Add split violin

[ahuang11]

#4096
Not ready yet. I don't know how to add a legend and change the colors. Not sure if this implementation is the best approach. Also, not sure how to make the widths as proportionate as seaborn. Feel free to edit/make commits directly. Also didn't handle for matplotlib/plotly.

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

import seaborn as sns
sns.set(style="whitegrid")
tips = sns.load_dataset("tips")

ax = sns.violinplot(x="day", y="total_bill", hue="sex",
                    data=tips, palette="Set2", split=True,
                    scale="count")

import pandas as pd
tips = pd.concat([
    tips.loc[tips['day'] == 'Thur'],
    tips.loc[tips['day'] == 'Fri'],
    tips.loc[tips['day'] == 'Sat'],
    tips.loc[tips['day'] == 'Sun']
])

hv.Violin(tips, 'day', ['total_bill', 'sex']).opts(ylim=(-10, 60), width=600)

[philippjfr]

Thanks for having a go at this, sadly as I said above this design doesn't really fit in with the rest of HoloViews.

[ahuang11]

No worries. I was basing it off this idea:
hv.Area((X, Y, Y2), vdims=['y', 'y2']) * hv.Area((X, Y, Y3), vdims=['y', 'y3'])

and I think I tried kdims first, but it created groups like:
hv.Violin(tips, ['day', 'sex'], ['total_bill']).opts(ylim=(-10, 60), width=600)

Any suggestions on how to deal with this group vs split violin? The closest I think is like stacked bar chart?

[ahuang11]

Okay, I swapped vdims for kdims. Still not sure how to add coloring/legend.

When I try:

        # if split_key is not None:
        #     legend_prop = 'legend_field' if bokeh_version >= '1.3.5' else 'legend'
        #     kde_map[legend_prop] = split_key

It says the legend must be a string

[philippjfr]

Thanks, I'll take a look at coloring and legend. Instead of split=True/False I think it should accept a dim mapping.

[philippjfr]

Initial attempt, still needs some cleanup:

[philippjfr]

Added a gallery example:

[ahuang11]

Really excited for this!

[ccneko]

Thanks for the really nice functionality. By the way, any chance of Plotly support? (Looking at Panel to use both Bokeh & Plotly meanwhile)

[ahuang11]

I don't think so, but maybe you can open a new issue for split violin plots in plotly.

[ccneko]

@ahuang11 Thanks for confirming. I will directly use the Plotly split violin plots first for the current usage.