Bug summary

Plotting a numpy masked array against a list of datetime objects, plt.plot functions as expected, but plt.scatter still displays the masked data. Auto-axis limits do respect the masked data though.

Code for reproduction

import matplotlib.pyplot as plt
import numpy.ma as ma
from datetime import datetime

# Create a list of datetime objects
time_unmasked = [datetime(2022, 11, 24, 4, 49, 52),
                 datetime(2022, 11, 24, 4, 49, 53),
                 datetime(2022, 11, 24, 4, 49, 54),
                 datetime(2022, 11, 24, 4, 49, 55),
                 datetime(2022, 11, 24, 4, 49, 56),
                 datetime(2022, 11, 24, 4, 49, 57),
                 datetime(2022, 11, 24, 4, 49, 58)]

# Create a masked array with three masked variables, one datapoint far from others
data_masked = ma.array([15,14,50,15,15,15,16], mask=[0,0,1,1,1,0,0])

# Line plot (expected behaviour)
plt.plot(time_unmasked, data_masked)
plt.show()

# Scatter plot (plots masked points, but ignores masked values when auto-setting axis limits)
plt.scatter(time_unmasked, data_masked)
plt.show()

Actual outcome

plot behaves as expected.

scatter unexpectedly draws the masked datapoints, but axes limits are still respecting the masked array as the masked y=50 value is not visible.

Expected outcome

Scatter should not plot the masked data even though all timestamps are unmasked.

Additional information

It seems as though there is a check within matplotlib.pyplot.scatter that is not catching. I'd expect that logic to check that if either x or y were masked, to not plot that point. The datetime object is able to slip through. I've also tried putting time_unmasked within a numpy masked array with the same mask as masked_data (so both x and y have the same mask), but the same unexpected behaviour occurs.

Operating system

MacOS 11.3.1 (20E241)

Matplotlib Version

3.5.3

Matplotlib Backend

module://ipykernel.pylab.backend_inline

Python version

3.8.11

Jupyter version

5.7.10

Installation

conda