prediction it - labels don't show
plot_predictions(mod, condition = "wt") +
  labs(dictionary = cardict) +
  theme_classic() # just to see that it did sth.

3. sessionInfo() output

R version 4.4.0 (2024-04-24)
Platform: aarch64-apple-darwin20
Running under: macOS 15.7

Matrix products: default
BLAS:   /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib 
LAPACK: /Library/Frameworks/R.framework/Versions/4.4-arm64/Resources/lib/libRlapack.dylib;  LAPACK version 3.12.0

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

time zone: Europe/Berlin
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] ggplot2_4.0.0          marginaleffects_0.31.0

loaded via a namespace (and not attached):
 [1] vctrs_0.6.5        cli_3.6.5          rlang_1.1.6        generics_0.1.4     ellmer_0.3.2       S7_0.2.0          
 [7] coro_1.1.0         jsonlite_2.0.0     data.table_1.17.99 labeling_0.4.3     glue_1.8.0         backports_1.5.0   
[13] scales_1.4.0       rappdirs_0.3.3     grid_4.4.0         tibble_3.3.0       lifecycle_1.0.4    httr2_1.2.1       
[19] insight_1.4.2.13   compiler_4.4.0     dplyr_1.1.4        RColorBrewer_1.1-3 pkgconfig_2.0.3    rstudioapi_0.17.1 
[25] farver_2.1.2       R6_2.6.1           tidyselect_1.2.1   pillar_1.11.0      curl_7.0.0         magrittr_2.0.4    
[31] checkmate_2.3.2    withr_3.0.2        tools_4.4.0        gtable_0.3.6      

Hope that helps!

[vincentarelbundock]

Thanks.

I don't know how this ggplot2 feature works, and TBH I doubt I will have time to look into it. If you or someone else can put together a Pull Request to implement it, I'd be happy to consider.

[brueckmann]

Thanks for letting me know! I might look into it (but if anyone else comes along, feel free!)

[etiennebacher]

The issue is that plot_predictions() calls labs(x = v_x, y = v_y) internally:

marginaleffects/R/plot_predictions.R

Line 210 in f758d40

labs_args <- list(x = v_x, y = dv)

marginaleffects/R/plot_predictions.R

Line 232 in f758d40

p <- p + do.call(ggplot2::labs, labs_args)

This overrides the call to labs(dictionary = cardict), and it can be also observed in pure ggplot2:

library(ggplot2)

cardict <- c(mpg = "Miles per gallon")

# This part is generated inside the function 
ggplot(mtcars, aes(mpg, hp, colour = wt)) +
    geom_point() +
    labs(x = "mpg") + # Commenting this out adds the label to the plot

    # I add this part myself 
    labs(dictionary = cardict)

One approach would be to remove list(x = v_x, y = dv) so that labs() only affects color, shape, etc. but this would probably have side effects (I don't know the codebase of marginaleffects enough for that).

I hope this helps.

[vincentarelbundock]

Thanks for the investigation!

Now that ggplot2 supports setting labels automatically based on column attributes, we may be able to use that internally. Set an attribute and get those labels by default. Then, we don't need to call labs() internally and, presumably, users are allowed to call it subsequently.