Repository navigation
contour edge case with all data below levels and a surrounding field of zeros #20203
Description
Bug report
Bug summary
If I attempt to make a contour plot with data that is entirely below the lowest level, and the data has a field of zeros around it, I get an unexpected contour between the zeros and anything positive.
Code for reproduction
import numpy as np
import matplotlib.pyplot as plt
y = np.linspace(-10, 10, 20)
x_2d, y_2d = np.meshgrid(y, y)
data = np.cos(x_2d * np.pi/10.) + np.cos(y_2d * np.pi/10.)
np.place(data, data < 0., 0.)
levels = [3, 6]
colors = ['b', 'r']
plt.contour(data, levels=levels, colors=colors)
plt.show()Actual outcome
Expected outcome
The plot should be blank, which is what happens if the call to np.place is skipped.
Matplotlib version
- Operating system: RHEL7.9
- Matplotlib version (import matplotlib; print(matplotlib.__version__)): 3.4.1
- Matplotlib backend (print(matplotlib.get_backend())): Qt5Agg
- Python version: 3.8.8
- Jupyter version (if applicable): N/A
- Other libraries: numpy 1.20.2
Activity
ianthomas23 commented on May 11, 2021
I've confirmed that this is a bug. If the levels for a line contour plot are all outside of the z limits, the levels are replaced by the lower z limit, here:
matplotlib/lib/matplotlib/contour.py
Lines 1141 to 1147 in a9c5224
Here is a simple reproducer:
Under most circumstances we'd get away with this behaviour. However, if there is a quad which has the lower-z value on 3 of its 4 corners then half the area of the quad is at the lower level and the contour line is calculated on the upper edge of this triangle (see #19691 for detailed explanation).
I presume that if the user specifies the contour levels then we should not modify them, even if this means that no contour lines are plotted. But I suspect there may be issues about colorbars being displayed without any contents?
rcomer commented on May 16, 2021
Thanks @ianthomas23 for taking a look at this and providing an explanation. With this knowledge we can make a simple workaround in the user code by setting a new, artificial, localised minimum before plotting. E.g.