go.Scatter ignores the masked array elements in numpy masked arrays

[ludwig-kastner]

Bug summary

Plotting a numpy masked array, px.scatter plots the non-masked array elements and limits the axis accordingly, while go.Scatter also displays the masked data and does also include the masked elements for the axis limits. The behavior from go.Scatter is unexpected in this regard and ignoring the mask defeats the purpose of masked arrays.

Minimal viable code

import numpy.ma as ma
import plotly.express as px
import plotly.graph_objects as go

# Create timestamps
time_unmasked = [1,2,3,4,5,6,7]

# 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])

fig_px = px.scatter(x=time_unmasked, y=data_masked, title='Plotly Express')
fig_px.show()

fig_go = go.Figure(layout=go.Layout(
        title=go.layout.Title(text="Plotly Graph Objects")))
fig_go.add_trace(go.Scatter(x=time_unmasked, y=data_masked))
fig_go.show()

Thanks to @Benblob688 for his code example in the issue mentioned below.

Actual outcome

Using plotly.graph_objects, the masked data entries at position 2,3,4 with the values 50,15,15 are drawn into the plot and the axis limits are set accordingly.
Expected outcome

Using plotly.express, the masked data entries are not shown and the axis limits are set accordingly to only include non-masked entries in the array.

Additional information
Matplotlib had a similar issue here, where the code has been adapted from to display this issue here. It seems that the usage of np.column_stack caused an issue there, that function has since been replaced with np.ma.column_stack.

Operating system

Ubuntu 20.04

Plotly Version

5.14.1
Numpy Version
1.24.1
Python version

3.11.3
Installation

conda