Awesome

ggreverse

ggreverse takes a ggplot object and returns the code to create that plot.

This package is written as a learning exercise to help me figure out the internal structure of a ggplot object.

Releases

• 0.1.0 - initial release
• 0.1.1 - improved theme handling

Installation

You can install from GitHub with:

# install.packages("remotes")
remotes::install_github("coolbutuseless/ggreverse")

Example ggreverse::convert_to_code()

1. Create a ggplot
2. Convert the ggplot back into code using ggreverse
3. Convert the code back into a plot

library(ggreverse)

plot_df <- mtcars

# Create a ggplot2 plot object
p <- ggplot(plot_df) +
  geom_point(aes(mpg, wt, colour = cyl), size = 3) +
  labs(title = "hello") +
  theme_bw() + 
  theme(legend.position = 'none') + 
  coord_equal()

# Convert the plot object back into code
plot_code <- ggreverse::convert_to_code(p)
print(plot_code)

#> ggplot(data = plot_df) +
#>   geom_point(mapping = aes(x = mpg, y = wt, colour = cyl), size = 3, position = position_identity(), stat = "identity") +
#>   labs(title = "hello", x = "mpg", y = "wt", colour = "cyl") +
#>   theme_bw(11) +
#>   theme(legend.position = "none") +
#>   coord_fixed()

# Parse the plot code back into a plot - which should match the original plot
eval(parse(text = plot_code))

Technical Notes

• the data arguments to ggplot() and geom() are evaluated at call time. There is no easy way to recover the name of the data argument.
  • ggreverse tries to match the actual data in the ggplot object against a named object in the plotting environment. Otherwise it uses a generic data name
• aesthethic mappings are evaluated at call time, so tidyeval and aes_string() mappings are supported, but ggreverse will only include the final variable name mapping.
• Layers are currently extracted as geom_x(stat = 'y') rather than stat_y(geom='x').
  I’m not sure if there are any cases where these aren’t equivalent.

ToDo

• Extracting facet and scales information.
• Complete themes which are customisations of built-in themes could be more compact if nested diffs where done between themes, rather than just a shallow_diff()
• Lots of other stuff :)

SessionInfo

Developed against:

• R 3.5.3
• ggplot2 v3.1.1