Plots.jl: Show columnames in legend of a grouped plot of a DataFrame

[crstnbr]

This has been discussed in a Slack thread.

Example from StatPlots.jl github page where the legend entries are fine because the entries in the DataFrame are nice strings:

using RDatasets
school = RDatasets.dataset("mlmRev","Hsb82")
@df school density(:MAch, group = :Sx)

If one has numerical data though, let's say something like

df = DataFrame(L = vcat([fill(i, 10) for i in [8,10]]...), T = vcat([fill(t,5) for t in [0.2, 0.1, 0.2, 0.1]]...), R = vcat([collect(1:5) for _ in 1:4]...), F = [1,2,3,4,5,5,4,3,2,1,1,1,2,2,3,3,3,3,3,3])

one gets a legend just containing numbers

@df df plot(:R, :F, group = (:L, :T), legend=:topleft, marker=:circle, frame=:box, grid=false)

It would be nice to have a dedicated option, which naively would be called showcolumnnamesinlegend , that would lead to this much nicer legend.

[piever]

Just to make sure: this issue only concerns StatPlots and @df or is there something you'd like to add to Plots as well to address this?

[piever]

After some more thinking, the best is to probably allow passing a NamedTuple to group, in which case the names are shown. For example:

plot(a, group = @NT(a = [1,1,2,2], b = [1,2,1,2]))

should display a = 1 b = 1 legends. Then in StatPlots you could do:

@df df plot(:R, :F, group = @NT(L = :L, T = :T), legend=:topleft, marker=:circle, frame=:box, grid=false)

and probably I could also add the {} from Query/JuliaDBMeta where {a, b} is short-hand for @NT(a = a, b = b) as more often then not the name of the column is the name you want to appear, meaning to get the same name as the column you should do:

@df df plot(:R, :F, group = {:L, :T}, legend=:topleft, marker=:circle, frame=:box, grid=false)

[piever]

Closed by #1429 and #132