Statistic errors for histograms with Plots recipes

Specific Domains Visualization
,
#1

I would love to have a recipe for Plots.jl that creates a histogram
with xerror of the size of the bin, and yerror according to e.g. sqrt(h.weight).

So, there are two problems:

  1. x-errors do not show up. The scatterhist is setting xerror in scatterbins
    Plots.jl/recipes.jl at master · JuliaPlots/Plots.jl · GitHub
    However, they do not appear
scatterhist(rand(100)) # does not have x-errors
# but!
h = fit(StatsBase.Histogram, ds)
plot(h, seriestype=:scatterbins) # does have x-errors
  1. y-errors do not appear, I cannot make the line yerror := sqrt.(h.weights) work

here is what I tried:

@recipe function f(::Type{Val{:errorhist}}, x, y, z)
    h = Plots._make_hist((y,), plotattributes[:bins], normed = plotattributes[:normalize], weights = plotattributes[:weights])
    edge = collect(h.edges)
    # 
    x := Plots._bin_centers(edge)
    y := h.weights
    xerror := diff(edge)/2
    yerror := sqrt.(h.weights)
    seriestype := :scatter
    linewidth := 2
    ()
end
"""
    errorhist
"""
@shorthands errorhist
#
errorhist(rand(1000), bins=100) # empty plot ...

EDIT: the plot was empty because edge = collect(h.edges) suppose to be edge = collect(h.edges[1]). After fixing, the scatter plot appears but without any errorbars.

#2

Does your code work without the recipe macro? If not, and you have a small number of bins, a bar and scatter plot combination might work. Perhaps like the following:

using Plots
gr(size=(1000,900), xtickfontsize=13, ytickfontsize=13, xguidefontsize=16, yguidefontsize=16, legendfontsize=12, dpi=100, grid=(:y, :gray, :solid, 1, 0.4))
counts = [5,7,8];
barswitherror = bar(["1-3", "4-10", "11 to 19"], counts, legend=false);
scatter!(collect(0.5:1:length(counts)), counts, xerr=[2,6,8] ./ maximum([2,6,8]), yerr=sqrt.(counts), ylim=(0,12), bar_width=5, xlabel="Bins", ylabel="Counts", title="Histogram with x- and y-errors", marker=stroke(1.5))
savefig(barswitherror, "barswitherror.png")

And while not perfect, it does produce a histogram-like figure:

barswitherror
barswitherror1000×900 16.9 KB

#3

thanks for the reply.

Sure, I can create a StatsBase.Histogram first and, then, plot using the barplot as you suggest. However, It is not the point. One should be able to say èrrorhist(rand(1000)) and get the right plot.

#4

Doesn’t StatsPlots work for you?

#5

sure, why?

#6

I could not figure out, looks like a bug somewhere