Backends

Backends are the lifeblood of Plots, and the diversity between features, approaches, and strengths/weaknesses was one of the primary reasons that I started this package.

For those who haven't had the pleasure of hacking on 15 different plotting APIs: first, consider yourself lucky. However, you will probably have a hard time choosing the right backend for your task at hand. This document is meant to be a guide and introduction to make that choice.

Persistent backend selection

Plots uses the Preferences mechanism to make the default backend choice persistent across julia restart.

$ JULIA_PKG_PRECOMPILE_AUTO=0 julia -e 'import Plots; Plots.set_default_backend!(:pythonplot)'
$ julia  # restart, show persistent mode
julia> using Plots
[ Info: Precompiling Plots [91a5bcdd-55d7-5caf-9e0b-520d859cae80]
[ Info: PythonPlot  # precompiles for this backend
julia> plot(1:2) |> display  # uses `PythonPlot` by default

You can clear preferences with Plots.set_default_backend!(). Alternatively, one can use the environment variable PLOTS_DEFAULT_BACKEND to select the default backend (but this will need to trigger manual precompilation using Base.compilecache(Plots)).

At a glance

My favorites: GR for speed, Plotly(JS) for interactivity, UnicodePlots for REPL/SSH and PythonPlot otherwise.

If you require...then use...
featuresGR, PythonPlot, Plotly(JS), Gaston
speedGR, UnicodePlots, InspectDR, Gaston
interactivityPythonPlot, Plotly(JS), InspectDR
beautyGR, Plotly(JS), PGFPlots/ PGFPlotsX
REPL plottingUnicodePlots
3D plotsGR, PythonPlot, Plotly(JS), Gaston
a GUI windowGR, PythonPlot, PlotlyJS, Gaston, InspectDR
a small footprintUnicodePlots, Plotly
backend stabilityPythonPlot, Gaston
plot+data -> .hdf5 fileHDF5

Of course this list is rather subjective and nothing in life is that simple. Likely there are subtle tradeoffs between backends, long hidden bugs, and more excitement. Don't be shy to try out something new !


GR

The default backend. Very fast with lots of plot types. Still actively developed and improving daily.

Pros:

  • Speed
  • 2D and 3D
  • Standalone or inline

Cons:

  • Limited interactivity

Primary author: Josef Heinen (@jheinen)

Fine tuning

It is possible to use more features of GR via the extra_kwargs mechanism.

using Plots; gr()

x = range(-3, 3, length=30)
surface(
  x, x, (x, y)->exp(-x^2 - y^2), c=:viridis, legend=:none,
  nx=50, ny=50, display_option=Plots.GR.OPTION_SHADED_MESH,  # <-- series[:extra_kwargs]
)

Supported :subplot :extra_kwargs

KeywordDescription
legend_hfactorVertical spacing factor for legends
legend_wfactorMultiplicative factor influencing the legend width

Supported :series :extra_kwargs

Series TypeKeywordDescription
:surfacenxNumber of interpolation points in the x direction
:surfacenyNumber of interpolation points in the y direction
:surface, :wireframedisplay_optionsee GR doc

Plotly / PlotlyJS

These are treated as separate backends, though they share much of the code and use the Plotly JavaScript API. plotly() is the only dependency-free plotting option, as the required JavaScript is bundled with Plots. It can create inline plots in IJulia, or open standalone browser windows when run from the Julia REPL.

plotlyjs() is the preferred option, and taps into the great functionality of Spencer Lyon's PlotlyJS.jl. Inline IJulia plots can be updated from any cell... something that makes this backend stand out. From the Julia REPL, it taps into Blink.jl and Electron to plot within a standalone GUI window... also very cool. Also, PlotlyJS supports saving the output to more formats than Plotly, such as EPS and PDF, and thus is the recommended version of Plotly for developing publication-quality figures.

"backends_plotlyjs.png"

Pros:

Cons:

  • No custom shapes
  • JSON may limit performance

Primary PlotlyJS.jl author: Spencer Lyon (@spencerlyon2)

MathJax

Plotly needs to load MathJax to render LaTeX strings, therefore passing extra keywords with extra_kwargs = :plot is implemented. With that it is possible to pass a header to the extra include_mathjax keyword. It has the following options:

  • include_mathjax = "" (default): no mathjax header
  • include_mathjax = "cdn" include the standard online version of the header
  • include_mathjax = "<filename?config=xyz>" include a user-defined file

These can also be passed using the extra_plot_kwargs keyword.

using LaTeXStrings
plotlyjs()
plot(
    1:4,
    [[1,4,9,16]*10000, [0.5, 2, 4.5, 8]],
    labels = [L"\alpha_{1c} = 352 \pm 11 \text{ km s}^{-1}";
              L"\beta_{1c} = 25 \pm 11 \text{ km s}^{-1}"] |> permutedims,
    xlabel = L"\sqrt{(n_\text{c}(t|{T_\text{early}}))}",
    ylabel = L"d, r \text{ (solar radius)}",
    yformatter = :plain,
    extra_plot_kwargs = KW(
        :include_mathjax => "cdn",
        :yaxis => KW(:automargin => true),
        :xaxis => KW(:domain => "auto")
   ),
)
"plotly_mathjax.html"

Fine tuning

It is possible to add additional arguments to the plotly series and layout dictionaries via the extra_kwargs mechanism. Arbitrary arguments are supported but one needs to be careful since no checks are performed and thus it is possible to unintentionally overwrite existing entries.

For example adding customdata can be done the following way scatter(1:3, customdata=["a", "b", "c"]). One can also pass multiple extra arguments to plotly.

pl = scatter(
    1:3,
    rand(3),
    extra_kwargs = KW(
        :series => KW(:customdata => ["a", "b", "c"]),
        :plot => KW(:legend => KW(:itemsizing => "constant"))
    )
)

PythonPlot

A Julia wrapper around the popular python package Matplotlib. It uses PythonCall.jl to pass data with minimal overhead.

2023-07-25T15:45:29.864777 image/svg+xml Matplotlib v3.7.2, https://matplotlib.org/

Pros:

  • Tons of functionality
  • 2D and 3D
  • Mature library
  • Standalone or inline
  • Well supported in Plots

Cons:

  • Uses Python
  • Dependencies frequently cause setup issues

Primary author: Steven G Johnson (@stevengj)

Fine tuning

It is possible to use more features of matplotlib via the extra_kwargs mechanism. For example, for a 3D plot, the following example should generate a colorbar at a proper location; without the extra_kwargs below, the colorbar is displayed too far right to see its ticks and numbers. The four coordinates in the example below, i.e., [0.9, 0.05, 0.05, 0.9] specify the colorbar location [ left, bottom, width, height ]. Note that for 2D plots, this fine tuning is not necessary.

using Plots; pythonplot()

x = y = collect(range(-π, π; length = 100))
fn(x, y) = 3 * exp(-(3x^2 + y^2)/5) * (sin(x+2y))+0.1randn(1)[1]
surface(x, y, fn, c=:viridis, extra_kwargs=Dict(:subplot=>Dict("3d_colorbar_axis" => [0.9, 0.05, 0.05, 0.9])))
2023-07-25T15:45:37.776512 image/svg+xml Matplotlib v3.7.2, https://matplotlib.org/

Supported :subplot :extra_kwargs

KeywordDescription
3dcolorbaraxisSpecifying the colorbar location [ left, bottom, width, height ] for a 3D plot

PGFPlotsX

LaTeX plotting, based on PGF/TikZ.

Successor backend of PGFPlots backend.

Has more features and is still in development otherwise the same.

Tip

To add save a standalone .tex file including a preamble use attribute tex_output_standalone = true in your plot command.

Pros:

  • Nice looking plots
  • Lots of functionality (though the code is still WIP)

Cons:

  • Tricky to install
  • Heavy-weight dependencies

Authors:

  • PGFPlots: Christian Feuersanger
  • PGFPlotsX.jl: Kristoffer Carlsson (@KristofferC89), Tamas K. Papp (@tpapp)
  • Plots <–> PGFPlotsX link code: Simon Christ (@BeastyBlacksmith), based on the code of Patrick Kofod Mogensen (@pkofod)

LaTeX workflow

To use the native LaTeX output of the pgfplotsx backend you can save your plot as a .tex or .tikz file.

using Plots; pgfplotsx()
pl  = plot(1:5)
pl2 = plot((1:5).^2, tex_output_standalone = true)
savefig(pl,  "myline.tikz")    # produces a tikzpicture environment that can be included in other documents
savefig(pl2, "myparabola.tex") # produces a standalone document that compiles by itself including preamble

Saving as .tikz file has the advantage, that you can use \includegraphics to rescale your plot without changing the size of the fonts. The default LaTeX output is intended to be included as a figure in another document and will not compile by itself. If you include these figures in another LaTeX document you need to have the correct preamble. The preamble of a plot can be shown using Plots.pgfx_preamble(pl) or copied from the standalone output.

Fine tuning

It is possible to use more features of PGFPlotsX via the extra_kwargs mechanism. By default it interprets every extra keyword as an option to the plot command. Setting extra_kwargs = :subplot will treat them as an option to the axis command and extra_kwargs = :plot will be treated as an option to the tikzpicture environment.

For example changing the colormap to one that is native to pgfplots can be achieved with the following. Like this it is possible to keep the preamble of latex documents clean.

using Plots; pgfplotsx()
surface(range(-3,3, length=30), range(-3,3, length=30),
        (x, y) -> exp(-x^2-y^2),
        label="",
        colormap_name = "viridis",
        extra_kwargs =:subplot)

Further more additional commands or strings can be added via the special add keyword. This adds a square to a normal line plot:

plot(1:5, add = raw"\draw (1,2) rectangle (2,3);", extra_kwargs = :subplot)

UnicodePlots

Simple and lightweight. Plot directly in your terminal. You won't produce anything publication quality, but for a quick look at your data it is awesome. Allows plotting over a headless node (SSH).

Pros:

  • Minimal dependencies
  • REPL plotting
  • Lightweight
  • Fast

Cons:

  • Limited precision, density

Primary author: Christof Stocker (@Evizero)

Fine tuning

It is possible to use more features of UnicodePlots via the extra_kwargs mechanism.

using Plots; unicodeplots()

extra_kwargs = Dict(:subplot=>(; border = :bold, blend = false))
p = plot(1:4, 1:4, c = :yellow; extra_kwargs)
plot!(p, 2:3, 2:3, c = :red)

Supported :subplot :extra_kwargs

KeywordDescription
widthPlot width
heightPlot height
projection3D projection (:orthographic, perspective)
zoom3D zoom level
up3D up vector (azimuth and elevation are controlled using Plots.jl's camera)
canvasCanvas type (see Low-level Interface)
borderBorder type (:solid, :bold, :dashed, :dotted, :ascii, :none)
blendToggle canvas color blending (true / false)

Supported :series :extra_kwargs

Series TypeKeywordDescription
allcolormapColormap (see Options)
heatmap, spyfix_arToggle fixing terminal aspect ratio (true / false)
surfaceplotzscalez axis scaling
surfaceplotlinesUse lineplot instead of scatterplot (monotonic data)

Gaston

Gaston is a direct interface to gnuplot, a cross platform command line driven plotting utility. The integration of Gaston in Plots is recent (2021), but a lot of features are supported.

Gnuplot Produced by GNUPLOT 5.4 patchlevel 2 0 2 4 6 8 0 1 2 3 4 5 y1 y2 y1 y1 y2 y2 0.0 0.2 0.4 0.6 0.8 0.5 1.0 1.5 2.0 2.5 3.0 3.5 y3 gnuplot_plot_2b gnuplot_plot_3b gnuplot_plot_4b y4 gnuplot_plot_6b gnuplot_plot_7b gnuplot_plot_8b y5 gnuplot_plot_10b gnuplot_plot_11b gnuplot_plot_12b y3 y3 gnuplot_plot_2b gnuplot_plot_3b gnuplot_plot_4b y4 y4 gnuplot_plot_6b gnuplot_plot_7b gnuplot_plot_8b y5 y5 gnuplot_plot_10b gnuplot_plot_11b gnuplot_plot_12b 0 2 4 6 8 0 2 4 6 8 -10 -5 0 5 10 15 gnuplot_plot_1c -15 -10 -5 0 5 10 15 20 gnuplot_plot_1d 0 2 4 6 8 0 2 4 6 8 -15 -10 -5 0 5 10 15 20

InspectDR

Fast plotting with a responsive GUI (optional). Target: quickly identify design/simulation issues & glitches in order to shorten design iterations.

Pros:

  • Relatively short load times / time to first plot.
  • Interactive mouse/keybindings.
    • Fast & simple way to pan/zoom into data.
  • Drag & drop &Delta;-markers (measure/display &Delta;x, &Delta;y & slope).
  • Designed with larger datasets in mind.
    • Responsive even with moderate (>200k points) datasets.
    • Confirmed to handle 2GB datsets with reasonable speed on older desktop running Windows 7 (drag+pan of data area highly discouraged).

Cons:

  • Mostly limited to 2D line/scatter plots

Primary author: MA Laforge (@ma-laforge)

HDF5 (HDF5-Plots)

Write plot + data to a single HDF5 file using a human-readable structure that can easily be reverse-engineered.

Write to .hdf5 file

hdf5() # Select HDF5-Plots "backend"
p = plot(...) # Construct plot as usual
Plots.hdf5plot_write(p, "plotsave.hdf5")

Read from .hdf5 file

pythonplot() # Must first select some backend
pread = Plots.hdf5plot_read("plotsave.hdf5")
display(pread)

Pros:

  • Open, standard file format for complex datasets.
  • Human readable (using HDF5view).
  • Save plot + data to a single binary file.
  • (Re)-render plots at a later time using your favourite backend(s).

Cons:

  • Currently missing support for SeriesAnnotations & GridLayout.
    • (Please open an "issue" if you have a need).
  • Not yet designed for backwards compatibility (no proper versioning).
    • Therefore not truly adequate for archival purposes at the moment.
  • Currently implemented as a "backend" to avoid adding dependencies to Plots.jl.

Primary author: MA Laforge (@ma-laforge)


Deprecated backends

PyPlot

matplotlib based backend, using PyCall.jl and PyPlot.jl. Superseded by PythonCall.jl and PythonPlot.jl. Whilst still supported in Plots 1.X, users are advised to transition to the pythonplot backend.

PGFPlots

LaTeX plotting, based on PGF/TikZ.

Tip

To add save a standalone .tex file including a preamble use attribute tex_output_standalone = true in your plot command.

Pros:

  • Nice looking plots
  • Lots of functionality (though the code is still WIP)

Cons:

  • Tricky to install
  • Heavy-weight dependencies

Authors:

  • PGFPlots: Christian Feuersanger
  • PGFPlots.jl: Mykel Kochenderfer (@mykelk), Louis Dressel (@dressel), and others
  • Plots <–> PGFPlots link code: Patrick Kofod Mogensen (@pkofod)

Gadfly

A Julia implementation inspired by the "Grammar of Graphics".

Pros:

  • Clean look
  • Lots of features
  • Flexible when combined with Compose.jl (inset plots, etc.)

Cons:

  • Does not support 3D
  • Slow time-to-first-plot
  • Lots of dependencies
  • No interactivity

Primary author: Daniel C Jones

Immerse

Built on top of Gadfly, Immerse adds some interactivity and a standalone GUI window, including zoom/pan and a cool "point lasso" tool to save Julia vectors with the selected data points.

Pros:

  • Same as Gadfly
  • Interactivity
  • Standalone or inline
  • Lasso functionality

Cons:

  • Same as Gadfly

Primary author: Tim Holy

Qwt

My package which wraps PyQwt. Similar to PyPlot, it uses PyCall to convert calls to python. Though Qwt.jl was the "first draft" of Plots, the functionality is superceded by other backends, and it's not worth my time to maintain.

Primary author: Thomas Breloff

Bokeh

Unfinished, but very similar to PlotlyJS... use that instead.

Winston

Functionality incomplete... I never finished wrapping it, and I don't think it offers anything beyond other backends. However, the plots are clean looking and it's relatively fast.