Skip to content

scatter3D assigns wrong color to points for some plot orientations #18287

Closed

Description

@MarkusSchwarz1980

Bug report

Bug summary

When colors are based on coordinates in scatter3D, the color changes when the orientation of the plot is changed.

Code for reproduction

This codes plots 10 points along the x-axis from 0 to x_max, using scatter3D with their color depending on their x-position (dark purple for 0, yellow for x_max). The first view_point gives the expected result, whereas the second does not. For comparison, the same data plotted with plot (and offset by 1 in the z-direction for clarity) always give the expected result.

import numpy as np
from matplotlib import pyplot as plt

import matplotlib
backend = matplotlib.get_backend()

# two view points to show the problem
view_points = [[15, -70],[15, -110]]

# create points along the x-axis
num_points, x_max = 10, 4
x_coord = np.linspace(0, x_max, num_points)
y_coord = np.zeros_like(x_coord)
z_coord = np.zeros_like(x_coord)

coords_3d = np.array([x_coord, y_coord, z_coord])

for view_point in view_points:

    fig = plt.figure(f'view point {view_point}', clear=True)
    ax3d = fig.gca(projection='3d')
    ax3d.set_title(f'backend {backend}, view point {view_point}')
    
    ax3d.set_xlabel('x')
    ax3d.set_ylabel('y')
    ax3d.set_zlabel('z')

    # color points according to x-coordinate
    scatter = ax3d.scatter3D(*coords_3d, c=x_coord, depthshade=False)

    # draw same points using `plot` and offset by 1 in z-direction for clarity
    for coord in coords_3d.T:
        ax3d.plot(*coord + [0,0,1], 'x',
                  color=scatter.cmap(scatter.norm(coord[0])) )

    fig.colorbar(scatter, label='x-coordinate')

    ax3d.view_init(*view_point)

    ax3d.legend(ax3d.get_children()[:2], ['scatter','plot'])
    fig.tight_layout()
    plt.show()

Actual outcome

Qt5Agg_-110

The view_point=[15,-110] assigns the color in the wrong (reversed) order: the point at (0,0,0) has the bright color corresponding to an x-coordinate of x_max, which is clearly wrong. This occurs for azimuth angles between [-180,-90] and [90, 180] degrees.

Expected outcome

Qt5Agg_-70

The view_point=[15,-70] produces the expected outcome, i.e. dark colors for points close to the origin at (0,0,0) and bright colors for points farther away. This should be independent on the orientation of the plot.

Matplotlib version

  • Operating system: maxOS 10.15.6 (19G2021)
  • Matplotlib version: 3.3.1
  • Matplotlib backend (print(matplotlib.get_backend())): tested with Qt5Agg and MacOSX
  • Python version: 3.7.8
  • Jupyter version (if applicable):
  • Other libraries:

Activity

  1. timhoffm commented on Aug 18, 2020

    @timhoffm
    Member

    Maybe introduced via #17378? This sounds similar to #18135.

  2. QuLogic commented on Aug 19, 2020

    @QuLogic
    Member

    It is indeed caused by the merge commit from #17378, though the fix in #18135 was to start sorting sizes, while facecolors and edgecolors are already sorted the same way.

  3. QuLogic commented on Aug 19, 2020

    @QuLogic
    Member

    Oops, sorry, I was mislead by view_init not setting the Axes stale correctly. That is not the commit that introduces the issue. It is actually #18156.

  4. added a commit that references this issue on Aug 19, 2020
    5bd536b
  5. added this to the v3.3.2 milestone on Aug 19, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions