Colors lost in 3D scatter plot

Question

For a project I am making some 3D scatter plots with the three corresponding projections under it. I use different colors to indicate a fourth parameter. First I plot data with a certain color and then I overplot that with other data with a different color, so that in the end the order is such that I can see everything as I want:

This is what I want!

In the beginning this worked fine, but when I try to do the same thing with slightly different data, the colors get messed up. The colors shown in the projections are the right ones, but some of them are missing in the 3D plot so they don't match anymore:

Colors messed up :(

When I rotate the 3D plot in a funny way, the colors are recovered and I can see them as they were supposed to be:

Funny rotation

However, I don't want a 3D plot that is rotated in a funny way, because the axes get messed up and it's impossible to read it properly like that.

I found one solution to the problem here: plotting 3d scatter in matplotlib. It basically says that I should replace my ax.scatter(X,Y) with ax.plot(X,Y,'o'). When I do this the colors are shown the way they were supposed to be, but the plot is much messier and uglier this way. Basically I just want to be able to do this with a scatter plot.

Does anyone know how to solve this?

Here's a minimum example of my code, for only two colors:

from mpl_toolkits.mplot3d import art3d
import numpy as np
from mpl_toolkits.mplot3d import Axes3D
import matplotlib.pyplot as plt
from matplotlib import gridspec

art3d.zalpha = lambda *args:args[0]

numcols = 20
percentage = 50

def load(Td, pc):
    T = np.load(str(pc) + 'pctTemperaturesTd=' + str(Td) + '.npy')
    D = np.load(str(pc) + 'pctDensitiesTd=' + str(Td) + '.npy')
    CD = np.load(str(pc) + 'pctColDensitiesTd=' + str(Td) + '.npy')
    return T, D, CD

def colors(ax):
    colors = np.zeros((numcols, 4))
    cm = plt.get_cmap('gist_rainbow')
    ax.set_color_cycle([cm(1.*i/numcols) for i in range(numcols)])
    for i in range(numcols):
        color = cm(1.*i/numcols)
        colors[i,:] = color
    return colors

# LOAD DATA
T10, D10, CD10 = load(10, percentage)
T200, D200, CD200 = load(200, percentage)

# 3D PLOT
fig = plt.figure(1)
gs = gridspec.GridSpec(4, 4) 

ax = fig.add_subplot(gs[:-1,:-1], projection='3d')
colours = colors(ax)

ax.plot(T200/100., np.log10(D200), np.log10(CD200), '*', markersize=10,color=colours[10], mec = colours[10], label='Td = 200', alpha=1)
ax.plot(T10/100., np.log10(D10), np.log10(CD10), '*', markersize=10,color=colours[0], mec = colours[0], label='Td = 10', alpha=1)

ax.set_xlabel('\nTg/100', fontsize='x-large')
ax.set_ylabel('\nlog(nH)', fontsize='x-large')
ax.set_zlabel('\nlog(colDen)', fontsize='x-large')
ax.set_xlim(0,5)
#ax.set_zlim(0,)
ax.set_ylim(2,6)

# PROJECTIONS

# Tg, nH
ax2 = fig.add_subplot(gs[3,0])

ax2.scatter(T200/100., np.log10(D200), marker='*', s=10, color=colours[10], label='Td = 200', alpha=1, edgecolor=colours[10])
ax2.scatter(T10/100., np.log10(D10), marker='*', s=10, color=colours[0], label='Td = 10', alpha=1, edgecolor=colours[0])

ax2.set_xlabel('Tg/100')
ax2.set_ylabel('log(nH)')
ax2.set_xlim(0,6)

# Tg, colDen 
ax3 = fig.add_subplot(gs[3,1])

ax3.scatter(T200/100., np.log10(CD200), marker='*', s=10, color=colours[10], label='Td = 200', alpha=1, edgecolor=colours[10])
ax3.scatter(T10/100., np.log10(CD10), marker='*', s=10, color=colours[0], label='Td = 10', alpha=1, edgecolor=colours[0])

ax3.set_xlabel('Tg/100')
ax3.set_ylabel('log(colDen)')
ax3.set_xlim(0,6)

# nH, colDen
ax4 = fig.add_subplot(gs[3,2])

ax4.scatter(np.log10(D200), np.log10(CD200), marker='*', s=10, color=colours[10], label='Td = 200', alpha=1, edgecolor=colours[10])
ax4.scatter(np.log10(D10), np.log10(CD10), marker='*', s=10, color=colours[0], label='Td = 10', alpha=1, edgecolor=colours[0])

ax4.set_xlabel('log(nH)')
ax4.set_ylabel('log(colDen)')

# LEGEND
legend = fig.add_subplot(gs[:,3])

text = ['Td = 10', 'Td = 20', 'Td = 30', 'Td = 40', 'Td = 50', 'Td = 60', 'Td = 70', 'Td = 80', 'Td = 90', 'Td = 100', 'Td = 110', 'Td = 120', 'Td = 130', 'Td = 140', 'Td = 150', 'Td = 160', 'Td = 170', 'Td = 180', 'Td = 190', 'Td = 200']

array = np.arange(0,2,0.1)
for i in range(len(array)):
    legend.scatter(0, i, marker='*', s=100, c=colours[numcols-i-1], edgecolor=colours[numcols-i-1])
    legend.text(0.3, i-0.25, text[numcols-i-1])
    legend.set_xlim(-0.5, 2.5)
    legend.set_ylim(0-1, i+1)

legend.axes.get_xaxis().set_visible(False)
legend.axes.get_yaxis().set_visible(False)

gs.tight_layout(fig)

plt.show()

Can you share code? Can you specify data structures used, or represent the shape of data you are using? What's more, how different are the two data sets that you are using (are some features transposed in one data set compared to the other? - this would explain the appearance of red when you rotate charts.. but it is still weird)? — kasparg, May 21 '15 at 12:18
I included my code. The data sets are just numpy arrays, the second one should be a subset of the first one so to say, it only consists of points that also appear in the first one but they are just fewer. Apart from that nothing changes. I think it's a bug, I found no way of changing the order in which it shows my data. For now I just solved it using 'plot' instead of 'scatter' after all. — Nikki, May 22 '15 at 13:16
First thing I'd look at is [`zorder`](http://matplotlib.org/examples/pylab_examples/zorder_demo.html). — Matt Hall, Nov 23 '15 at 14:03
@Nikki does it happen with all [backends](http://matplotlib.org/faq/usage_faq.html#what-is-a-backend)? Could be a work-around to keep the scatter — rll, Nov 29 '15 at 19:05
As far as I can see, the "funny" angle that shows your colors, is from the other side. So the problem is your "wrong" markers are on top. As kwinkunks suggest, check out zorder. — Koebmand STO, Jan 13 '16 at 10:16
Instead of zorder, have you tried sorting the data differently? I'm thinking of sorting the elements farthest from the camera so they end up drawn first and so on. — Hugo O. Rivera, Jan 31 '16 at 17:33
After sorting the data, maybe you can pass all your data and a large color array to `.plot()`. [See this 2D example](http://matplotlib.org/examples/shapes_and_collections/scatter_demo.html). I'm not sure if this works in 3D. — Hugo O. Rivera, Jan 31 '16 at 17:50
The `load()` function seems to depend on six `.npy` files. Can you post some minimal version of those files, or include some sample data in the script? — Don Kirkby, Feb 06 '16 at 03:53
My impression is that overlapping is the issue as mentioned in other comments. I would suggest down sampling or adding a jitter or dither to your data to disambiguate data points, see [this answer] (http://stackoverflow.com/questions/8671808/matplotlib-avoiding-overlapping-datapoints-in-a-scatter-dot-beeswarm-plot) Sorting out the order may allow displaying of all the colors but it would end up being a less accurate representation of the data, e.g. areas would show no TD=80 even though regions contain those results. This may be more or less a concern given the nature of your work. — Seth_P, Feb 28 '16 at 17:44
Please may you provide the '50pctTemperaturesTd=10.npy' file content for debug purpose. — A. STEFANI, Mar 10 '16 at 16:29

score 1 · Answer 1 · answered Apr 01 '16 at 20:19

1

Rather than using ax.plot(x,y, 'o') try ax.plot(x,y,'.') or ax.plot(x,y,'*'. The 'o' is specifying the marker to use, and the 'o' marker is a large filled circle, which is why your plot looks ugly.

answered Apr 01 '16 at 20:19

wesanyer

982
1
6
27

Colors lost in 3D scatter plot

1 Answers1