Marker sizes in Axes3D scatter plot are changing all the time

Bug report

Bug summary

The marker sizes that specified with s= argument in scatter function are wrong, and even worse, the markers are changing their sizes every time the plot is re-rendered (panning or resizing). It seems that the sizes are randomly reassigned to these markers.

Code for reproduction

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

fig = plt.figure()

# matplotlib 3.4
ax = Axes3D(fig, auto_add_to_figure=False)
fig.add_axes(ax)
# matplotlib 3.3.4
# Axes3D(fig)

x = np.random.rand(200)
y = np.random.rand(200)
z = np.random.rand(200)
ax.scatter(x, y, z, s=x*50)

plt.show()

Actual outcome

matplotlib_axes3d_marker_size_issue.mp4

Expected outcome

As shown in the screen recording, the sizes are correct and not changing. matplotlib was 3.3.4 in the following example.

matplotlib_axes3d_marker_size_expected.mp4

Matplotlib version

Operating system: Arch Linux
Matplotlib version (import matplotlib; print(matplotlib.__version__)): 3.4.0
Matplotlib backend (print(matplotlib.get_backend())): Qt5Agg
Python version: 3.9.2
Jupyter version (if applicable): Not using it
Other libraries: numpy 1.20.1

I installed both python, numpy and matplotlib with pacman package manager from Arch's official repository.

@tacaswell

This bisects to e6d9cbb (#18189) Ping @tacaswell.

@jklymak

Well, that is quite bad.

Thank you for the bisect @jklymak Looking at this now.

The issue is that when we re-compute the 3D projection we resort the values based on their depth from the viewer (so markers "in the back" do not render "in the front"). We have a bunch of logic to make sure that the colors also track, but we need the same logic to be added to the size.

if you add color to this example you see that they keep their colors even if their sizes jump around.

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

fig = plt.figure()

# matplotlib 3.4
ax = Axes3D(fig, auto_add_to_figure=False)
fig.add_axes(ax)
# matplotlib 3.3.4
# Axes3D(fig)

x = np.random.rand(200)
y = np.random.rand(200)
z = np.random.rand(200)
ax.scatter(x, y, z, s=x*50, c=x**2)

plt.show()

@QuLogic

Attn @QuLogic This appears to be super broken.

I have been testing with:

th = np.linspace(0, 2*np.pi*6, 2000)

x = np.sin(th)
y = np.cos(th)
z = th
sc = ax.scatter(x, y, z, s=(5 + th), c=th**2)

which has both a color and size structure I can predict what should be.

The color mapping was broken by 4c3d644 via #18480

There are enough coupled changes here (both to the mplot3d code and the colors), I do not think this is something we are going to be able to revert our way out of.

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

fig = plt.figure()



try:
    # matplotlib 3.4
    ax = Axes3D(fig, auto_add_to_figure=False)
    fig.add_axes(ax)
except AttributeError:
    # matplotlib 3.3.4
    ax = Axes3D(fig)
    if ax not in fig.axes:
        fig.add_axes(ax)

th = np.linspace(0, 2*np.pi*6, 2000)

x = np.sin(th)
y = np.cos(th)
z = th
sc = ax.scatter(x, y, z, s=(5 + th), c=th**2)
fig.savefig('/tmp/a.png')
fig.savefig('/tmp/b.png')
plt.show()

OK, figured out why this passed tests (at least as far as the size goes, the first render is right!):

It also looks like all of our tests with 3D scatter than do not have fixed color are using the figure comparison method so they are wrong together and we missed the color regression 😞 .

v3.3.4

#19812

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

fig = plt.figure()


try:
    # matplotlib 3.4
    ax = Axes3D(fig, auto_add_to_figure=False)
    fig.add_axes(ax)
except AttributeError:
    # matplotlib 3.3.4
    ax = Axes3D(fig)
    if ax not in fig.axes:
        fig.add_axes(ax)

th = np.linspace(0, 2 * np.pi * 6, 2000)

x = np.sin(th)
y = np.cos(th)
z = th
sc = ax.scatter(x, y, z, s=(1 + th * 5), c=th ** 2)
fig.suptitle(mpl.__version__)
fig.savefig("/tmp/a.png", bbox_inches="tight")
fig.savefig("/tmp/b.png")
fig.savefig("/tmp/c.png")
plt.show()

FIX: size and color rendering for Path3DCollection

Navigation Menu

Repository navigation

Marker sizes in Axes3D scatter plot are changing all the time #19787

Bug report

v3.3.4

#19812

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Marker sizes in Axes3D scatter plot are changing all the time #19787

Description

Bug report

Activity

jklymak commented on Mar 27, 2021

tacaswell commented on Mar 28, 2021

tacaswell commented on Mar 28, 2021

tacaswell commented on Mar 28, 2021

tacaswell commented on Mar 28, 2021

tacaswell commented on Mar 28, 2021

tacaswell commented on Mar 29, 2021

v3.3.4

#19812

tacaswell commented on Mar 29, 2021

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions