Way to make contour plot from 3 vectors?

General Usage
1

Just ran into this problem with Julia. It is a repeat of these questions though:

Summarized, you have:

  • two vectors that hold points on an irregularly spaced grid: (x,y)
  • one vector with function values at each of those points: z = f(x,y)

How do you plot a heatmap of this?

// using Plots.jl (and maybe some Interpolation package)

2

Here’s a Gadfly example:
https://github.com/GiovineItalia/Gadfly.jl/pull/1150#issue-190151930
The function rsm there could be used to give you a regular grid for any plotting package. Note it was written for Julia 0.6 or earlier, so would need updating for Julia 0.7/1.x.

3

What would need updating besides linspaces? This seems like a relatively common modeling problem

4

Here is a Julia 1.x update for a response surface model:

import Cairo
using DataFrames, Gadfly, RDatasets

module t
using DataFrames, Statistics

VecOrSA = Union{Vector, SubArray}
expand_grid(xv::AbstractRange, yv::AbstractRange) = 
    vcat([[x y] for x in xv, y in yv]...)

function rsm(xv::VecOrSA, yv::VecOrSA, zv::VecOrSA, k::Int; 
        xlim::Tuple=extrema(xv), ylim::Tuple=extrema(yv))
    # k is the order of the polynomial surface
    y = Vector{Float64}(zv)
    μx, μy = mean(xv), mean(yv)
    p = reshape([1:k;],1,:)
    X = hcat(ones(length(xv)), (xv.-μx).^p, (yv.-μy).^p)    
    B = X'X \ X'y
    xp = range(xlim[1], xlim[2], length=11)
    yp = range(ylim[1], ylim[2], length=11)
    g = expand_grid(xp, yp)
    X = hcat(ones(size(g,1)), (g[:,1].-μx).^p, (g[:,2].-μy).^p)
    return DataFrame(x=g[:,1], y=g[:,2], z=X*B)
end

end

Doz = dropmissing(dataset("datasets","airquality"), disallowmissing=true)
Doz[:SolarRad] = Doz[:,2] .> 200
labelD = Dict(false=>"SolarRad ≤200", true=>"SolarRad >200")

Dpredict = by(Doz, :SolarRad, [:Wind, :Temp, :Ozone] =>
    x->t.rsm(x.Wind, x.Temp, x.Ozone, 2, xlim=(0, 21), ylim=(50,100)) )

set_default_plot_size(16cm, 8cm)
p = plot(Doz, x=:Wind, y=:Temp, xgroup=:SolarRad, color=:Ozone, 
    Geom.subplot_grid(
        layer( Geom.point),
        layer(Dpredict, x=:x, y=:y, z=:z, xgroup=:SolarRad, Geom.contour),
        Guide.xticks(ticks=[0:5:20;]) ),
    Scale.color_continuous(minvalue=0, maxvalue=200),
    Scale.xgroup(labels=x->labelD[x], levels=[false, true])
    )

rsm

1 Like