How to add point size to a Gadfly scatter plot based on a variable

New to Julia
,
1

I am trying to reproduce this Seaborn plot using Gadfly.

The code I have so far is:

using CSV, DataFrames, Gadfly

download("https://raw.githubusercontent.com/mwaskom/seaborn-data/master/mpg.csv", "mpg.csv");
mpg = DataFrame(CSV.File("mpg.csv"));

p = plot(
     mpg, 
     x = :horsepower, 
     y = :mpg, 
     color = :origin, 
     #size = :weight, 
     alpha = [0.5], 
     Geom.point,
     Theme(background_color = "white"),
     )

How do I add the size of the circles to match the car weight and add this as a second legend?

enter image description here

2

This part of the Gadfly documentation is pretty confusing, because Gadfly has several methods for setting sizes and the distinctions therein aren’t particularly explicitly laid out. In particular, size attributes can refer to (a) multiples of Theme.point_size_{min or max}, (b) physical distances like 2mm, (c) distances in plot-axis units.

You may have noticed that with your above snippet, but with the #size line not commented out, the plot looks nonsensical:

image
image879Ă—540 20.1 KB

That’s because extrema(mpg.weight) == (1613, 5140). The Gadfly docs on Geom.point aesthetic size state:

size : Categorical data and vectors of Ints will interpolate between Theme.point_size_{min,max} . A continuous vector of AbstractFloats or Measures of length(x) specifies the size of each data point explicitly. A vector of length one specifies the size to use for all points. Default is Theme.point_size.

I’m not entirely sure how the interpolation between Theme.point_size_{min,max} works, but definitely larger integers in the weight attribute will give you larger circles. The above nonsense plot has a bunch of giant overlapping circles, and we’re “zoomed out” so far we can’t really see them all. Zoom in really far and you’ll see a reasonable plot.

Gadfly can size areas according to a function of the data used for size, analogous to how when coloring points, a data attribute is passed through a function to get the color for each point. The relevant scale here is Scale.size_area (or radius, though probably you want area?).

 p = plot(
    mpg,
    x = :horsepower,
    y = :mpg,
    color = :origin,
    size = :weight,
    alpha = [0.5],
    Geom.point,
    Theme(background_color = "white"),
    Scale.size_area(maxvalue=maximum(mpg.weight)),
)

image
image855Ă—544 68.7 KB

1 Like
3

Thanks for your detailed answer, highly appreciated!