Dodge in scatter,lines

[behinger]

Problem description

I often encountered the issue, that there is no way to dodge datapoints in a scatter (or less important imho line) plot.

That is going from here:

To here:

Something similar is asked here #230 I think

Proposed solution

I dont understand enough of the internals, but as I understand Makie.jl implements the dodge feature for violin & histograms.

In ggplot2 dodging is possible by specifing a position-function (position = position_dodge(width=0.2)). I was expecting something like
data(df) * mapping(x,y,color="categoryA",dodge="categoryA") * visual(Scatter) to "just work", by doding the colored datapoints.

Let's just hope I missed something ;-)

[behinger]

ok - I needed this again and couldnt find a solution. I implemented a new Makie Recipe Dodge, which allows you to do exactly this.

Please find the gist here: https://gist.github.com/behinger/8df41a3e051979a8e8ee0068f1aac6b8

I don't really know whether @recipe Dodge should be rather in Makie.jl @SimonDanisch what do you think?

I also dont know if this is the nicest way, but one can dodge all kinds of plots. tried with scatter and errorbars. dodging in y instead of x is supported as well.

MWE

begin
	using Makie
	using CairoMakie
	using HTTP
HTTP.download("https://gist.githubusercontent.com/behinger/8df41a3e051979a8e8ee0068f1aac6b8/raw/4c6f44603d4432b19c8abe2955eab4962a8ae45f/MakieDodge.jl",tempdir()) |>include
	
	#----
	x = [1,2,3,  1,2,3.,  3,2,1]
	y = [1.1,1.9,2.8,1,2,3,3.05,2.05,1.05]
	c = ["red","red","red","green","green","green","blue","blue","blue"]
	low,high = repeat([0.2],length(x))
	dodge(x,y,low,high;plot_fun=errorbars!,color=c)
	
	dodge!(x,y;color=c, plot_fun=scatter!)
	current_figure()
	

end

[behinger]

ah and of course with AoG

let
		d = DataFrame(:x=>["x2","x1","x2","x1","x3","x3","x4"],:y=>[2.,1.1,2.1,1.,3.,3.1,4],:c=>["a","a","b","b","b","b","b"])
	d.high = repeat([0.1],nrow(d))
	d.low = repeat([0.1],nrow(d))
	
	data(d)* 
		mapping(color=:c,dodge=:c) *
		(
			mapping(:x,:y,:low,:high,)* visual(Dodge,plot_fun=errorbars!) +
	 		mapping(:x,:y)			  * visual(Dodge)
		) |> draw
	
end

[pdeffebach]

@jkrumbiegel I would be super interested in this feature! Not sure how to add it. But if it could go higher on the priority list I would appreciate it!

[jkrumbiegel]

I'm not sure, yet, either, how exactly this could be implemented. But in principle one can add arbitrary keywords to the aesthetic mapping of a plotting function, which just have to be removed before everything is transferred to Makie. Maybe one could add some sort of generic logic to to_entry which detects the dodge keyword and modifies the data belonging to the X or Y aesthetic accordingly. The generic to_entry functions are here

AlgebraOfGraphics.jl/src/algebra/layers.jl

Lines 314 to 351 in a61e361

	function to_entry(p::ProcessedLayer, categoricalscales::Dictionary, continuousscales::Dictionary)
	entry = to_entry(p.plottype, p, categoricalscales, continuousscales)
	insert!(entry.named, :cycle, nothing)
	for key in (:group, :layout, :row, :col)
	if haskey(entry.named, key)
	delete!(entry.named, key)
	end
	end
	return entry
	end
	function to_entry(P, p::ProcessedLayer, categoricalscales::Dictionary, continuousscales::Dictionary)
	aes_mapping = aesthetic_mapping(p)
	scale_mapping = p.scale_mapping
	positional = map(eachindex(p.positional)) do i
	full_rescale(p.positional[i], i, aes_mapping, scale_mapping, categoricalscales, continuousscales)
	end
	named = map(pairs(p.named)) do (key, value)
	full_rescale(value, key, aes_mapping, scale_mapping, categoricalscales, continuousscales)
	end
	primary = map(pairs(p.primary)) do (key, values)
	if key in (:group, :layout, :col, :row)
	return values
	end
	# seems that there can be vectors here for concatenated layers, but for unconcatenated these should
	# be scalars
	if values isa AbstractVector
	full_rescale(values, key, aes_mapping, scale_mapping, categoricalscales, continuousscales)
	else
	values = [values]
	rescaled = full_rescale(values, key, aes_mapping, scale_mapping, categoricalscales, continuousscales)
	rescaled[]
	end
	end
	Entry(P, positional, merge(p.attributes, named, primary))
	end

. Then the plotting functions wouldn't need to know about dodging at all, they would just receive modified x and y data.

[pdeffebach]

I think the issue is that Makie itself has no concept of dodge for certain things like lines and errorbars. So this has to be fixed upstream, right?

[jkrumbiegel]

Well as I wrote above, if you consider dodging in data space, then you don't really have to add it to plotting functions. You could, but you don't have to, depending on what exactly is plotted and whether there's some advanced interactions like in barplots where width needs to change as well.

[pdeffebach]

Yes sorry, my particular application is bar plots with error bars on the top.

[pdeffebach]

As a concrete example, here is what I want to be able to do with AoG:

julia> df = DataFrame(
           g = ["a", "b", "c", "a", "b", "c"],  
           country = ["X", "X", "X", "Y", "Y", "Y"],
           val = [1.0, 1.5, 2.0, 4.3, 3.6, 2.9], 
           err = [0.2, 0.4, 0.1, 0.3, 0.1, 0.4]);

julia> data(df) * (mapping(:g, :val, color = :country, dodge = :country) * visual(BarPlot) + mapping(:g, :val, :err; dodge = :country) * visual(Errorbars)) |> draw

I get

ERROR: ArgumentError: ProcessedLayer with plot type Plot{Makie.errorbars} did not have :dodge in its AestheticMapping. The mapping was {1 = AlgebraOfGraphics.AesX, 2 = AlgebraOfGraphics.AesY, 3 = AlgebraOfGraphics.AesDeltaY, :color = AlgebraOfGraphics.AesColor}