Plots.jl: Fill overlaps with x/y-axis

[scipham]

I use the following for plotting with Plots.jl:

using Plots
using HDF5
using Statistics

all_predictions = []
all_trueValues = []

h5open("3to12benchmark_corrected_r.dat", "r") do fid

for qn in 3:10

    predictions = []
    trueValues = []
    for runi in 1:3
        push!(predictions, read(fid[string("predictionArray",qn,"q", runi)]))
        push!(trueValues, read(fid[string("truePredictionArray", qn, "q", runi)]))
    end
    push!(all_predictions, trueValues)
    push!(all_trueValues, predictions)
end

limits = [-0.7, 0.7]
trueRelationX, trueRelationY = collect(limits[1]-1:limits[2]+1), collect(limits[1]-1:limits[2]+1)

num_measurements = 400
err = sqrt(1/num_measurements) *sqrt(3^3) #Chebyshev
    
p = [3,5,8,10]

#Using Subplots:
l = @layout grid(2,2){0.5w}
plt = plot( titlefont=5,
  labelfont=5,
  legend=false, 
  layout=l)

for (pi,pe) in enumerate(p)

    plot!(plt[pi], trueRelationX, trueRelationY .- err, 
    fillrange=trueRelationY .+ err, 
    fillalpha= 0.1, 
    fillcolor="blue", 
    linealpha=0.1,
    label="1-sigma",
    markersize=1,
    xlabel="True expectation",
    ylabel="Estimated expectation"

    )

    #Plot the linear y=x line for orientation
    plot!(plt[pi],trueRelationX, trueRelationY,
          linecolor=:black,
          linestyle=:dash,
          guidefontsize = 8
            )

    for runi in 1:3
        plot!(plt[pi], all_trueValues[pe .- 2][runi], all_predictions[pe .- 2][runi],
        seriestype = :scatter, 
        label="Iteration $runi",
        legend=false,
        title=["a", "b", "c", "d"][pi],
        xlim = [-1/sqrt(pi), 1/sqrt(pi)],
        ylim = [-1/sqrt(pi), 1/sqrt(pi)])
    end

end
    
display(plt)

end

My problem is that running this gives:

so the top-left figure gets a fill that shoots over the x-axis (and just doesn’t hit the y-axis). Can anyone say what I am possibly doing wrong here? Or is this probably a bug? I got this in similar plots already so I am not sure how common this is.
(As there is no direct upload function here, the code runs directly with this datafile: 3to12benchmark_corrected_r.dat - Google Drive)

[rafael.guerra]

Is this issue fixed if you enforce tight plots with the argument widen=false?

NB:
Better if you can provide a MWE with input data.

[scipham]

Thanks @rafael.guerra . No, but it’s closer.

I indeed wanted to provide the .dat file but didn’t see a way to upload it here. So maybe this way: 3to12benchmark_corrected_r.dat - Google Drive

[rafael.guerra]

@scipham, tested your code here, added an earlier end to the h5open command, and did not observe any leakage using Plots gr() back-end:

NB:
Julia 1.6.1 on Win10, Plots v1.15.2

[scipham]

Ah, yes in REPL it now works as well. It seems I’ve been relying too much on Jupyter notebooks once again. Now I know again why I didn’t like them Thanks a lot for helping out!