2

I have this dataset: https://dl.dropboxusercontent.com/u/73950/data.csv The dataset contains 3 variables.

Here's how I visualize the data right now:

library(ggplot2)
library(reshape2)
library(RColorBrewer)

dat = read.csv("data.csv", header = FALSE)

myPalette <- colorRampPalette(rev(brewer.pal(11, "Spectral")))
sc <- scale_colour_gradientn(colours = myPalette(100))

ggplot(dat, aes(x=V1, y=V3, colour = V2))+ geom_point(alpha = .2,size = 3) + sc

Instead of just one figure, I'd like to facet the figure to display 3 different ways to attribute variables to each axis and color. As such:

  • x = V1, y = V2, color = V3
  • x = V1, y = V3, color = V2
  • x = V2, y = V3, color = V1

How to do this kind of things with ggplot2's faceting?

Lucien S.
  • 5,123
  • 10
  • 52
  • 88

1 Answers1

3

You can get this by putting the data in the format ggplot likes. In this case, a column that can be used to split the data into facets (called var below). To do that, I just repeated the data three times, choosing the appropriate x and y variables for each 2-way combo, and using the variable left out of each combination as the coloring variable.

## Rearrange the data by 2-way combinations, the coloring is the remaining column
res <- do.call(rbind, combn(1:3, 2, function(ii)
    cbind(setNames(dat[,c(ii, setdiff(1:3, ii))], c("x", "y", "color")),
                   var=paste(ii, collapse=".")), simplify=F))

ggplot(res, aes(x=x, y=y, color=color))+ geom_point(alpha = .2,size = 3) + 
  facet_wrap(~ var, scales="free") + sc

enter image description here

Rorschach
  • 31,301
  • 5
  • 78
  • 129
  • Fantastic! is it possible to give each of these facet a separate color legend? – Lucien S. Jul 31 '15 at 00:15
  • 1
    @Rodolphe I'm not sure, I was just trying to make sense of the colors, will update if I figure anything out. What you can do easily is put three plots (not facetted) on the same canvas with different legends, but that totally defeats the purpose of this. – Rorschach Jul 31 '15 at 00:18
  • 1
    @Rodolphe this answer here from hadley, http://stackoverflow.com/a/3809071/2415684, makes me think a hack might be necessary... Maybe color by distance from each variables mean (by `scale`ing)? I dont know sry – Rorschach Jul 31 '15 at 00:31