In jonocarroll/ggghost: Capture the Spirit of Your 'ggplot2' Calls

knitr::opts_chunk$set(
  collapse = TRUE,
  message = FALSE,
  warning = FALSE,
  comment = "#>",
  fig.path = "README_supp/README-"
)

:ghost: Oh, no! I think I saw a ... g-g-ghost

Capture the spirit of your ggplot2 calls.

Motivation

ggplot2::ggplot() stores the information needed to build the graph as a grob, but that's what the computer needs to know about in order to build the graph. As humans, we're more interested in what commands were issued in order to build the graph. For good reproducibility, the calls need to be applied to the relevant data. While this is somewhat available by deconstructing the grob, it's not the simplest approach.

Here is one option that solves that problem.

ggghost stores the data used in a ggplot() call, and collects ggplot2 commands (usually separated by +) as they are applied, in effect lazily collecting the calls. Once the object is requested, the print method combines the individual calls back into the total plotting command and executes it. This is where the call would usually be discarded. Instead, a "ghost" of the commands lingers in the object for further investigation, subsetting, adding to, or subtracting from.

Installation

You can install ggghost from CRAN with:

install.packages("ggghost")

or the development version from github with:

# install.packages("devtools")
devtools::install_github("jonocarroll/ggghost")

Usage

use %g<% to initiate storage of the ggplot2 calls then add to the call with each logical call on a new line (@hrbrmstr style)

tmpdata <- data.frame(x = 1:100, y = rnorm(100))
head(tmpdata)

library(ggplot2)
library(ggghost)
z %g<% ggplot(tmpdata, aes(x, y))
z <- z + geom_point(col = "steelblue")
z <- z + theme_bw()
z <- z + labs(title = "My cool ggplot")
z <- z + labs(x = "x axis", y = "y axis")
z <- z + geom_smooth()

This invisibly stores the ggplot2 calls in a list which can be reviewed either with the list of calls

summary(z)

or the concatenated call

summary(z, combine = TRUE)

The plot can be generated using a print method

which re-evaluates the list of calls and applies them to the saved data, meaning that the plot remains reproducible even if the data source is changed/destroyed.

The call list can be subset, removing parts of the call

subset(z, c(1,2,6))

Plot features can be removed by name, a task that would otherwise have involved re-generating the entire plot

z2 <- z + geom_line(col = "coral")
z2 - geom_point()

Calls are removed based on matching to the regex \\(.*$ (from the first bracket to the end of the call), so arguments are irrelevant.

The object still generates all the grob info, it's just stored as calls rather than a completed image.

str(print(z))
#> [... truncated ...]

Since the grob info is still produced, normal ggplot2 operators can be applied after the print statement, such as replacing the data

xvals <- seq(0,2*pi,0.1)
tmpdata_new <- data.frame(x = xvals, y = sin(xvals))
print(z - geom_smooth()) %+% tmpdata_new

ggplot2 calls still work as normal if you want to avoid storing the calls.

ggplot(tmpdata) + geom_point(aes(x,y), col = "red")

Since the object is a list, we can stepwise show the process of building up the plot as a (re-)animation

lazarus(z, "mycoolplot.gif")

A supplementary data object (e.g. for use in a geom_* or scale_* call) can be added to the ggghost object

myColors <- c("alpha" = "red", "beta" = "blue", "gamma" = "green")
supp_data(z) <- myColors

These will be recovered along with the primary data.

For full reproducibility, the entire structure can be saved to an object for re-loading at a later point. This may not have made much sense for a ggplot2 object, but now both the original data and the calls to generate the plot are saved. Should the environment that generated the plot be destroyed, all is not lost.

saveRDS(z, file = "README_supp/mycoolplot.rds")
rm(z)
rm(tmpdata)
rm(myColors)
exists("z")
exists("tmpdata")
exists("myColors")

Reading the ggghost object back to the session, both the relevant data and plot-generating calls can be re-executed.

z <- readRDS("README_supp/mycoolplot.rds")
str(z)

recover_data(z, supp = TRUE)
head(tmpdata)

myColors

z

We now have a proper reproducible graphic.

Caveats

The data must be used as an argument in the ggplot2 call, not piped in to it. Pipelines such as z %g<% tmpdata %>% ggplot() won't work... yet.
~~Only one original data set will be stored; the one in the original ggplot(data = x) call. If you require supplementary data for some geom then you need manage storage/consistency of that.~~ (fixed)
For removing labs calls, an argument must be present. It doesn't need to be the actual one (all will be removed) but it must evaluate in scope. TRUE will do fine.