Cannot plot errorbars extending below 0 in log plot

[swiesend]

I think this is a very similar issue to #920 and #455.

When I try to use yscale=:log10 with negative numbers involved, than plotting fails. Is this to be expected?

I expected that the negative yaxis also could be scaled to log10.

julia> using Plots

julia> y = cumsum(randn(1000))
1000-element Array{Float64,1}:
  1.8348617340698719
  1.9606676421249605
  4.333292609221781
  4.09586047315016
  2.98930280831738
  2.357884780473308
  2.6189274704623235
  2.83913285661845
  4.344737082863868
  4.763790453265313
  7.7348590368246235
  6.200257461690212
  7.013165122758999
  ⋮
 52.712163269110924
 52.068582279483046
 53.78077131388864
 53.11248976727025
 52.746426694762135
 52.0656869461403
 52.80934531825999
 52.79565060643303
 52.97480165785222
 52.36469731394966
 51.51959490723907
 53.491865989913464

julia> plot(y, label="data", yscale=:log10)

julia> y = cumsum(randn(1000))
1000-element Array{Float64,1}:
  1.6817492628520394
  0.3680270914647106
 -1.2270393491574023
 -1.5711404626304561
 -0.5348526196816741
 -0.8200693718466388
  0.39696492411542406
  0.08383529336848139
 -1.134012164398457
 -1.6296405703629382
 -1.0146955231435941
 -0.5772690171903598
 -0.15132812426836884
  ⋮
 37.10157525765415
 36.023787355352816
 35.17354783593564
 36.08233783147756
 35.803529277885154
 35.337439611497885
 34.62101719705245
 34.330446662122334
 34.882637867410864
 34.22854522933675
 35.754676632885015
 35.52025132181647

julia> plot(y, label="data", yscale=:log10)
Error showing value of type Plots.Plot{Plots.GRBackend}:
ERROR: DomainError with -28.45252942385645:
NaN result for non-NaN input.
Stacktrace:
 [1] nan_dom_err at ./math.jl:339 [inlined]
 [2] log10(::Float64) at ./math.jl:485
 [3] (::getfield(Plots, Symbol("##106#107")){Symbol})(::Float64) at /home/sebastian/.julia/packages/Plots/UQI78/src/axes.jl:151
 [4] widen(::Float64, ::Float64, ::Symbol) at /home/sebastian/.julia/packages/Plots/UQI78/src/axes.jl:451
 [5] axis_limits(::Plots.Axis, ::Bool) at /home/sebastian/.julia/packages/Plots/UQI78/src/axes.jl:510
 [6] axis_limits at /home/sebastian/.julia/packages/Plots/UQI78/src/axes.jl:483 [inlined]
 [7] axis_drawing_info(::Plots.Subplot{Plots.GRBackend}) at /home/sebastian/.julia/packages/Plots/UQI78/src/axes.jl:590
 [8] _update_min_padding!(::Plots.Subplot{Plots.GRBackend}) at /home/sebastian/.julia/packages/Plots/UQI78/src/backends/gr.jl:584
 [9] iterate at ./generator.jl:47 [inlined]
 [10] _collect(::Array{AbstractLayout,2}, ::Base.Generator{Array{AbstractLayout,2},typeof(Plots._update_min_padding!)}, ::Base.EltypeUnknown, ::Base.HasShape{2}) at ./array.jl:619
 [11] collect_similar at ./array.jl:548 [inlined]
 [12] map at ./abstractarray.jl:2018 [inlined]
 [13] _update_min_padding!(::Plots.GridLayout) at /home/sebastian/.julia/packages/Plots/UQI78/src/layouts.jl:310
 [14] prepare_output(::Plots.Plot{Plots.GRBackend}) at /home/sebastian/.julia/packages/Plots/UQI78/src/plot.jl:261
 [15] display(::Plots.PlotsDisplay, ::Plots.Plot{Plots.GRBackend}) at /home/sebastian/.julia/packages/Plots/UQI78/src/output.jl:144
 [16] display(::Any) at ./multimedia.jl:287
 [17] #invokelatest#1 at ./essentials.jl:742 [inlined]
 [18] invokelatest at ./essentials.jl:741 [inlined]
 [19] print_response(::IO, ::Any, ::Any, ::Bool, ::Bool, ::Any) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/REPL/src/REPL.jl:155
 [20] print_response(::REPL.AbstractREPL, ::Any, ::Any, ::Bool, ::Bool) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/REPL/src/REPL.jl:140
 [21] (::getfield(REPL, Symbol("#do_respond#38")){Bool,getfield(REPL, Symbol("##48#57")){REPL.LineEditREPL,REPL.REPLHistoryProvider},REPL.LineEditREPL,REPL.LineEdit.Prompt})(::Any, ::Any, ::Any) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/REPL/src/REPL.jl:714
 [22] #invokelatest#1 at ./essentials.jl:742 [inlined]
 [23] invokelatest at ./essentials.jl:741 [inlined]
 [24] run_interface(::REPL.Terminals.TextTerminal, ::REPL.LineEdit.ModalInterface, ::REPL.LineEdit.MIState) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/REPL/src/LineEdit.jl:2273
 [25] run_frontend(::REPL.LineEditREPL, ::REPL.REPLBackendRef) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/REPL/src/REPL.jl:1035
 [26] run_repl(::REPL.AbstractREPL, ::Any) at /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.1/REPL/src/REPL.jl:192
 [27] (::getfield(Base, Symbol("##734#736")){Bool,Bool,Bool,Bool})(::Module) at ./client.jl:362
 [28] #invokelatest#1 at ./essentials.jl:742 [inlined]
 [29] invokelatest at ./essentials.jl:741 [inlined]
 [30] run_main_repl(::Bool, ::Bool, ::Bool, ::Bool, ::Bool) at ./client.jl:346
 [31] exec_options(::Base.JLOptions) at ./client.jl:284
 [32] _start() at ./client.jl:436

[anandijain]

I am also getting this bug ERROR: ArgumentError: At least one finite value must be provided to formatter when using yscale=:log10.

[samuela]

I just ran into what I believe to be the same thing. Here's a reproduction:

using Plots
xs = [
    125.52,
    145.44,
    190.32,
    245.64,
    332.4,
    475.32,
    678.0,
    1125.36,
    4134.6,
    31917.84,
]
ys = [
    1307.3872007891846,
    31042.239707177123,
    1181.3060910085344,
    896.9340954470773,
    1.9948095908444798,
    0.0014904280995787927,
    28.499027488786066,
    0.09929698822471894,
    17.774974274675294,
    0.018928728905538334,
]
yerror = [
    7330.035378771072,
    191504.15596389078,
    8299.491801063763,
    5832.231598360875,
    13.323568058590752,
    0.005526929674038933,
    148.13836172133148,
    0.5653059411083314,
    125.6863542615931,
    0.0997831262556895,
]
display(plot(xs, ys, yaxis = :log, yerror = yerror))

gives me the error:

ERROR: LoadError: DomainError with -160461.91625671365:
NaN result for non-NaN input.
Stacktrace:
 [1] nan_dom_err at ./math.jl:404 [inlined]
 [2] log10(::Float64) at ./math.jl:550
 [3] (::RecipesPipeline.var"#11#12"{Symbol})(::Float64) at /Users/skainswo/.julia/packages/RecipesPipeline/5RD7m/src/utils.jl:200
 [4] widen(::Float64, ::Float64, ::Symbol) at /Users/skainswo/.julia/packages/Plots/SMabf/src/axes.jl:420
 [5] axis_limits(::Plots.Subplot{Plots.GRBackend}, ::Symbol, ::Bool, ::Bool) at /Users/skainswo/.julia/packages/Plots/SMabf/src/axes.jl:482
 [6] axis_limits(::Plots.Subplot{Plots.GRBackend}, ::Symbol) at /Users/skainswo/.julia/packages/Plots/SMabf/src/axes.jl:452
 [7] optimal_ticks_and_labels(::Plots.Subplot{Plots.GRBackend}, ::Plots.Axis, ::Nothing) at /Users/skainswo/.julia/packages/Plots/SMabf/src/axes.jl:127 (repeats 2 times)
 [8] get_ticks(::Plots.Subplot{Plots.GRBackend}, ::Plots.Axis) at /Users/skainswo/.julia/packages/Plots/SMabf/src/axes.jl:231
 [9] _update_min_padding!(::Plots.Subplot{Plots.GRBackend}) at /Users/skainswo/.julia/packages/Plots/SMabf/src/backends/gr.jl:894
 [10] iterate at ./generator.jl:47 [inlined]
 [11] _collect(::Array{AbstractLayout,2}, ::Base.Generator{Array{AbstractLayout,2},typeof(Plots._update_min_padding!)}, ::Base.EltypeUnknown, ::Base.HasShape{2}) at ./array.jl:678
 [12] collect_similar(::Array{AbstractLayout,2}, ::Base.Generator{Array{AbstractLayout,2},typeof(Plots._update_min_padding!)}) at ./array.jl:607
 [13] map(::Function, ::Array{AbstractLayout,2}) at ./abstractarray.jl:2072
 [14] _update_min_padding!(::Plots.GridLayout) at /Users/skainswo/.julia/packages/Plots/SMabf/src/layouts.jl:310
 [15] prepare_output(::Plots.Plot{Plots.GRBackend}) at /Users/skainswo/.julia/packages/Plots/SMabf/src/plot.jl:184
 [16] display(::Plots.PlotsDisplay, ::Plots.Plot{Plots.GRBackend}) at /Users/skainswo/.julia/packages/Plots/SMabf/src/output.jl:149
 [17] display(::Any) at ./multimedia.jl:323
 [18] top-level scope at /Users/skainswo/dev/research/julia/odecontrol/plots_bug.jl:38
 [19] include(::Module, ::String) at ./Base.jl:377
 [20] exec_options(::Base.JLOptions) at ./client.jl:288
 [21] _start() at ./client.jl:484
in expression starting at /Users/skainswo/dev/research/julia/odecontrol/plots_bug.jl:38

Does this affect all backends? Could I switch backends to hack around this issue?

Edit: Well using the pyplot backend doesn't do any good...

[isentropic]

How do you think should be it be drawn, if value - error < 0. That is how should the errorbar look like in case it protrudes into an unreachable log domain? We can fix this, but do not see how errorbar can extent into infinity

[isentropic]

I believe your plot would just work if you use 0 errorbars

[mcabbott]

I expected that the negative yaxis also could be scaled to log10.

This would be very strange, IMO. Negative numbers are infinitely far away, and if you'd like a second plot pane then that is your choice. Or you could plot negative values on the same axes in a different colour.

However I think simply not plotting those points would be much better than giving an error. For instance plot([1,2,NaN,3,4,Inf]) is happy to ignore some points. I tried at some point to figure out where to call NaNMath.jl's version of log10 instead...

You can do this, but note that (unlike NaNs) there are some extra horizontal line segments drawn:

y = cumsum(randn(1000));
mask = y .> 0;
plot((1:1000)[mask], y[mask], yaxis=:log10, lab="positive")
plot!((1:1000)[.!(mask)], -y[.!(mask)], lab="negative", legend=:bottomright)

How do you think should be it be drawn, if value - error < 0.

I think such points should not be dropped, but the error bar drawn should extend all the way to the edge of the picture.

[isentropic]

Yeah, those extra horizontal lines connecting nearest points are annoying. I could suggest using st=:scatter and make the marker size fairly small. Or use st=:step where you can see the missing points better.
But the best solution is to separate your data into discrete lines yourself.

[isentropic]

Or

[mcabbott]

Perhaps I should have written plot(ifelse.(y.>0, y, NaN), yaxis=:log10), to avoid the lines.

[mcabbott]

This: mcabbott/RecipesPipeline.jl@3d4d7d4 should allow this to work:

plot(cumsum(randn(1000)), yaxis=(:log10, [0.01, 30]))

And this: mcabbott@cad5a96 is a too-crude attempt to allow it to set the axis range.

[Moelf]

#2962 should have closed this?