Violinplot shows groups swapped contrary to boxplot

[rita-o]

Hello,

I am seeing different behavior in seaborn.violinplot compared with seaborn.boxplot when using the same dataframe.

In my case, boxplot looks correct, but violinplot lookes swapped. I found a workaround, but I do not understand the underlying issue: the violin plot only matches the data after I explicitly change the column's type I use for groupping to str instead of the original type which was category. I included an example below.

I am using pandas 3.0.0 and seaborn 0.13.2.

I also saw issue #3893, which seems possibly related.

Thank you in advance,
Rita

import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd

data_pat = df_long[df_long["Pattern"] == 'MD']
data_pat = data_pat[data_pat["ROI"].isin(["Isocortex", "Cerebellum"])]
data_pat = (
    data_pat
    .groupby(["Subject", "ROI", "Group", "Pattern"], as_index=False)
    .agg({"Value": "mean"})
)

# ----------  !! Violinplot looks wrong unless group is cast to str !! ----------
# data_pat["Group"] = data_pat["Group"].astype(str)

hue_order = ['WT', 'HTZ']

fig, axes = plt.subplots(1, 2, figsize=(10,4), sharey=True)

# ---------- LEFT: Violin ----------
ax = axes[0]
sns.violinplot(
    data=data_pat,
    x="ROI",
    y="Value",
    hue="Group",
    width=0.6,
    linewidth=0.3,
    gap=0.25,
    linecolor='grey',
    saturation=0.9,
    hue_order=hue_order,
    ax=ax,
)
sns.stripplot(
    data=data_pat,
    x="ROI",
    y="Value",
    hue="Group",
    hue_order=hue_order,
    dodge=True,
    jitter=False,
    size=3.5,
    alpha=0.5,
    palette=["#bfbfbf"] * 2,
    ax=ax,
)
ax.set_title("Violinplot")
  
# ---------- RIGHT: Boxplot -----------
ax = axes[1]

sns.boxplot(
    data=data_pat,
    x="ROI",
    y="Value",
    hue="Group",
    width=0.6,
    fliersize=0,
    linewidth=0.3,
    gap=0.25,
    linecolor='grey',
    saturation=0.9,
    hue_order=hue_order,
    ax=ax,
)
sns.stripplot(
    data=data_pat,
    x="ROI",
    y="Value",
    hue="Group",
    hue_order=hue_order,
    dodge=True,
    jitter=False,
    size=3.5,
    alpha=0.5,
    palette=["#bfbfbf"] * 2,
    ax=ax,
)
ax.set_title("Boxplot")

[mycroftsnm]

Hi! I'm facing the same issue, it seems to be related to the ordering of the dataset, i workarounded it by sorting the df by the hue variable.

[bidyut18]

Hello Everyone, I have initiated a pull request with a fix which ensures the integrity of the the hue_order in case of pd.Categorical.

Solution

My solution is to manually add a check to ensure that the hue column’s category order is aligned with the mapping’s levels thus fixing the bug.

[oryktos]

This also happens for me using boxenplots. @mycroftsnm your solution worked for me, I sorted the datafame by that column and the plots are fine.

Tangentially (probably unrelated), I ran into a different thing where I had to sort after adding a new column. The new one was a modified version of an existing column. If I sorted first, the new column was just a copy of the source column. This was the same whether sorting in-place or not. I fixed it by sorting the dataframe last, before plotting.

---

Library	Version
Seaborn	0.13.2
Pandas	3.0.3
Matplotlib	3.10.8