Heatmap of data not on a grid

General Usage
, ,
1

I have been trying to make a heatmap of some experimental data using Makie. When the x,y positions of the datapoints are on a grid, this works just fine:

using CairoMakie

i = 1
for x in 1:5
    for y in 1:5
        xys[i,:] = [x, y]
        i = i+1
    end
end
xs = xys[:,1]
ys = xys[:,2]
zs = rand(25)
heatmap(xs,ys,zs)

image
image590×439 5.74 KB

But when the grid points are randomly places (ie. experimental data), the plot gets all wonky:

using CairoMakie

i = 1
for x in 1:5
    for y in 1:5
        xys[i,:] = [x+rand()/5-0.1, y+rand()/5-0.1]
        i = i+1
    end
end
xs = xys[:,1]
ys = xys[:,2]
zs = rand(25)
heatmap(xs,ys,zs)

image
image585×442 7.18 KB

My first thought was to solve this by interpolating the data, but this seems to also work only with data that is already on a grid…

Any ideas on how to solve this?

2

See for instance this example using the KernelInterpolation.jl package.

1 Like
3

You can use PyPlot.jl (or PythonPlot.jl) or GMT.jl, for example, both of which support this by triangulating irregular data onto a mesh and then linearly interpolating from that.

(There are also various packages that can do unstructured-points interpolation, see e.g. 2D Interpolation on an irregular grid, also Dierckx.jl and BasicInterpolators.jl and KernelInterpolation mentioned above.)

4

Your code doesn’t run, but cant you just use tricontourf as an alternative?

2 Likes
5

When you create a heatmap using three vectors in makie, then it infers the structure of the matrix from that data. If your data is not actually on a rectangular grid, the output is necessarily wonky. Pass the x and y values yourself and your z data as a matrix if there are too many missing datapoints. Otherwise tricontourf might be a better fit, as mentioned above.

6

I ended up going with KernelInterpolation.jl as suggested by Rafael. using tricontourf also works, but the plot does not look as nice as a heatmap.
Thanks you all for the suggestions!

2 Likes
7

I’m happy to see you used KernelInterpolation.jl. Would you mind sharing the code you used for future reference?