Jitter (feature request)

jdm204 commented 3 years ago

AlgebraOfGraphics is very nice!

I was wondering if support for jitter is planned, i.e. adding a bit of randomness to the positions of scatter points to help with overplotting.

For example, when plotting continuous data (y) against categorical (x) it can be nice to jitter the points in the horizontal x direction:

df = PalmerPenguins.load() |> DataFrame |> dropmissing
data(df) * mapping(:sex, :body_mass_g) * (visual(BoxPlot) + visual(Scatter)) |> draw

In ggplot2 for example this is implemented as parameters to the point visual and by a jitter visual like geom_jitter(width=0.4, height=0).

SimonDanisch commented 3 years ago

@kimlaberinto created a function for this:

function create_jitter_array(length_data_array; jitter_width = 0.1, clamped_portion = 0.1)
    jitter_width < 0 && ArgumentError("`jitter_width` should be positive.")
    !(0 <= clamped_portion <= 1) || ArgumentError("`clamped_portion` should be between 0.0 to 1.0")

    # Make base jitter, note base jitter minimum-to-maximum span is 1.0
    base_min, base_max = (-0.5, 0.5)
    jitter = [rand_localized(base_min, base_max) for _ in 1:length_data_array]

    # created clamp_min, and clamp_max to clamp a portion of the data
    @assert (base_max - base_min) == 1.0
    @assert (base_max + base_min) / 2.0 == 0
    clamp_min = base_min + (clamped_portion / 2.0)
    clamp_max = base_max - (clamped_portion / 2.0)

    # clamp if need be
    clamp!(jitter, clamp_min, clamp_max)

    # Based on assumptions of clamp_min and clamp_max above
    jitter = jitter * (jitter_width / clamp_max)

    return jitter
end

Maybe we can figure out how to upstream these... This was for a raincloud plot recipe, which might as well live in Makie itself...