Generating a consistent dynamic color palette in ggplot within a loop?

Question

I have a data set called d1 similar to:

location, depth.from, depth.to, val, type

I have a loop that creates a fairly complex plot for each unique location (it's glueing together many things using grid.arrange, which is why I can't use facet_wrap on the location to keep the legend/color consistent of one part of the plot).

Say there are 4 categories for "type", the problem when one location has a different number of "types" than the other, the colors assigned are not consistent between each plot. I can manually force them to be the same, but I am trying to generalize this function. Google has failed me.

For the following block, d1 is a subset of the data based on the location type, e.g.

d1 <- subset(myData, location == location.list[i])

Looking at the plot, which is within the loop:

p1 <- ggplot(data = d1, aes (y=val, x=depth.from))+
layer(geom = "point", size = 2) + 
geom_rect(data=d1, aes(xmin=Depth.to, xmax=Depth.from, ymin=0, ymax=100, fill = type), linetype =0, alpha=0.3)+
scale_fill_brewer(palette="Set1")

the geom_rect command is going through the data and based on the depth from and depth to, creating a overlay based on fill type. I can use scale_fill_manual("Lith", c("Val1" = "DodgerBlue4"...) etc to manually set it, but that defeats the purposes. If I have:types like where I want something like:

Bird_one = blue
Bird_two = red
Bird_three = green

I want bird_three to be green, even if bird_two doesn't exist, without having to explicitly set it using scale_fill_manual. Is there a way to set a global list of names for the color palette? Perhaps by providing the array from something like:

myData <- read.csv("mydata.csv" 
typeList <- unique(myData$type)

Some really good answers here! I appreciate the help. I've gone with the scale_fill_brewer(limits = levels(dat$type), palette = "Set1") solution which seems to be working for everything that I'm throwing at it. — mat4nier
– mat4nier, Commented Jan 31, 2014 at 18:21

BrodieG · Accepted Answer · 2014-01-30 17:52:08Z

6

Pretty late, but there actually is a trivial solution by setting scale_fill_discrete(drop=F)

plots <- lapply(dfs, function(df) {
  ggplot(df, 
    aes(
      x=location, fill=type, 
      y=(depth.from + depth.to) / 2, 
      ymin=depth.from, ymax=depth.to
  ) ) +
  geom_crossbar() + scale_fill_discrete(drop=F)
})
library(gridExtra)
do.call(grid.arrange, plots)

enter image description here

And here is dummy data I used:

set.seed(12)
items <- 2
dfs <- 
  replicate(2, simplify=F,
    data.frame(
      location=sample(letters, items), 
      depth.from=runif(items, -10, -5),
      depth.to=runif(items, 5, 10),
      val=runif(items),
      type=factor(
        sample(c("Bird_one", "Bird_two", "Bird_three"), items),
        levels=c("Bird_one", "Bird_two", "Bird_three")
  ) ) )

answered Jan 30, 2014 at 17:52

BrodieG

52.8k9 gold badges99 silver badges148 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

BrodieG Over a year ago

@hadley, I'm not sure what you mean by "set the limits to the scale" in the previous answer, but wouldn't this be the correct way to do it?

alexwhan Over a year ago

that's neat - what I would like is something similar, but doesn't maintain the absent levels in the legend... I'll do some investigating

alexwhan · Accepted Answer · 2014-01-27 22:18:10Z

6

It is not important whether or not they are in a loop, you just need to associate each level with a colour. In your case:

colourList <- c(bird_one = "red", bird_two = "blue", bird_three = "green")

In a simple example:

#Make some data
dat <- data.frame(location = rep(1:4, c(3,2,2,3)), val = rnorm(10), 
  depth.from = sample(1:5, 10, replace = TRUE), depth.to = sample(6:10, 10, replace = TRUE),
  type = factor(LETTERS[c(1:3, 1,3,1,3,1:3)]))

#Associate levels with colours
colourList <- c(A = "red", B = "blue", C = "green")

p <- list()
for(i in 1:4) {
  d <- dat[dat$location == i,]
  p[[i]] <- ggplot(data = d, aes (y=val, x=depth.from))+
    layer(geom = "point", size = 2) + 
    geom_rect(aes(xmin=depth.to, xmax=depth.from, ymin=0, ymax=100, fill = type), linetype =0, alpha=0.3) +
    #This is where the assignment works
    scale_fill_manual(values=colourList)
}
grid.arrange(p[[1]], p[[2]])

You can see that level C is green in both plots.

In response to @BrodieG, here is a way to set the colours semi-automatically. It is creating the named vector using the levels of type and colour values from the RColorBrewer package. This could be developed pretty easily to include in a function:

library(RColorBrewer)
colourList <- setNames(brewer.pal(length(levels(dat$type)), "Set1"), levels(dat$type))

As @hadley points out, in this instance it's even more straightforward to set the limits of the scale, although in my typical use I find it more useful to setup an object like colourList which can be used across multiple plots by just setting values. Setting limits also maintains the levels for the legend, which may or may not be what is wanted:

scale_fill_brewer(limits = levels(dat$type), palette = "Set1")

enter image description here

As @hadley points out, in this instance it's even more straightforward to set the limits of the scale, although in my typical use I find it more useful to setup an object like colourList which can be used across multiple plots by just setting values. Setting limits also maintains the levels for the legend, which may or may not be what is wanted:

scale_fill_brewer(limits = levels(dat$type), palette = "Set1")

enter image description here

edited Jan 27, 2014 at 22:18

answered Jan 24, 2014 at 2:37

alexwhan

16.1k7 gold badges54 silver badges67 bronze badges

3 Comments

BrodieG Over a year ago

But is there a way to do this without specifying the color scale manually, even if only once. I think this is what the OP is trying to do. For example, I was hoping that by using a factor with the same levels for both plots, that the color mapping would be done on the levels, not on the unique values in the factor. Unfortunately it is the latter.

BrodieG Over a year ago

Yeah, I think that's as good as it gets without actually modifying ggplot. Ideally there would be a direct way of doing this. Since ggplot derives so much of it's behavior from levels (series ordering, etc.), I would have thought it would do this (though to be fair, missing elements in factors don't show up as empty spots on a categorical bar plot either).

BrodieG Over a year ago

@alexwhan, fyi, I stumbled onto the answer (see above).

Collectives™ on Stack Overflow

Generating a consistent dynamic color palette in ggplot within a loop?

2 Answers 2

2 Comments

3 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

2 Comments

3 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related