Band behavior & NaN values

What is the expected behavior when using a Band when NaN values occur for the lower or upper value? Currently the band seems to collapse into the lower/upper value which is not missing. Naturally I would have expected that Band will not be drawn at all for such a lower/upper value combination. Of course the data can be cleaned upfront however having an option to control this behavior would be great.

import numpy as np
from bokeh.io import curdoc
from bokeh.models import ColumnDataSource, DataTable, DateFormatter, TableColumn, Band
from bokeh.plotting import figure

doc = curdoc()

n_rows = 10

data = dict(
        downloads=[np.sin(i)+2 for i in range(n_rows)],
        x=[i for i in range(n_rows)]
    )

data["downloads"][:1] = [np.nan, np.nan]

source = ColumnDataSource(data)

f = figure()
l = f.line(x="x", y="downloads", source=source)
s2_band = Band(base="x", lower=0, upper="downloads", source=source,
               level='underlay', fill_alpha=0.15, fill_color="green")
f.add_layout(s2_band)

doc.add_root(f)

bokeh_band_nan619×621 17.9 KB

I expect this is currently undefined behavior. AFAIK the band values are simply added/subtracted from the base series and passed on the HTML canvas APIs to draw fill regions from the results. It’s not really clear what would be useful to do in this case. Neither skipping the point or treating it as a zero value seem generally appropriate.

Would it be reasonable to have a user configuration here to enable some control over what should happen in this scenario?

I haven’t ever seen any other questions about this. Since there are multiple possible ways to handle things, but not one clear “right” way, it would take some complexity to add this in a configurable way. Until/unless there’s more user interest expressed, I think it’s probably a case where users can best just process their own data to suit their specific needs.

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.