User defined colour palette in R and ggpairs

Question

I am trying to use another color palette for the scatter plots from ggpairs from the GGally library in R. See similar question here.

library(ggplot2)
library(GGally)

Works

ggplot(iris, aes(x=Sepal.Width, colour=Species)) + stat_ecdf() + scale_color_brewer(palette="Spectral")

Also works

ggplot <- function(...) ggplot2::ggplot(...) + scale_color_brewer(palette="Spectral")
ggplot(iris, aes(x=Sepal.Width, colour=Species)) + stat_ecdf()

Does not work

ggplot <- function(...) ggplot2::ggplot(...) + scale_color_brewer(palette="Spectral")

ggpairs(iris, 
    columns=, c("Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width"),
    colour='Species',
    lower=list(continuous='points'), 
    axisLabels='none',  
    upper=list(continuous='blank')
)

but adding

putPlot(p, ggplot(iris, aes(x=Sepal.Length, colour=Species)) + stat_ecdf(), 1,1)

adds a plot in the right colors.

Workaround

I can change the plots afterwards with getPlot, but that's not pretty..

subplot <- getPlot(a, 2, 1) # retrieve the top left chart
subplotNew <- subplot + scale_color_brewer(palette="Spectral")
a <- putPlot(a, subplotNew, 2, 1)

How can I change the color scheme for the scatter plots in ggpairs? More specifically, I'd like to manually define the colors like so

scale_colour_manual(values=c("#FF0000","#000000", "#0000FF","#00FF00"))

Thanks!

Answer 1

Here is a hack that works:

ggplot <- function(...) ggplot2::ggplot(...) + scale_color_brewer(palette="Spectral")
unlockBinding("ggplot",parent.env(asNamespace("GGally")))
assign("ggplot",ggplot,parent.env(asNamespace("GGally")))

When you assign a new value to the ggplot function, it is in the global environment. Now, GGally imports everything including ggplot when it loads (it didn't have to be that way). At that point, changing the ggplot function in your global environment has no effect, because imports from GGally have precedence. Instead, you need to update the ggplot function on the GGally:imports. There is only one problem: once a package is loaded, its bindings are locked. But we can unlock them (I am guessing this is frowned upon, hence labeling the solution a hack).

See Josh O'Brien's answer under Replace definition of built-in function in R? for more info.