color2

[Mattriks]

• 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 built the docs and confirmed these changes don't cause new errors

This PR:

• Adds color2. color2 is a second color scale (discrete only, which covers the use cases in Contour plot with all lines in one color #527, Color facetting fails when combined with a contour layer #1098, Exception in apply_scales() when using multiple layers+DataFrame+color keys #1171)
• plus Scale.color2_discrete and Guide.color2key
• color2 has been enabled for Geom.point. color2 may be enabled for other Geoms.
• fixes Contour plot with all lines in one color #527, fixes Color facetting fails when combined with a contour layer #1098, fixes Exception in apply_scales() when using multiple layers+DataFrame+color keys #1171

Example

# plot p1 is the same type of plot that reopened #527

iris = dataset("datasets", "iris")
gp = Geom.polygon(preserve_order=true, fill=true)
p1 = plot(iris, x=:SepalLength, y=:SepalWidth, color2=:Species,
    layer(Geom.density2d(levels=[0.1:0.1:0.4;]),  order=1),
    Geom.point, Scale.color_none,  Guide.color2key(title="", pos=[6.9,5.7]), 
    Theme(default_color="gray", point_size=3pt, key_swatch_shape=Shape.circle, 
        key_position=:inside))
p2 = plot(iris, 
    layer(x=:SepalLength, y=:SepalWidth, color2=:Species, alpha=[1.0]),
    layer(x=:SepalLength, y=:SepalWidth, Stat.density2d(levels=7), gp,  order=1,
        Theme(alphas=[0.2],lowlight_color=identity)), Scale.color_continuous,
    Guide.colorkey, Guide.color2key(title="\n\nIris"), 
    Theme(point_size=3pt, key_swatch_shape=Shape.circle, alphas=[0.7],
    lowlight_color=identity), Guide.ylabel(nothing)
)
hstack(p1, p2)

[bjarthur]

color2 seems like a hack to me. it would be more intuitive to a user i think to permit Guide.colorkey to be used within a layer. how difficult would that be to implement.

@tlnagy thoughts?

[tlnagy]

Agreed. This might be a pragmatic solution, but associating scales/colorkey with layers seems much cleaner, discoverable, and intuitive.

[Mattriks]

It's not a hack. In Gadfly, aesthetics can be transferred between layers.
How would you do this plot, if there was 1 color aesthetic, and Guides needed to be put in layers?

plot(Coord.cartesian(xmin=0, xmax=5),
    layer(x=0.5:3.5, y=0.5*rand(4), color=0:3),
     layer(x=0.5:3.5, y=rand(4), color=5:8),
    layer(x=1:4, y=rand(4), color2=["Group 1"]),
     layer(x=1:4, y=rand(4), color2=["Group 2"]),
    Theme(point_size=5pt, color2s=[colorant"hotpink", colorant"violet"])
)

[Mattriks]

Also, aesthetic color2 and Scale.color2_discrete follows the suggestion of x2 and Scale.x2_continuous in implementing secondary x,y axes (#373 (comment)) (which @bjarthur gave thumbs up! - and has since removed). Allowing "scales per layer" was seen to be problematic, earlier in that discussion (#373 (comment))

[codecov-io]

Codecov Report

Merging #1436 into master will decrease coverage by 0.88%.
The diff coverage is 80.00%.

@@            Coverage Diff             @@
##           master    #1436      +/-   ##
==========================================
- Coverage   90.26%   89.38%   -0.89%     
==========================================
  Files          39       39              
  Lines        4140     4389     +249     
==========================================
+ Hits         3737     3923     +186     
- Misses        403      466      +63     

Impacted Files	Coverage Δ
src/aesthetics.jl	77.77% <ø> (-2.06%)	⬇️
src/data.jl	0.00% <ø> (ø)
src/theme.jl	65.07% <ø> (-4.42%)	⬇️
src/guide/keys.jl	79.86% <69.56%> (-3.19%)	⬇️
src/Gadfly.jl	76.00% <100.00%> (-1.94%)	⬇️
src/geom/point.jl	100.00% <100.00%> (ø)
src/scale.jl	97.58% <100.00%> (+0.08%)	⬆️
src/geom/rectbin.jl	80.55% <0.00%> (-6.95%)	⬇️
src/geom/label.jl	96.26% <0.00%> (-2.74%)	⬇️
... and 24 more

---

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 40c708b...36a2555. Read the comment docs.

[bjarthur]

my thumbs up in #373 was not for y2 but for the ability to have multiple axes in general. there, i agreed with iq2luc's voltage & current plot (as did three others) that the two axes units need not even be convertible. the API needed there needs to be able to handle the case for even more than two axes, as in the OP which had four. i don't think it reasonable to have an x3 and x4 to handle this, as then you'd be limited to a fixed number.

i didn't realize aes could be tranferred btw layers! seems quite useful. nevertheless, from a user perspective, i think a cleaner API would be to permit sth like this:

plot(Coord.cartesian(xmin=0, xmax=5),
     layer(x=0.5:3.5, y=0.5*rand(4), color=0:3),
     layer(x=0.5:3.5, y=rand(4), color=5:8),
     (layer(x=1:4, y=rand(4), color=["Group 1"]),
      layer(x=1:4, y=rand(4), color=["Group 2"]),
      Theme(colors=[colorant"hotpink", colorant"violet"])),
     Theme(point_size=5pt)
)

don't know how hard that would be to implement, but the technical debt taken on by color2 would be too burdensome to refactor later on IMO.

[Mattriks]

That's an interesting suggestion. It looks like a new plot element i.e. SuperLayer <: Element, and a SuperLayer can have its own layers, scales, guides, themes (and there could be some inheritance from the plot).