Apparent inconsistent plotting of histograms in Gadfly

[evan-wehi]

Why is the histogram for the WildType normalised data differently when plotted with other data compared to being plotted alone?
This plot command

plot(df[df.genotype .== "WildType", :], x=:level, Geom.histogram(bincount=200), Scale.x_log10, color=:genotype, Coord.cartesian(xmin=-2, xmax=1, ymin=0, ymax=8000)

produces this plot

while this plot

plot(df, x=:level, Geom.histogram(bincount=200), Scale.x_log10, color=:genotype, Coord.cartesian(xmin=-2, xmax=1, ymin=0, ymax=8000))

produces

The WildType data are roughly 3x larger in the second plot. This does not happen to the KO data if I plot it by itself. Am I missing somethong about how Gadfly histograms work?

[evan-wehi]

Here is a useable example the reproduces the behaviour.

using Distributions, Gadfly, DataFrames

d1 = Normal(0, 1)
d2 = Normal(1, 0.5)

x = rand(d1, 10000)
df = DataFrame(x=x, class="Class 1")

x = 0.5 .* rand(d2, 10000)

append!(df, DataFrame(x=x, class="Class 2"))

spike = -2.0 .* ones(500)
append!(df, DataFrame(x=spike, class="Class 2"))

h = plot(df[df.class .== "Class 1", :], 
  x=:x, 
  color=:class, 
  Geom.histogram(bincount=200), 
  Coord.cartesian(xmin=-4, xmax=4, ymin=0, ymax=1000)
  )
display(h)

h = plot(df,
  x=:x, 
  color=:class, 
  Geom.histogram(bincount=200), 
  Coord.cartesian(xmin=-4, xmax=4, ymin=0, ymax=1000)
  )
display(h)

This produces:

It looks like it has plotted Class 1 as the sum of Class 1 and Class 2. Is this to be expected? Is there a way to turn this behaviour off?

[juliohm]

@evan-wehi try to reach the Gadfly.jl maintainers in their community channels. They use Gitter instead of Discourse.

[evan-wehi]

My bad - I missed it in the documentation. This behaviour is intended and is controlled by the position argument in Geom.histogram. I need to use position=:dodge.