Support automatic control of labels density to avoid overlap

[guidocioni]

In geographical plots (but I think this would also benefit other types of plot types) it is useful to automatically reduce the density of labels so that the plot become more readable. The density should be then automatically adjusted depending on the zoom level. This is something already implemented in libraries like plotly or leaflet: here is a demonstration.

Screen.Recording.2025-09-29.at.11.12.02.mov

Just for reference this is the current behaviour in hvplot (I don't know if the density reduction while zooming is expected or just a bug in my renderer).

Screen.Recording.2025-09-29.at.11.14.10.mov

This could be activated with a boolean.

[ahuang11]

As a workaround, I suspect you can do something like https://discourse.holoviz.org/t/show-individual-points-when-zoomed-else-datashade/2204

import holoviews as hv
import numpy as np

from holoviews.operation import decimate
from holoviews.operation import apply_when
hv.extension("bokeh")

points = hv.Labels(np.random.randn(5000, 3))

apply_when(points, operation=decimate.instance(max_samples=10), predicate=lambda x: len(x) > 100)

Then replace the decimate operation with something that is more aware of its neighbors and depends more on the zoom level

[guidocioni]

As a workaround, I suspect you can do something like https://discourse.holoviz.org/t/show-individual-points-when-zoomed-else-datashade/2204

import holoviews as hv
import numpy as np

from holoviews.operation import decimate
from holoviews.operation import apply_when
hv.extension("bokeh")

points = hv.Labels(np.random.randn(5000, 3))

apply_when(points, operation=decimate.instance(max_samples=10), predicate=lambda x: len(x) > 100)
Then replace the decimate operation with something that is more aware of its neighbors and depends more on the zoom level

Thanks for the suggestion @ahuang11, I didn't know about decimate as I'm kind of new to hvplot. I'll try it out

[ahuang11]

You can replace the operation with a generic function too IIRC (if not, subclass operation)

[guidocioni]

Hey @ahuang11 I tried to use your example but I'm not sure if I'm using it correctly because I'm seeing this kind of behaviour and the apply_when operation does not seem to do anything special to the behaviour of the plot.

labels = hv.Labels(np.random.randn(5000, 3))
apply_when(labels, operation=decimate.instance(max_samples=10), predicate=lambda x: len(x) > 100)
labels

Screen.Recording.2025-10-02.at.11.08.44.mov

[ahuang11]

You didn't save the apply_when line to a variable.

[guidocioni]

You didn't save the apply_when line to a variable.

Ah ok, I see. I couldn't find any concrete example in the doc on how to use apply_when so I was a little bit lost.

Do you know if it's possible to convert whatever is returned by the pandas wrapper of hvplot to an object that can be used by apply_when?

This doesn't work

labels = df.hvplot.labels("longitude", "latitude", geo=True, tiles="OSM")
w = apply_when(labels, operation=decimate.instance(max_samples=10), predicate=lambda x: len(x) > 100)
w

as it fails with AttributeError: 'Overlay' object has no attribute interface.. I think what is returned by the hvplot method on the pandas dataframe may not be the same type as what is returned from labels = hv.Labels(np.random.randn(5000, 3)). I tried with labels.Labels but it throws the same error.

[ahuang11]

tiles=True returns an overlay. Add tiles with geoviews afterward.
https://geoviews.org/gallery/bokeh/tile_sources.html