Change colours to defined palette for ggplot objects

Question

I would like to change the default colours to a specific palette for all geom_* type objects by using a function.

Below is an example for geom_line() and using the function change_colours()

# load ggplot2 and tidyr library
require(ggplot2)
require(tidyr)

# create a mock data frame
df <- data.frame(cbind(var1=500*cumprod(1+rnorm(300, 0, 0.04)),
                       var2=400*cumprod(1+rnorm(300, 0, 0.04)),
                       var3=300*cumprod(1+rnorm(300, 0, 0.04))))
df$TS <- as.POSIXct(Sys.time()+seq(300))
df <- gather(df, stock, price, -TS)

# create basic base graph
p <- ggplot(df, aes(x=TS, y=price, group=stock))+geom_line(aes(colour=stock))

# custom pallet
custom_pal <- c("#002776", "#81BC00", "#00A1DE", "#72C7E7", "#3C8A2E", "#BDD203", 
                "#313131", "#335291", "#9AC933", "#33B4E5", "#8ED2EC", "#63A158", 
                "#CADB35", "#575757", "#4C689F", "#A7D04C", "#4CBDE8", "#9DD8EE", 
                "#76AD6D", "#D1DF4F", "#8C8C8C", "#7F93BA", "#C0DE80", "#80D0EE", 
                "#B8E3F3", "#9DC496", "#DEE881", "#B4B4B4", "#99A9C8", "#CDE499", 
                "#99D9F2", "#C7E9F5", "#B1D0AB", "#E5ED9A", "#DCDCDC")

# the function the change colours
change_colours <- function(ggplot_obj, pal){
  p <- ggplot_obj
  group_data <- p$data[, as.character(p$mapping$group)]
  n_groups <- length(unique(group_data))
  group_data_cols <- pal[group_data]

  p + theme_light()+ geom_line(colour=group_data_cols)
}

p
p1 <- change_colours(ggplot_obj=p, pal=custom_pal)
p1

I am hoping to change the change_colours() function to be more generic for all geom_* type layers

Any help would be much appreciated.

Myotonia

I don't get the question. It seems to be that your code is already doing what you asked for. What is your request? – Loadstar

Ah, I get it. For all of the primitives... – Loadstar

Answer 1

The following should do what you're after. Note that it only changes colours that are mapped to variables. Colour passed directly to the geom_* won't be affected (there's an example below). For an approach that modifies colour or fill (whichever is mapped first), see the bottom half of this post.

change_colours <- function(p, palette) {
  n <- nlevels(p$data[[deparse(p$mapping$group)]])
  tryCatch(as.character(palette), 
           error=function(e) stop('palette should be a vector of colours', call.=FALSE))
  if(n > length(palette)) stop('Not enough colours in palette.')
  pal <- function(n) palette[seq_len(n)]
  p + theme_light() + discrete_scale('colour', 'foo', pal)
}

# Here, df is from the OP's post
p <- ggplot(df, aes(x=TS, y=price, group=stock)) 

Examples:

# NB: custom_pal is defined in the OP's post
change_colours(p + geom_line(aes(colour=stock)), custom_pal)

change_colours(p + geom_point(aes(colour=stock)), custom_pal)

And with a different palette:

change_colours(p + geom_smooth(aes(colour=stock)), 
               c('firebrick', 'midnightblue', 'violet', 'seagreen'))

As mentioned above, this will only change colour and fill that are mapped to variables. For example, it'll have no effect on the colours for the following:

change_colours(p + geom_point(colour=rep(c('tomato', 'hotpink', 'cadetblue'), each=300)), 
               custom_pal)

---

In response to the OP's comment, you can easily detect what types of mappings are being used (e.g. alpha, colour, fill). Just look at p$layers[[1]]$mapping.

If we assume that the first fill or colour mapping of the first layer is the mapping for which you want to change colours, you can do:

change_colours <- function(p, palette, type) {
  n <- nlevels(p$data[[deparse(p$mapping$group)]])
  tryCatch(as.character(palette), 
           error=function(e) stop('palette should be a vector of colours', call.=FALSE))
  if(n > length(palette)) stop('Not enough colours in palette.')
  if(missing(type)) 
    type <- grep('colour|fill', names(p$layers[[1]]$mapping), value=TRUE)[1]
  pal <- function(n) palette[seq_len(n)]
  p + theme_light() + discrete_scale(type, 'foo', pal)
}

# Here, df is from the OP's post
p <- ggplot(df, aes(x=TS, y=price, group=stock))

Examples:

Changing fill instead of colour:

change_colours(p + geom_point(aes(fill=stock), pch=21), 
               c('white', 'grey50', 'grey80'))

Showing priority of first mapped colour/fill aesthetic:

change_colours(p + geom_point(aes(fill=stock, color=stock), pch=21) +
                 geom_smooth(aes(color=stock)), 
               c('black', 'grey50', 'grey80'))

change_colours(p + geom_point(aes(color=stock, fill=stock), pch=21) +
                 geom_smooth(aes(color=stock)), 
               c('black', 'grey50', 'grey80'))

Override the priority of the first mapped aesthetic with the type argument, e.g.:

change_colours(p + geom_point(aes(color=stock, fill=stock), pch=21) +
                 geom_smooth(aes(color=stock)), 
               c('black', 'grey50', 'grey80'), type='fill')

Startle

Is there a way to check to see if/detect if fill is being used or colour is being used as the 'main' colouring 'brush'? e.g. when trying: df2 <- df %>% group_by(stock) %>% summarise(mean_px=mean(price)); p2 <- ggplot(df2,aes(x=stock,y=mean_px))+geom_bar(aes(fill=stock),stat='identity') then the new change_colours function doesn't quite work... – Myotonia