Inconsistent behaviour with hv.Image hv.QuadMesh when plotting in logspace

[zhiihan]

ALL software version info

(this library, plus any other relevant software, e.g. bokeh, python, notebook, OS, browser, etc should be added within the dropdown below.)

Software Version Info

bokeh = 3.8.0
holoviews = 1.21.0
numpy = 1.26.4
python = 3.10.19

Description of expected behavior and the observed behavior

Hi, I am trying to generate some plots in logspace. My expectation is that if I have data in logspace and my axes are in logspace, then I should be able to avoid rtol or anything when using hv.Image.

A workaround is to plot the exponent of the data: i.e.

hv.Image

import holoviews as hv
import numpy as np

hv.extension('bokeh')

npoints = 50

x = np.linspace(1, 10, 5)
y = np.logspace(0.1, 2, npoints)
transformed_y = np.log10(y)

data = np.random.rand(npoints, 5)
img = hv.Image((x, transformed_y, data), kdims=['x', 'y']).opts(
    colorbar=True,
    ylim=(0.1, 2)
)
img

But I think this should work out of the box if the user uses logy = True, no? i.e. this is the plot I expect to see (note that the y axes is linear, not logspace). I now run two tests for when using log axes, i.e. logy = True when y is logspace data.

Complete, minimal, self-contained example code that reproduces the issue

Test 1: hv.Image with npoints = 50, 5

import holoviews as hv
import numpy as np

hv.extension('bokeh')

npoints = 50 # change this to 5 later

x = np.linspace(1, 10, 5)
y = np.logspace(0.1, 2, npoints)

data = np.random.rand(npoints, 5)

img = hv.Image((x, y, data), kdims=['x', 'y']).opts(
    logy=True,
    colorbar=True,
    ylim=(1, 100)
)
img

when npoints = 50, we see the expected plot logspace but with a warning. When npoints = 5, we get a blank plot.
Furthermore, when y has negative exponents, i.e. y = np.logspace(-4, -2, npoints), even for npoints = 50, we also get a blank plot.

Test 2: hv.QuadMesh

Here, at low npoints=5 count, we use hv.QuadMesh. The bottom is missing and a white strip appears.

import holoviews as hv
import numpy as np

hv.extension('bokeh')

npoints = 5

x = np.linspace(1, 10, 5)
y = np.logspace(-5, -1, npoints)

data = np.random.rand(npoints, 5)

img = hv.QuadMesh((x, y, data), kdims=['x', 'y']).opts(
    logy=True,
    colorbar=True,
    ylim=(1e-5, 1e-1)
)
img

Observed:

Expected: (generated with transformed_y)

Code that generates the expected plot:

import holoviews as hv
import numpy as np

hv.extension('bokeh')

npoints = 5

x = np.linspace(1, 10, 5)
y = np.logspace(-5, -1, npoints)

data = np.random.rand(npoints, 5)

img = hv.QuadMesh((x, y, data), kdims=['x', 'y']).opts(
    logy=True,
    colorbar=True,
    ylim=(1e-5, 1e-1)
)
img

Stack traceback and/or browser JavaScript console output

Screenshots or screencasts of the bug in action

• I may be interested in making a pull request to address this

[hoxbro]

It looks weird to me; I will mark it as a bug for now. The transformation of data is around these lines and edges=True:

holoviews/holoviews/plotting/bokeh/raster.py

Lines 599 to 600 in b5781ca

	xc, yc = (element.interface.coords(element, x, edges=True, ordered=True),
	element.interface.coords(element, y, edges=True, ordered=True))