How to extract an interaction to an external variable from a ggplot graph?

Question

I have a ggplot graph defined like this:

x <- seq(0, 10, by = 0.1)
y1 <- cos(x)
y2 <- sin(x)
df1 <- data.frame(x = x, y = y1, type = "sin", id = 1)
df2 <- data.frame(x = x, y = y2, type = "cos", id = 2)
df3 <- data.frame(x = 2, y = 0.5, type = "constant", id = 3)
df4 <- data.frame(x = 4, y = 0.2, type = "constant", id = 4)

combined <- rbind(df1, df2, df3, df4)

ggplot(combined, aes(x, y, colour = interaction(type, id))) + geom_line() +
  geom_point(data = subset(combined, type == "constant"))

This works very well as illustrated below:

Now I would like to extract the interaction in a variable to reuse it later (e.g. customize the legend style or labels).

I did that in a very naïve way:

my.interaction <- interaction(combined$type, combined$id)

ggplot(combined, aes(x, y, colour = my.interaction)) + geom_line() +
  geom_point(data = subset(combined, type == "constant"))

But then I have an error:

Error: Aesthetics must be either length 1 or the same as the data (2): x, y, colour

Edit: Here is the kind of manipulation I could do: edit the linetype of the legend

displayed <- levels(factor(my.interaction))
line.style <- rep(1, length.out = length(displayed))
line.style[grep("constant", displayed)] <- 0

That works:

ggplot(combined, aes(x, y, colour = interaction(type, id))) + geom_line() +
    geom_point(data = subset(combined, type == "constant")) +
    guides(colour=guide_legend(override.aes=list(linetype = line.style)))

That does not:

ggplot(combined, aes(x, y, colour = my.interation) + geom_line() +
    geom_point(data = subset(combined, type == "constant")) +
    guides(colour=guide_legend(override.aes=list(linetype = line.style)))

In the end, I could also edit the shapes or the legend labels (e.g. "Id: 1 / Type: sin" or any other advanced transformation of the labels based on the interaction values).

Are you opposed to adding my.interaction to the combined data.frame? You could do, combined$my.interaction <- interaction(combined$type, combined$id) and your approach will work. — JasonAizkalns
Yes, I want to avoid this. Because depending on what I want to plot, I might use other interactions in different plots... Thus making my data frame grow with somewhat redundant information. — Ben
Can you be more specific and provide an example of your comment? You may want to edit/update your question. I think this might be an XY problem. — JasonAizkalns

Edward R. Mazurek Edward R. Mazurek · Accepted Answer · 2016-05-25T14:08:38

This'll work. What's wrong with adding a column to your data frame?

combined %>% mutate(my.interaction = paste(type, id, sep='.')) %>%
  ggplot(aes(x, y, colour = my.interaction)) + geom_line() +
  geom_point(data = subset(combined, type == "constant"))

How to extract an interaction to an external variable from a ggplot graph?

1 Answers