Missing tests for norm=Normalize(...)

Currently, we don't have enough tests including the norm=Normalize(...) argument. It should work for matplotlib and also datshader (after #309), but I think we need some more tests and maybe also a bit more documentation.

Also, I noted that both seem to give the same result, not sure if that's a bug

from spatialdata.datasets import blobs
from matplotlib.colors import Normalize
blob=blobs()

blob.pl.render_points(color="instance_id", size=40, norm=Normalize(4, 7, clip=True)).pl.show()
blob.pl.render_points(color="instance_id", size=40, norm=Normalize(4, 7, clip=False)).pl.show()

Thanks for reporting. The instance_id column has values in [0, 1, ..., 9]. With clip=True one would expect the values [0, 1, 2, 3, 4] to have the same color, and the values [7, 8, 9] to have the same color. So it seems like that there is a bug.

Did you notice this only for matplotlib or also for datashader?

When using datashader, both plots look identical as well.

After looking into this a bit more, I would say the behavior is not a bug: according to the matplotlib documentation (https://matplotlib.org/stable/api/_as_gen/matplotlib.colors.Normalize.html), with clip=False, values outside of [vmin, vmax] are mapped to values outside [0, 1] (and not exactly to 0 or 1 as would happen with clip=True). Then, the respective colormap can "can mark these under and over values with specific colors".
In the examples above, we use the default colormap viridis, which uses its minimal value for under values and its maximal value for over values (see below).

So, if you would set over and under differently, the result looks like this for clip=False (the clip=True plot looks like above)

from matplotlib import rcParams
import matplotlib

cmap = matplotlib.colormaps[rcParams["image.cmap"]]
cmap.set_under("grey")
cmap.set_over("grey")

blob.pl.render_points(color="instance_id", size=40, norm=Normalize(4, 7, clip=False), cmap=cmap).pl.show()

You are right! Thanks for checking this 😊

Notes on the datashader behavior here: ds.tf.shade() has an argument span that we can use for passing [norm.vmin, norm.vmax]. Unfortunately, there is no way to pass the norm.clip value and datashader always behaves like clip=True (described here: https://datashader.org/getting_started/Pipeline.html#controlling-ranges-for-colormapping). This also means that doing a manual normalization before calling shade() won't lead to the correct result, since these values would be clipped to [0, 1] automatically.

Looks like this could be a workaround: https://discourse.holoviz.org/t/specify-datashader-colors-outside-of-color-limits/8561/4

Navigation Menu

Repository navigation

Missing tests for norm=Normalize(...) #372

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Missing tests for norm=Normalize(...) #372

Description

Activity

LucaMarconato commented on Jan 5, 2025

Sonja-Stockhaus commented on Jan 6, 2025

Sonja-Stockhaus commented on Jan 14, 2025

LucaMarconato commented on Jan 15, 2025

Sonja-Stockhaus commented on Feb 4, 2025

Sonja-Stockhaus commented on Feb 10, 2025

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions