matplotlib Slow 3D scatter rotation

Question

I am using matplotlib to scatter plot a 3D matrix of points. I am using the following code:

import pylab as plt
from mpl_toolkits.mplot3d import Axes3D
import numpy as np

my_data = np.random.rand(6500,3)  # toy 3D points

fig = plt.figure()
ax = Axes3D(plt.gcf())
ax.scatter(my_data[:,0],my_data[:,1],my_data[:,2])
plt.show()

It works, so it opens a window where I can see my points.

However if I try to rotate the plot with the mouse (clicking and dragging it) it rotates REALLY slow.

I think 6500 points are not a lot for such a slow and laggy rotation, so I'm wondering if there is any pre-configuration to be done to speed it up.

Note: I tried to use Matlab and I can rotate a way bigger scatter plot without any lag, so it's not a computer limitation.

Can someone run this code and see if also experiences the slow rotation?

EDIT: Using the System monitor I can see that when rotating the points, only one CPU is used, so matplotlib is not parallelizing the process.

My computer specs:

Intel(R) Core(TM) i5-3550 CPU @ 3.30GHz
Graphic Card: Xeon E3-1200 v2/3rd Gen Core processor Graphics Controller
Memory: 10GB RAM
HDD: Samsung SSD 500 GB

running Ubuntu 16.10

I'm afraid that matplotlib was not designed with speed in mind, and slow performance in your case is not a specific quirk of your local installation - it's the way mplot3d was written. If you want to use fast 3d scatterplots with python you can check out `plotly.py` lib - it's very fast, although its python capabilities are somewhat limiting. — Vlas Sokolov, Apr 03 '17 at 15:25
Echo what @VlasSokolov says - matplotlib is not designed for speedy interactive plotting, especially in 3D (which is not a real full-fledged 3D capability, more just for convenience). You might also look into mayavi for 3D plotting. Also, as a Python tip - I would recommend against `import pylab as plt`. `plt` is typically used for `matplotlib.pyplot` *not* `pylab`. Any examples you may have found that do so are old. Also, using pylab in a script at all is not recommended as good practice - use matplotlib and numpy separately. — Ajean, Apr 03 '17 at 15:56
See https://stackoverflow.com/questions/8955869/why-is-plotting-with-matplotlib-so-slow — thomaskeefe, Jul 18 '17 at 18:03

anon01 · Accepted Answer · 2021-01-28T08:51:27.130

(revised 2021)

Matplotlib was not really designed to be interactive. Plotly (among a few others) is a plotting package that is rather feature complete and uses a webGL backend for scatter3D that will render in your browser (and is blazing fast). More on why you might want to consider it as a go-to replacement at the bottom:

# pip install plotly
import numpy as np
import plotly.graph_objects as go

my_data = np.random.rand(6500,3)  # toy 3D points
marker_data = go.Scatter3d(
    x=my_data[:,0], 
    y=my_data[:,1], 
    z=my_data[:,2], 
    marker=go.scatter3d.Marker(size=3), 
    opacity=0.8, 
    mode='markers'
)
fig=go.Figure(data=marker_data)
fig.show()

I use plotly as a replacement for matplotlib because:

it has great coverage (bar, scatter, line, hist, etc.)
intuitive interactive components that are highly configurable
portability: it can generate html plots which are easily viewed across a network(!) or saved as static file formats
it has a clearer api than matplotlib (for full config, jsonish)
it integrates strongly with pandas (you can use it as the backend)
it is well supported and adoption is increasing

This is nice! I had to drastically reduce the marker size to get anything like your plot, though. — Mohan, Jan 26 '21 at 14:35

matplotlib Slow 3D scatter rotation

1 Answers1

Linked