ggplot2: Automatic scaling to include complete contour lines in geom_density_2d

Question

Hopefully this will be quick.

I have plotted the following chart using ggplot.

ggplot chart

with the code:

ggplot(ContourDummy,aes(x=Measure.Name1,y=Measure.Name2,colour=Category.Name))
+geom_density_2d()

My issue is that some of the contour lines are not complete.

Now if I scale my axis by adding the following...

+ scale_x_continuous(minor_breaks=0, breaks=seq(14,26,12),limits=c(14,26)) 
+ scale_y_continuous(minor_breaks=0, breaks=seq(50,100,50),limits=c(50,100)

I get the desired output.

But is there any way of automatically setting the limits? I want to be able to replicate this chart type automatically just by switching the data source, x, y and colour.

I don't particularly want to be fiddling around with scales every time.

Can you please make a [reproducible example](http://stackoverflow.com/questions/5963269/how-to-make-a-great-r-reproducible-example)? — Heroka, Mar 13 '16 at 19:37
`geom_density_2d()` uses the limits of the data by default. You would need to calculate the densities outside ggplot, calculate the required limits based on the relevant density level and supply that to the scales in ggplot. — Thierry, Mar 13 '16 at 19:44

eipi10 · Accepted Answer · 2016-03-14T15:21:16.747

Here's a function that expands the x and y ranges to include the maximum extent of the density contours. The function works as follows:

Create a plot object with x and y ranges expanded well beyond the data range, so that we can be sure the plot will include complete contour lines.
Use ggplot_build to determine the min and max x and y values among all the density contours.
Set the x and y ranges of the plot to the min and max x and y values determined in step 2.

The exp parameter is there to expand the final range by a tiny amount (1% by default) because a small piece of contour line can still be cut off without that small bit of extra padding (in the example below, try plotting the mtcars data frame with exp=0 and you'll see what I mean).

d2d = function(data, var1, var2, col, exp=0.005) {

  # If the colour variable is numeric, convert to factor
  if(is.numeric(data[,col])) {
    data[,col] = as.factor(data[,col])
  }

  # Create plot, but expand x and y ranges well beyond data
  p=ggplot(data, aes_string(var1, var2, colour=col)) +
    geom_density_2d() +
    scale_x_continuous(limits=c(min(data[,var1]) - 2*diff(range(data[,var1])),
                                max(data[,var1]) + 2*diff(range(data[,var1])))) +
    scale_y_continuous(limits=c(min(data[,var2]) - 2*diff(range(data[,var2])),
                                max(data[,var2]) + 2*diff(range(data[,var2]))))

  # Get min and max x and y values among all density contours
  pb = ggplot_build(p)

  xyscales = lapply(pb$data[[1]][,c("x","y")], function(var) {
    rng = range(var)
    rng + c(-exp*diff(rng), exp*diff(rng))
  })

  # Set x and y ranges to include complete density contours
  ggplot(data, aes_string(var1, var2, colour=col)) +
    geom_density_2d() +
    scale_x_continuous(limits=xyscales[[1]]) +
    scale_y_continuous(limits=xyscales[[2]]) 
}

Try out the function on two built-in data sets:

d2d(mtcars, "wt","mpg", "cyl")
d2d(iris, "Petal.Width", "Petal.Length", "Species")

Here's what the plots look like with the default x and y ranges:

ggplot(mtcars, aes(wt, mpg, colour=factor(cyl))) + geom_density_2d()

ggplot(iris, aes(Petal.Width, Petal.Length, colour=Species)) + geom_density_2d()

If you want to control the number of axis tick marks as well, you can, for example, do something like this:

d2d = function(data, var1, var2, col, nx=5, ny=5, exp=0.01) {

  require(scales)

  # If the colour variable is numeric, convert to factor
  if(is.numeric(data[,col])) {
    data[,col] = as.factor(data[,col])
  }

  # Create plot, but expand x and y ranges well beyond data
  p=ggplot(data, aes_string(var1, var2, colour=col)) +
    geom_density_2d() +
    scale_x_continuous(limits=c(min(data[,var1]) - 2*diff(range(data[,var1])),
                                max(data[,var1]) + 2*diff(range(data[,var1])))) +
    scale_y_continuous(limits=c(min(data[,var2]) - 2*diff(range(data[,var2])),
                                max(data[,var2]) + 2*diff(range(data[,var2]))))

  # Get min and max x and y values among all density curves
  pb = ggplot_build(p)

  xyscales = lapply(pb$data[[1]][,c("x","y")], function(var) {
    rng = range(var)
    rng + c(-exp*diff(rng), exp*diff(rng))
  })

  # Set x and y ranges to include all of outer density curves
  ggplot(data, aes_string(var1, var2, colour=col)) +
    geom_density_2d() +
    scale_x_continuous(limits=xyscales[[1]], breaks=pretty_breaks(n=nx)) +
    scale_y_continuous(limits=xyscales[[2]], breaks=pretty_breaks(n=ny)) 
}

Thanks for providing me with such a detailed reply. I made my own temp solution by using the following which finds me the maximum and min values for the points on each axis and then extends by +-20% +scale_x_continuous(limits=c(min(ContourDummy$Measure.Name1)*0.8,max(ContourDummy$Measure.Name1)*1.2) +scale_y_continuous(limits=c(min(ContourDummy$Measure.Name2)*0.8,max(ContourDummy$Measure.Name2)*1.2) Which worked in all cases I put through it but I suppose long term there are potential issues (negative numbers for instance). I will try and integrate this piece above into my flow tomorrow. — Benjamin Moss, Mar 13 '16 at 23:10
My original answer used a function that worked in a similar way, with the expansion factor as an argument to the function. In my updated answer, the range is found automatically, so you don't have to tweak the percentage expansion by hand. — eipi10, Mar 14 '16 at 15:23

ggplot2: Automatic scaling to include complete contour lines in geom_density_2d

1 Answers1