如何基于百分位数设置Seaborn箱线图的须？

Got it, let's figure out how to set whiskers by percentiles in Seaborn boxplots—since Seaborn doesn’t support passing direct percentile values like whis=[5, 95] (unlike Matplotlib’s native boxplot), and your earlier approach of mapping percentiles to IQR multiples only works for perfectly normal data, which doesn’t fit your use case. Here are two solid workarounds:

Method 1: Manually Calculate Percentiles and Adjust Whiskers/Caps #

This approach precomputes your target percentiles for each group, then updates the Seaborn boxplot's whisker and cap positions directly. It keeps Seaborn's native styling while giving you full control over whisker placement.

Step-by-Step Code Example: #

import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np

# Use the tips dataset as a stand-in for your data
tips = sns.load_dataset("tips")

# Define your target percentiles
lower_percentile = 5
upper_percentile = 95

# Calculate percentiles for each group (e.g., grouped by 'day' here)
grouped_data = tips.groupby('day')['total_bill']
percentile_values = grouped_data.agg([
    lambda x: np.percentile(x, lower_percentile),
    lambda x: np.percentile(x, upper_percentile)
])
percentile_values.columns = ['lower', 'upper']

# Create the base Seaborn boxplot
fig, ax = plt.subplots(figsize=(8, 6))
sns.boxplot(x='day', y='total_bill', data=tips, ax=ax)

# Extract whisker and cap elements from the plot
whiskers = ax.lines[::2] + ax.lines[1::2]  # Split into lower and upper whiskers
caps = ax.lines[len(ax.lines)//2:]         # Caps sit at the end of whiskers

# Update each group's whiskers and caps to match your percentiles
for i, (group, pcts) in enumerate(percentile_values.iterrows()):
    # Update lower whisker and cap
    whiskers[2*i].set_ydata([pcts['lower'], pcts['lower']])
    caps[2*i].set_ydata([pcts['lower'], pcts['lower']])
    # Update upper whisker and cap
    whiskers[2*i + 1].set_ydata([pcts['upper'], pcts['upper']])
    caps[2*i + 1].set_ydata([pcts['upper'], pcts['upper']])

plt.title(f"Boxplot with Whiskers at {lower_percentile}th and {upper_percentile}th Percentiles")
plt.show()

Method 2: Use Matplotlib's Boxplot + Seaborn Styling #

Since Seaborn is built on top of Matplotlib, you can create a Matplotlib boxplot with percentile-based whiskers first, then apply Seaborn's theme to match your desired aesthetic. This is simpler if you don't need all of Seaborn's boxplot-specific styling.

Code Example: #

import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np

tips = sns.load_dataset("tips")
sns.set_theme(style="whitegrid")  # Apply Seaborn's clean theme

# Prepare grouped data for Matplotlib's boxplot
group_labels = tips['day'].unique()
grouped_values = [tips[tips['day'] == day]['total_bill'].values for day in group_labels]

fig, ax = plt.subplots(figsize=(8, 6))
# Use Matplotlib's boxplot with direct percentile whisker definition
ax.boxplot(grouped_values, whis=[lower_percentile, upper_percentile], labels=group_labels)

# Add Seaborn-style polish
sns.despine(ax=ax)
ax.set_ylabel('Total Bill')
ax.set_xlabel('Day')
plt.title(f"Boxplot with Whiskers at {lower_percentile}th and {upper_percentile}th Percentiles")
plt.show()

Key Notes: #

• Both methods work for non-normal data because they directly use your specified percentiles—no reliance on IQR multiples that only make sense for normal distributions.
• Method 1 preserves Seaborn's default color palettes and group spacing, while Method 2 gives you direct Matplotlib control with easy Seaborn theming.

内容的提问来源于stack exchange，提问作者Hauzero