New colorscheme for Plots.jl 2.0

[isentropic]

@BeastyBlacksmith @t-bltg
So I've been on search on new colorschemes and what we could choose. I think we should also update the color gradient while at it too. Upon hearing feedback I think we should settle with colorblind friendly scheme:
https://juliagraphics.github.io/ColorSchemes.jl/stable/basics/#For-CVD-(color-vision-deficient-or-%22color-blind:)-users
Upon which tol_bright looks good to me
https://juliagraphics.github.io/ColorSchemes.jl/stable/basics/#For-CVD-(color-vision-deficient-or-%22color-blind:)-users
looks good to ne

For the colorgradient, upon watching this and this setting viridis or some of the other 3 would be good in my opinion.

I'm hoping we could reach some consensus here

[isentropic]

These look really good as suggested by @cormullion esp the batlow gradient and catergorical if it is as good as advertised

[BeastyBlacksmith]

@cormullion can I also get the categorial palette of the scientific gradients somehow in ColorSchemes.jl?

[BeastyBlacksmith]

Its hard to judge without examples, so here is some code and pictures:

using StatsPlots
@userplot BackendPlot
@recipe function f(bp::BackendPlot; n = 4, colorgradient = :inferno)
           t = range(0, 3π, length = 100)
           d = rand(3, 3)

           plot_title := "p = $(plotattributes[:color_palette]), cg = $colorgradient"
           layout := n

           @series begin
               subplot := 1
               f = s -> -cos(s) * log(s)
               g = t -> sin(t) * log(t)
               f1 = s -> -cos(s) * log(s) + 5
               g1 = t -> sin(t) * log(t) + 5
               f2 = s -> -cos(s) * log(s) + 10
               g2 = t -> sin(t) * log(t) + 10
               linewidth --> 3
               [f g f1 g1 f2 g2]
           end

           @series begin
               subplot := 2 + (n > 2)
               RecipesBase.recipetype(:groupedbar, d)
           end

           if n > 2
               @series begin
                   subplot := 2
                   line_z := t
                   label := false
                   seriestype := surface
                   seriescolor := colorgradient
                   t, t, (x, y) -> x * sin(x) - y * cos(y)
               end

               @series begin
                   subplot := 4
                   seriestype := contourf
                   seriescolor := colorgradient
                   t, t, (x, y) -> x * sin(x) - y * cos(y)
               end
           end
       end

schemes = [:batlowW, :batlowK, :lipari, :navia, :imola, :devon, :lajolla, :bamako, :davos, :bilbao, :lapaz, :acton]

mkdir("plots_colors")
for scheme in fschemes
     savefig(backendplot(colorgradient = scheme, palette = :tol_bright), joinpath("plots_colors", "$scheme.png"))
end

[BeastyBlacksmith]

I like navia, lajolla, devon and lipari

[cormullion]

ColorSchemes.jl is very minimalist - the SVG swatches in the docs are drawn with Luxor.

Code

using Luxor, ColorSchemes

function draw_scheme(schemename, pos=O;
  swatchwidth=800,
  swatchheight=20)
  @layer begin
      translate(pos)
      cols = colorschemes[schemename].colors
      setline(0.5)
      t = Tiler(swatchwidth, swatchheight, 1, length(cols), margin = 0)
      for (i, c) in enumerate(cols)
          sethue(c)
          box(t, i, :fillstroke)
      end
      sethue("white")
      text(string(schemename), Point(-swatchwidth/2 - 20, 0), halign=:right)
  end
end

schemes = [
  :mpetroff_10,
  :tol_YlOrBr,
  :tol_land_cover,
  :tol_light,
  :mk_15,
  :mpetroff_6,
  :tol_ylorbr,
  :tol_PRGn,
  :tol_medcontrast,
  :tol_prgn,
  :okabe_ito,
  :tol_highcontrast,
  :ibm_cvd,
  :tol_BuRd,
  :mk_12,
  :tol_rainbow,
  :tol_muted,
  :tol_iridescent,
  :tol_bright,
  :tol_incandescent,
  :tol_vibrant,
  :tol_bu_rd,
  :tol_sunset,
  :mpetroff_8,
  :mk_8,
  :tol_nightfall,
]

Drawing(650, 700,  "/tmp/cs.svg")
origin()
background("#121212")
translate(boxtopcenter() + (0, 30))
for cs in schemes
  draw_scheme(cs, swatchwidth=450)
  translate(0, 25)
end
finish()
preview()

(I would use Plots.jl if I were a Plots.jl user. :))

I think seeing them in action -in a plot - is a good idea. But don't forget Dark Mode (like often happens)!

[BeastyBlacksmith]

ColorSchemes.jl is very minimalist - the SVG swatches in the docs

Thats not what I meant. On their homepage they have sequential continuous palettes, discrete palettes and categorial palettes. But in ColorSchemes.jl I can only find the first two.

[isentropic]

lajola looks great and inline with notion of heatmap (as in heat). tol_bright and lajola are a good candidate for the new default. Maybe we should see about this batlow and the categorical variant of it too

[cormullion]

@BeastyBlacksmith Ah, I’m with you now.

ColorSchemes.jl is mostly Crameri-v5.0 (ca 2020), with minor corrections. Incorporating the complete Crameri-v8.0 set wouldn’t be too much work - could be done by next week.

Version 3.26 includes the -S scientific schemes now - eg batlowS.

[isentropic]

batlowS looks weird in that the primary color isnt bright blue

https://juliagraphics.github.io/ColorSchemes.jl/stable/catalogue/#Color-Vision-Deficient-friendly-schemes :tol_bright looks good
I like :tol_bright and :lajolla colors. I would also go one step further and recommend :managua usage for diverging colorschemes.

I also like :seaborn_colorblind as it resembles matplotlib default and would be more famliar to most users.

They also have other variants including gradients that look good. Link here too.

Overall seaborn colors and :tol_bright + scientific gradients both look good to me. I wouldn't mind having either of those. I'm slightly leaning toward seaborn as the look a little more conventional (close to matplotlib).

[BeastyBlacksmith]

I'm slightly leaning toward seaborn as the look a little more conventional (close to matplotlib).

I think, being distinct is a good thing.

I made some more examples for palettes using the gradients identified before:

tol_bright

tol_light

seaborn_colorblind

There is something neat about having the same colors in gradient and palette, but this would possibly only work with batlow:

batlow

code

```julia using StatsPlots @userplot BackendPlot @recipe function f(bp::BackendPlot; n = 4, colorgradient = :inferno) t = range(0, 3π, length = 100) d = rand(3, 3)

       plot_title := "p = $(plotattributes[:color_palette]), cg = $colorgradient"
       layout := n

       @series begin
           subplot := 1
           f = s -> -cos(s) * log(s)
           g = t -> sin(t) * log(t)
           f1 = s -> -cos(s) * log(s) + 5
           g1 = t -> sin(t) * log(t) + 5
           f2 = s -> -cos(s) * log(s) + 10
           g2 = t -> sin(t) * log(t) + 10
           linewidth --> 3
           [f g f1 g1 f2 g2]
       end

       @series begin
           subplot := 2 + (n > 2)
           RecipesBase.recipetype(:groupedbar, d)
       end

       if n > 2
           @series begin
               subplot := 2
               line_z := t
               label := false
               seriestype := surface
               seriescolor := colorgradient
               t, t, (x, y) -> x * sin(x) - y * cos(y)
           end

           @series begin
               subplot := 4
               seriestype := contourf
               seriescolor := colorgradient
               t, t, (x, y) -> x * sin(x) - y * cos(y)
           end
       end
   end

schemes = [:lipari, :navia, :devon, :lajolla, :batlow, :batlowK]

mkdir("plots_colors")

for (scheme, palette, background) in Iterators.product(schemes, [:tol_bright, :tol_light, :seaborn_colorblind, :batlowS], [:white, :black])
savefig(backendplot(colorgradient = scheme, palette = palette, bg = background), joinpath("plots_colors", "$(palette)$(scheme)$(background).png"))
end

</details>

[BeastyBlacksmith]

but I think we narrowed the options enough, that I could make a poll on discourse