Quantile bars

[rikhuijzer]

This adds a new statistic to Gadfly, namely for showing quantiles for distributions. In the docs, this currently looks as follows:

Stat.quantile_bars

using DataFrames, Gadfly, Distributions
set_default_plot_size(14cm, 8cm)
n = 400
group = [i ≤ n/2 ? -1 : 1 for i in 1:n]
x = randn(n) .+ group

plot(x=x, color=string.(group), Geom.density, Guide.colorkey(title="", pos=[3.5,0.7]),
    Guide.title("Density with bars showing the central 90% CI"),
    Guide.ylabel("Density"), Coord.cartesian(xmin=-4, xmax=4),
    layer(Stat.density, Geom.polygon(fill=true, preserve_order=true), Theme(alphas=[0.4])),
    layer(Stat.quantile_bars(quantiles=[0.05, 0.95], bar_width=0.05), Geom.bar)
)

Contributor checklist:

• I've updated the documentation to reflect these changes
• I've added an entry to NEWS.md
• I've added and/or updated the unit tests
• I've run the regression tests
• I've squash'ed or fixup'ed junk commits with git-rebase
• I've built the docs and confirmed these changes don't cause new errors

[Mattriks]

Two design issues here are:

1. For a manual bandwidth, bandwidth needs to be specified in both Geom.density and Stat.quantile_bars.
   Note that layers can have more than 1 Geom, if a Stat outputs the aesthetics for those Geoms (see the subsection on layer inputs in Layers). But in this case something like:

   layer(Stat.density(quantiles=[0.05, 0.95]), Geom.line, Geom.bar)

won't work because both Geom.line and Geom.bar want to use the y aesthetic. I don't see a way around this atm.

2. Should Stat.quantile_bars be used with Geom.segment or Geom.bar? For Geom.bar you have to define bar_width,
   but with Geom.segment width can be changed with Theme(line_width), and in the future linewidth should be developed into a proper aesthetic.
   Making Stat.quantile_bar to be used with Geom.segment seems a better way forward.

Also (from point 1) your example can be tidied up using 2 geoms in the Stat.density layer:

n = 400
group = repeat([-1, 1], inner=200)
x = randn(n) .+ group

plot(x=x, color=string.(group), Guide.colorkey(title="", pos=[3.5,0.7]),
  layer(Stat.density, Geom.line, Geom.polygon(fill=true, preserve_order=true), alpha=[0.4]),
    layer(Stat.quantile_bars(quantiles=[0.05, 0.95], bar_width=0.05), Geom.bar)
    Guide.title("Density with bars showing the central 90% CI"),
    Guide.ylabel("Density"), Coord.cartesian(xmin=-4, xmax=4)
)

Note for polygons, alpha=[0.4] was enabled in #1511.
I'll make some other comments on the code.

[codecov-io]

Codecov Report

Merging #1521 (12c8ada) into master (04d931f) will decrease coverage by 0.02%.
The diff coverage is 50.00%.

@@            Coverage Diff             @@
##           master    #1521      +/-   ##
==========================================
- Coverage   89.74%   89.72%   -0.03%     
==========================================
  Files          39       39              
  Lines        4633     4672      +39     
==========================================
+ Hits         4158     4192      +34     
- Misses        475      480       +5     

Impacted Files	Coverage Δ
src/statistics.jl	95.69% <50.00%> (-0.77%)	⬇️
src/scale.jl	95.63% <0.00%> (+0.03%)	⬆️
src/varset.jl	97.87% <0.00%> (+0.04%)	⬆️
src/Gadfly.jl	76.44% <0.00%> (+0.23%)	⬆️
src/misc.jl	65.00% <0.00%> (+0.40%)	⬆️
src/theme.jl	73.61% <0.00%> (+3.74%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 04d931f...12c8ada. Read the comment docs.

[rikhuijzer]

It seems that something upstream is broken. Probably, it is related to the new PooledArrays 1.2.0 release of 6 hours ago.

[rikhuijzer]

Apart from that, thank you for the feedback @Mattriks! I have now implemented your comments. Let me know if you have more comments.

[rikhuijzer]

Thanks to a PooledArrays update by Bogumił Kamiński, the tests passed again in 0aa5feb.

[rikhuijzer]

@Mattriks Is there anything I can do to move this PR forwards?

[Mattriks]

Be patient 😃

[rikhuijzer]

Friendly reminder.

By the way, I understand that reviewing code takes time and is not the greatest job in the world, but it's also quite demotivating to see PRs being stalled. I have many ideas for, hopefully, good PRs for Gadfly. However, with each day of delay on the review, the new PRs become less important for me.

[Mattriks]

Sorry for the delay! Super busy with other stuff, I will attempt to get back to this PR this week. More PRs for Gadfly are very welcome! To discuss your new PRs/ideas opening a new issue is good, or message me directly on Julia discourse!

[rikhuijzer]

Okay 👍 👍 Thanks for letting me know!

[Mattriks]

PR looks good! You should also add a test to /test/testscripts: call it stat_quantile_bars.jl, and don't use random data.
Also run the Regression Tests.

[rikhuijzer]

PR looks good! You should also add a test to /test/testscripts: call it stat_quantile_bars.jl, and don't use random data.
Also run the Regression Tests.

Thanks! I've implemented your suggestions and also ran the regression tests. Of course, there isn't something to check against, but the image looks as expected:

If this PR is merged, would it then also be possible to register a new version in the JuliaRegistries? That would be nice, because I then don't have to use Gadfly.jl#master in an upcoming paper of mine 😄

[rikhuijzer]

Good idea of the hstack, much clearer now:

My bad about the randomness. I copied it from the density example.