Handling promises (rlang) with for loops and ggplot in R - r

Question

The aim of this script was to replicate something like the figure below:
found on: https://robjhyndman.com/hyndsight/tscv/
The problem I have encountered relates to (I think) how R is handling my promises in ggplot.
Below is an example which reproduces my problem.
library(tidyverse)
process_starting_row <- 600
per_validation_period <- 30
number_of_validations <- 5
graphical_data <- data.frame(x= 1:(process_starting_row + 1 + (number_of_validations)*per_validation_period))
for (it in 1:number_of_validations) {
# For this graph there is always a line and then a colour component explaining each one...
graphical_data[,paste0("iteration",it,"line")] <- c(it)
# First make the whole row grey and then "dolly up" the colours.
graphical_data[,paste0("iteration",it,"colour")] <- "grey"
graphical_data[1:(process_starting_row + (it-1)*per_validation_period), paste0("iteration",it,"colour")] <- "blue"
graphical_data[(process_starting_row + 1 + (it)*per_validation_period), paste0("iteration",it,"colour")] <- "red"
}
#graphical_data
The above code creates a dataframe object which could be used to create the desired figure. For each iteration (in the original figure a different line) it creates a vector corresponding to the iterations "height" above the axis (that columns name is always iteration#line and a corresponding character vector, iteration#colour, with the colour code for each of the dots.
The next bit is to create a base ggplot object.
ggbase <- ggplot(data = graphical_data, aes(x=x)) +
coord_cartesian(xlim = c(process_starting_row-1*per_validation_period, nrow(graphical_data))) +
theme_bw()
It is upon this base object that I wish to iterate.
I wrote a function which would add each iteration gg_adding() and then another ggaddfor() which runs the for loop.
gg_adding <- function(data, iteration_sub, color_sub){
iteration_promise <- enquo(iteration_sub)
colour_promise <- enquo(color_sub)
gg <- geom_point(data = data, aes(x= x, y= !! iteration_promise, color = !! colour_promise))
return(gg)
}
ggaddfor <- function(data, gg){
ggout <- gg
for (it in 1:number_of_validations) {
#print(it)
iterationsub <- paste0("iteration",it,"line")
coloursub <- paste0("iteration",it,"colour")
ggout <- ggout + gg_adding(data, iterationsub, coloursub)
}
return(ggout)
}
When I run this function I get the following:
# Not working
ggaddfor(graphical_data, ggbase)
Which produces output that looks like this:
Clearly that's not what I was hoping for...
In order to test things I stipulated each iteration explicitly.
# Working...
ggadd <- ggbase
ggadd <- ggadd + gg_adding(graphical_data, iteration1line, iteration1colour)
ggadd <- ggadd + gg_adding(graphical_data, iteration2line, iteration2colour)
ggadd <- ggadd + gg_adding(graphical_data, iteration3line, iteration3colour)
ggadd <- ggadd + gg_adding(graphical_data, iteration4line, iteration4colour)
ggadd <- ggadd + gg_adding(graphical_data, iteration5line, iteration5colour)
This produces the desired output:
I want to put these functions into a package I'm currently writing and so explicitly stipulating the additions (as I do directly above) is not going to work...
I'm not sure why my earlier code is not producing the same results. I'm somewhat new to handling promises with the rlang package and I suspect my mistake could be there...

Answer 1

What worked for me is to replace your enquo() calls in your gg_adding() function by as.symbol(), so that the new function would look like this:
gg_adding <- function(data, iteration_sub, color_sub){
iteration_promise <- as.symbol(iteration_sub)
colour_promise <- as.symbol(color_sub)
gg <- geom_point(data = data, aes(x= x, y= !! iteration_promise, color = !! colour_promise))
return(gg)
}
However, in order to not duplicate your data every iteration, I would suggest this as your geom_point() call.
gg <- geom_point(aes(y= !! iteration_promise, color = !! colour_promise))
I'm tangentially familiar with tidy evaluation and quotation, but not fully. The thing that I understand is that whatever you put in aes(), will always be evaluated in context of data column names, first in the layer's data, next in the global data, unless the user is explicit in his calls (e.g aes(fill = "black") or something). Because a value for x and data are already specified in your ggbase construction, we do not need it in your geom_point() call.
I know this is maybe an unsollicited tip and I apologise, but ggplot seems to prefer to work with long data more than with wide data. What I mean with 'wide' data is that your iterations are sort of cbind()-ed together. Therefore, if you first calculate each iteration and then rbind() them together, you could shorten your script by quite a bit and circumvent the (quasi)quotation stuff altogether to produce a similar plot:
new_gr_dat <- lapply(seq_len(number_of_validations), function(it){
df <- data.frame(x= 1:(process_starting_row + 1 + (number_of_validations)*per_validation_period),
line = it, # doubles as y-value and iteration tracker
colour = "grey")
df[1:(process_starting_row + (it-1)*per_validation_period), "colour"] <- "blue"
df[(process_starting_row + 1 + (it)*per_validation_period), "colour"] <- "red"
return(df)
})
new_gr_dat <- do.call(rbind, new_gr_dat)
ggplot(new_gr_dat, aes(x = x, y = line, colour = colour)) +
geom_point() +
coord_cartesian(xlim = c(process_starting_row-1*per_validation_period, max(new_gr_dat$x)))