heatmap(density) from scattered dots?

[xindi-dumbledore]

Hi~ I'm wondering whether there is a easy way to plot "density" from pure dots? So I'm able to make a scatter plot from dots but what I would like to have is to get a density heat map. But from my understanding, the current heat map function needs a dictionary feed in.

Thanks a lot!

[marcharper]

There's no built in function like matplotlib provides. The easiest way is to write an interpolation function and use heatmapf or to compute your own dictionary.

[fbragman]

I have tried to compute a 3D histogram based on scatter data of the format [0, 0.8, 0.2], [0.3, 0.4 ,0.3] etc.. I then smooth this histogram with a Gaussian kernel. I finally create a dictionary of the smoothed histogram pointing to each coordinate e.g. (0, 0, 0), (0, 0, 1) ... (1, 1, 1) to be mapped to the simplex space.

data.txt

xyz = np.loadtxt('data.txt', delimiter=',')
nbins = 11
import numpy as np
H, b = np.histogramdd((xyz[:, 0], xyz[:, 1], xyz[:, 2]),
                      bins=(nbins, nbins, nbins), range=((0, 1), (0, 1), (0, 1)))
H = H / np.sum(H)

# 3D smoothing and interpolation
from scipy.ndimage.filters import gaussian_filter
kde = gaussian_filter(H, sigma=2)
interp_dict = dict()
binx = np.linspace(0, 1, nbins)
for i, x in enumerate(binx):
    for j, y in enumerate(binx):
        for k, z in enumerate(binx):
            interp_dict[(i/10, j/10, k/10)] = kde[i, j, k]

fig, tax = ternary.figure(scale=1)
tax.heatmap(interp_dict)
tax.show()

I've attached my result. I'm not entirely too sure what is going on and whether I am calling the scale variable correctly. My understanding was that if scale=1, (i, j, k) had to sum to 1. However, looking at the heatmap example here, the x that is passed to the heatmap function f sum to 1 and not 10.

Any help would be appreciate @marcharper and thanks again for a beautiful package :)

[marcharper]

For heatmaps i + j + k = scale. These functions create a partition of the simplex and color each sub-polygon (triangles or hexagons) with the lattice point value. The scale parameter is how it defines the lattice.

I think you can simply scale up your points -- for scale = 1 there's only one subtriangle (0, 0, 1), hence the monocolor plot. Try scale = 10 and interp_dict[(i, j, k)] = kde[i, j, k].

[fbragman]

Yup! Works perfectly - see below.

Thanks again - your package is really helping my upcoming paper submissions

[marcharper]

Glad to hear it! Let me know when you get published and I'll add your paper to the citations page.

[ShiveshM]

From here, is there a way to scale back the labels on the axis so they go between 0 and 1?

[marcharper]

Yes, you can directly overwrite the axes' values using matplotlib's functions for the outer plot (the usual x-axis and y-axis), or the analogous ternary functions if you are using the ternary axes markings on the three triangle sides. See the example here for the matplotlib case.

[manuelbaum]

This was a very useful piece of code, I think it would be great if it would be part of the ternary codebase!

[nehasrikn]

+1 to this! Supremely helpful, and definitely something that would be a nice addition if possible!

This was a very useful piece of code, I think it would be great if it would be part of the ternary codebase!