[ENH]: Add split feature for violin plots

[anjabeck]

Problem

Recently, I wanted to get split violin plots at defined x-positions (see image).
My problem is that while it's possible to specify the x-position in pyplot.violinplot(), there is no option for split violins.
seaborn.violinplot() on the other hand allows split violins but seems to enforce discrete x-positions with fixed intervals. This is not true anymore since v0.13

This example shows two data sets (black and red/blue) in four categories defined by $A=0,1$ and $B=0,1$.

The code that produces the plot based on this answer on stackoverflow.

import matplotlib.pyplot as plt
import numpy as np

np.random.seed(0)

cols = ["firebrick","dodgerblue"]
leg = {"patches": [], "labels": []}

# Loop over categories A and B
for A in range(0,2):
    xpos   = []
    dLeft  = []
    dRight = []

    for B in range(0,2):
        # Create data for the left and right violins in the current category
        mu = np.random.randint(10,100)
        dLeft.append(np.random.poisson(mu,1000))
        dRight.append(np.random.normal(mu+10,np.sqrt(mu+10),1000))

        # Create x-positions for the violins
        xpos.append(B+(A-0.5)*0.15)

        # Create right violins (colors)
        vRight = plt.violinplot(dRight,positions=xpos,widths=0.1,showextrema=False,showmedians=False)
        for vr in vRight["bodies"]:
            # Get the center
            m = np.mean(vr.get_paths()[0].vertices[:, 0])
            # Get the rightmost point
            r = np.max(vr.get_paths()[0].vertices[:, 0])
            # Only allow values right of the center
            vr.get_paths()[0].vertices[:, 0] = np.clip(vr.get_paths()[0].vertices[:, 0], m, r)
            # Set category A color
            vr.set_color(cols[A])
            vr.set_alpha(1.)

        # Create left violins (black)
        vLeft = plt.violinplot(dLeft,positions=xpos,widths=0.15,showextrema=False,showmedians=False)
        for vl in vLeft["bodies"]:
            # Get the center
            m = np.mean(vl.get_paths()[0].vertices[:, 0])
            # Get the leftmost point
            l = np.min(vl.get_paths()[0].vertices[:, 0])
            # Only allow values left of the center
            vl.get_paths()[0].vertices[:, 0] = np.clip(vl.get_paths()[0].vertices[:, 0], l, m)
            # Set color
            vl.set_color("k")
            vl.set_alpha(1.)

    # Add legend entry
    leg["patches"].append(vRight["bodies"][0])
    leg["labels"].append(fr"$A={A}$")

plt.legend(leg["patches"],leg["labels"])
plt.xlabel(r"$B$")
plt.xticks([0,1],["0","1"])
plt.xlim(-0.5,1.5)
plt.tight_layout()
plt.savefig("example_violins.png")

Proposed solution

An easy-to-use solution would be to add an argument in the violinplot() function through which the user can define which side to plot. Something like asymetric=None, "below", "above", where "below"/"above" only plots the side below/above the values given to positions.

This would mean that calling

vRight = plt.violinplot(dRight,positions=xpos,widths=0.1,showextrema=False,showmedians=False,asymetric="above")

internally executes this loop

for vr in vRight["bodies"]:
    # Get the center
    m = np.mean(vr.get_paths()[0].vertices[:, 0])
    # Get the rightmost point
    r = np.max(vr.get_paths()[0].vertices[:, 0])
    # Only allow values right of the center
    vr.get_paths()[0].vertices[:, 0] = np.clip(vr.get_paths()[0].vertices[:, 0], m, r)

[timhoffm]

Thanks for the suggestion. This is a reasonable addition. 👍 Do you want to make a pull request?

Technical remarks:

• I would call the parameter side with values "low" / "high" / "both"
• Rather than modifying the bodies afterwards, the implementation should change the creation of the bodies here:
  

  matplotlib/lib/matplotlib/axes/_axes.py

  Line 8459 in 88d64e5

  bodies += [fill(stats['coords'], -vals + pos, vals + pos,

[anjabeck]

Sweet. I'll create a pull request :)

[mwaskom]

seaborn.violinplot() on the other hand allows split violins but seems to enforce discrete x-positions with fixed intervals.

BTW this is not true as of seaborn v0.13

[anjabeck]

seaborn.violinplot() on the other hand allows split violins but seems to enforce discrete x-positions with fixed intervals.

BTW this is not true as of seaborn v0.13

Oh that's cool actually. I completely missed that!