Spikes element does not overlay properly

[jlstevens]

As shown here:

I get the same issue with matplotlib.

[jlstevens]

Actually, this issue may be a duplicate of the issue described in #1480.

[jlstevens]

Looks like it is fixed?

This screenshot shows a different issue (unexpected colormapping) described in #1669.

[philippjfr]

I'm not sure this was ever a real issue, it's just to do with the semantics of Spikes. By default Spikes do not have a height so you have to set the height as a plot option. In the plot above the height of the Spikes is likely just too small to see. In your new example you have given them a height so there is no issue.

[jlstevens]

Philipp and I discussed this about maybe the height plot options should be 1 by default as a relative quantity (fraction of the y-axis). That would make more sense to me I think.

[neighthan]

@jlstevens I agree. I just learned that VLine can't appear in the legend, so I replaced it with Spikes in my code. However, it's not easy to get spikes that overlay properly when I want them to be the full height of the plot. Now I have to modify my spikes based on what I'm overlaying them with, which isn't nice. Unless explicit heights are given, I think having full-height spikes is the most reasonable default. These are the two workarounds I've found for getting a full-height vertical line that also appears (more or less) in the legend

import numpy as np
import holoviews as hv
hv.extension("bokeh")

x = np.linspace(0, 1)
curve = hv.Curve((x, np.sin(x)), label="sin(x)")
height = curve.data["y"].max() - curve.data["y"].min()
spikes = hv.Spikes(([0.5], [height]), vdims="height", label="mid")
spikes * curve

# need to make sure the colors are the same for spikes/vlines
# would look a bit better if I adjusted the spike thickness too
spikes = hv.Spikes([0.5], label="mid").opts(color="black")
vline = hv.VLine(0.5).opts(color="black")

spikes * curve * vline

[SandervandenOord]

Philipp and I discussed this about maybe the height plot options should be 1 by default as a relative quantity (fraction of the y-axis). That would make more sense to me I think.

Is there already something like relative spike_length, so you don't have to think yourself about what absolute length the spikes should get?
Right now I would do something like scatter * spikes.opts(spike_length='very large number') to be sure it goes over the whole y range of scatter. Edit: I thought this would work, but y-axis just gets the spike range, so this is also not a workaround.
But it would also be nice when you just put the spikes to be at 25% of the y-range?

[jlstevens]

@philippjfr I'm wondering if there is a way to use the dim object to find the y-range and scale as a proportion of that. Should be possible, right?

[philippjfr]

I'm not sure if spike_length accepts a dim transform atm but in theory that should be possible yes. Something like:

spike_length=np.ptp(dim('y'))/10.

[philippjfr]

Actually no that doesn't make sense, that would just get the range of the y-values, not the overall range of the plot.