Add HoloViews' subcoordinate_y

[droumis]

subcoordinate_y was recently added to HoloViews to plot on subcordinates. This should now be incorporated into hvPlot when working with xarray.

The API could be something as simple as: da.hvplot(subcoordinate_y=True).

This would produce something like:

The minimum data are:

• 1D time array (common to all channel sources)
• 2D stack (time x channel source) of 1D amplitude arrays
• 1D array of channel source labels

There should also be the option for adjusting subcoordinate_scale (as in HoloViews).

Whether or not the minimap should be included with the simple default API call is something that needs to be discussed and considered further.

[maximlt]

@droumis I can see the project board that this is planned for Q4 of this year. Is this a hard deadline?

But my main question would be: is this feature not too niche to be exposed in hvPlot? Increasing hvPlot's API comes with a cost, it makes it more difficult for users to navigate through its many parameters. I'm not -1 on this, but I'd like to make sure this is discussed.

---

Whether or not the minimap should be included with the simple default API call is something that needs to be discussed and considered further.

Yes I wouldn't tie the "minimap" (or however we call it in the end) with subcoordinate_y. We should open a separate issue to discuss how we'd generally like to expose the minimap in hvPlot.

---

If it's eventually implemented, I guess we could also support a DataFrame with a datetime index, using the column names as the source labels.

[philippjfr]

But my main question would be: is this feature not too niche to be exposed in hvPlot? Increasing hvPlot's API comes with a cost, it makes it more difficult for users to navigate through its many parameters. I'm not -1 on this, but I'd like to make sure this is discussed.

Personally, I feel like this is more of an issue on how our doc(strings) are organized than a general issue. subplot_y can be quite a nice and general feature and if it isn't general enough to generate e.g. a joy plot then that should be improved at the HoloViews level.

[droumis]

@maximlt , I'm in the process of updating the project board (this week); Q4 is not a hard deadline for this task, I'm going to push it to later since we are also still working on tasks that would directly impact this implementation.

I agree with @philippjfr that a ridgeplot (aka joyplot) is a common enough type of plot to justify the inclusion burden for hvPlot. With subcoordinate_y this is possible using Curve, but Area is more commonly used for a ridgeplot and it looks like it needs the element order to be reversed by default. I'll file an issue with HoloViews.

Code

import numpy as np
import holoviews as hv; hv.extension('bokeh')
from scipy.stats import gaussian_kde

categories = ['A', 'B', 'C', 'D', 'E']
data = {cat: np.random.normal(loc=i-2, scale=1.0, size=100) for i, cat in enumerate(categories)}
x = np.linspace(-5, 5, 100)

curves = []
areas = []
for i, (cat, values) in enumerate(data.items()):
    pdf = gaussian_kde(values)(x)
    
    curve = hv.Curve((x, pdf), label=cat).opts(
        subcoordinate_y=True, 
        subcoordinate_scale=1.5,
    )
    curves.append(curve)
    
    area = hv.Area((x, pdf), label=cat).opts(
        subcoordinate_y=True, 
        subcoordinate_scale=1.5,
    )
    areas.append(area)

ridge_plot_curves = hv.Overlay(curves).opts(
    width=900,
    height=400,
)

ridge_plot_areas = hv.Overlay(areas).opts(
    width=900,
    height=400,
)

# ridge_plot_areas.opts(show_legend=False)
ridge_plot_curves.opts(show_legend=False)

[philippjfr]

While we're at it, HoloViews should also support grabbing the label for each subcoordinate from the NdOverlay key. So it becomes as simple as this:

import numpy as np
import hvplot.pandas
from scipy.stats import gaussian_kde

categories = ['A', 'B', 'C', 'D', 'E']
data = {cat: np.random.normal(loc=i-2, scale=1.0, size=100) for i, cat in enumerate(categories)}
pd.DataFrame(data).hvplot.kde(y=categories, subcoordinate_y=True, subcoordinate_scale=1.5)

Right now you have to manually insert the labels explicitly:

import numpy as np
import hvplot.pandas
from scipy.stats import gaussian_kde

categories = ['A', 'B', 'C', 'D', 'E']
data = {cat: np.random.normal(loc=i-2, scale=1.0, size=100) for i, cat in enumerate(categories)}

labels = iter(categories)
overlay = pd.DataFrame(data).hvplot.kde(y=categories).map(lambda el: el.relabel(next(labels)), specs='Distribution')
overlay.opts('Distribution', subcoordinate_y=True, subcoordinate_scale=1.5)

[droumis]

Discussion in meeting: Preferred API is probably 'ridge' kind, so like hvplot(kind='ridge')

[droumis]

(Bokeh#8128) discussion about naming ridge instead of joy